No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Real-Time Detection of Driver Cognitive Distraction Using Support Vector Machines.
... For example, Solovey et al. (14) reached 89% accuracy in classifying 2 levels of cognitive load (no-task versus an auditory recall 2-back task) using driving performance, GSR, and HR data collected in an on-road study. Liang et al. (15) reached 81.1% accuracy in identifying 2 levels of cognitive load (no-task versus an auditory stock ticker task) using driving performance and eye-tracking data collected in a simulator study. In general, driving performance measures used in these earlier studies (e.g., speed and lane position) are highly sensitive to traffic conditions and may require additional driving context-assessment to improve their utility in driver state detection (16). ...
... The distance metric dictates how the distances between the unknown sample and the voting samples are calculated. For SVM, a radial basis function (RBF) kernel was used, as it yielded the best performance in most of the previous attempts in classifying levels of cognitive load with SVM (e.g., Liang et al. [15], Wang et al. [22], He et al. [26]). Two additional hyperparameters were tuned for SVM. ...
... In this paper, two physiological measures that are available in consumer-grade wearable devices, that is HR and GSR, were fused with eye-tracking measures, leading to 97.8% accuracy with a RF model in classifying three levels of cognitive load (no task, lower difficulty 1-back task, and higher difficulty 2-back task). This result is promising when compared with the accuracies reached in previous research that combined driving performance with eye-tracking measures (e.g., 81.1% in Liang et al. [15]), and also with the accuracies reached in previous research that combined driving performance measures with physiological measures (e.g., 89% in Solovey et al. [14]), especially considering that our 3-class classification problem is more challenging than the 2-class problems tackled in these earlier studies. GSR contributed more to the performance of the models compared with HR: when HR was added to eye-tracking data, the model accuracies increased from 3.2% (with RF) to 17.9% (with SVM), whereas with HR, the increases ranged from 9.7% (with RF) to 29.3% (with SVM). ...
In-vehicle infotainment systems can increase cognitive load and impair driving performance. These effects can be alleviated through interfaces that can assess cognitive load and adapt accordingly. Eye-tracking and physiological measures that are sensitive to cognitive load, such as pupil diameter, gaze dispersion, heart rate (HR), and galvanic skin response (GSR), can enable cognitive load estimation. The advancement in cost-effective and nonintrusive sensors in wearable devices provides an opportunity to enhance driver state detection by fusing eye-tracking and physiological measures. As a preliminary investigation of the added benefits of utilizing physiological data along with eye-tracking data in driver cognitive load detection, this paper explores the performance of several machine learning models in classifying three levels of cognitive load imposed on 33 drivers in a driving simulator study: no external load, lower difficulty 1-back task, and higher difficulty 2-back task. We built five machine learning models, including k-nearest neighbor, support vector machine, feedforward neural network, recurrent neural network, and random forest (RF) on (1) eye-tracking data only, (2) HR and GSR, (3) eye-tracking and HR, (4) eye-tracking and GSR, and (5) eye-tracking, HR, and GSR. Although physiological data provided 1%–15% lower classification accuracies compared with eye-tracking data, adding physiological data to eye-tracking data increased model accuracies, with an RF classifier achieving 97.8% accuracy. GSR led to a larger boost in accuracy (29.3%) over HR (17.9%), with the combination of the two factors boosting accuracy by 34.5%. Overall, utilizing both physiological and eye-tracking measures shows promise for driver state detection applications.
... efforts to understand and reduce distracted behaviours. However, nearly all the focus has been placed on driver distractions, [7][8][9][10][11][12] and the research of phone-related pedestrian distractions has not been systemically studied. Previous studies [13,14] show that distracted walking with cell phones or other handheld devices will cause severe pedestrian safety problems and some interventions should be applied to improve the pedestrian safety. ...
... Many previous works to improve driving safety are about driver distractions. [7][8][9][10][11] Strayer et al. [7] found that using the cell phone during the simulated driving slowed reaction speed by approximately 18% and increased the risk of collision. Liang et al. [8] used eye movements and driving performance data to evaluate the driver distractions. ...
... [7][8][9][10][11] Strayer et al. [7] found that using the cell phone during the simulated driving slowed reaction speed by approximately 18% and increased the risk of collision. Liang et al. [8] used eye movements and driving performance data to evaluate the driver distractions. Pettitt et al. [9] tried to define driver distractions. ...
The distracted phone‐use behaviours among pedestrians, like Texting, Game Playing and Phone Calls, have caused increasing fatalities and injuries. However, the research of phone‐related distracted behaviour by pedestrians has not been systemically studied. It is desired to improve both the driving and pedestrian safety by automatically discovering the phone‐related pedestrian distracted behaviours. Herein, a new computer vision‐based method is proposed to detect the phone‐related pedestrian distracted behaviours from a view of intelligent and autonomous driving. Specifically, the first end‐to‐end deep learning based Two‐Branch Convolutional Neural Network (CNN) is designed for this task. Taking one synchronised image pair by two front on‐car GoPro cameras as the inputs, the proposed two‐branch CNN will extract features for each camera, fuse the extracted features and perform a robust classification. This method can also be easily extended to video‐based classification by confidence accumulation and voting. A new benchmark dataset of 448 synchronised video pairs of 53,760 images collected on a vehicle is proposed for this research. The experimental results show that using two synchronised cameras obtained better performance than using one single camera. Finally, the proposed method achieved an overall best classification accuracy of 84.3% on the new benchmark when compared to other methods.
... Empirical studies have shown that performing visual-manual tasks while driving may degrade drivers' performance in many aspects such as steering control and lane keeping performance (5)(6)(7)(8)(9) , headway control and braking behavior (10)(11) , and response to sudden or hazard events (12)(13) . Efforts have been made to develop in-vehicle system that could monitor and detect driving distraction based on measures of different categories that include (1) a driver's eye movements (14)(15)(16) or head orientations (14)(15) , (2) driver maneuvers such as steering wheel angle (14,17) , throttle position (14) , (3) vehicle kinematics such as lane position (14)(15)(16) or speed (14) , and (4) a driver's involvement of a secondary task itself (17) . A comprehensive review of driver inattention monitoring systems that include distraction detection can be also found (18) . ...
... Empirical studies have shown that performing visual-manual tasks while driving may degrade drivers' performance in many aspects such as steering control and lane keeping performance (5)(6)(7)(8)(9) , headway control and braking behavior (10)(11) , and response to sudden or hazard events (12)(13) . Efforts have been made to develop in-vehicle system that could monitor and detect driving distraction based on measures of different categories that include (1) a driver's eye movements (14)(15)(16) or head orientations (14)(15) , (2) driver maneuvers such as steering wheel angle (14,17) , throttle position (14) , (3) vehicle kinematics such as lane position (14)(15)(16) or speed (14) , and (4) a driver's involvement of a secondary task itself (17) . A comprehensive review of driver inattention monitoring systems that include distraction detection can be also found (18) . ...
... To develop such systems, it may not be feasible to directly measure other drivers' eye movements, head orientations, or their inputs to the vehicle such as steering wheel angle or throttle position. However, the vehicle kinematics such as its lane keeping performance has shown to contain useful information regarding the driver's distraction states (5)(6)(7)(8)(9)(14)(15)(16) , and it could be potentially measured by the host vehicle from a distance. An anecdotal illustration is that when a driver sees a nearby vehicle making ...
Distracted driving has become an emerging concern for road safety in the past decade. Efforts have been made to develop in-vehicle active safety systems that could detect driver distraction. However, most methods focused on detecting a distracted driver of the host vehicle (ego-vehicle). Given that a distracted driver poses increased crash risk not only to him/herself but also to other road users, it may be beneficial to investigate ways to detect a distracted driver from a surrounding vehicle. This paper proposes a method to estimate the kinematics of a lead vehicle solely based on the sensory data from a host vehicle. The estimated kinematics of the lead vehicle include its lane position, lateral speed, longitudinal speed, and longitudinal acceleration, all of which may be potentially useful to detect distracted driving. The method was developed and validated using an existing naturalistic driving study, Safety Pilot Model Deployment, which collected a large scale of driving data in real-world roadways. The method utilizes signals from a camera-based Mobileye® system and other host vehicle sensory channels such as speed and yaw rate. Sensor fusion techniques were used to improve the accuracy of the estimation. The validation results show that the method was able to capture the lead vehicle’s kinematics within a fairly small error range. The method could be potentially used to develop in-vehicle systems that are able to monitor the behaviors of its surrounding vehicles and detect distracted or impaired driving.
... In the non-wearable device eye detection method in a vehicle environment, a camera, usually installed inside the vehicle, obtains images of the driver's face, and through these images, the driver's eyes are either directly detected or detected from a limited search region after the face is detected. These methods can be largely divided into the single-camera-based method [12-17] and the multiple-camera-based method [18][19][20][21][22][23][24][25][26][27]. In the single-camera-based method, as the driver's eyes are detected through the image information from a single camera, the complexity of the computation is relatively low. ...
... To overcome these disadvantages of the single-camera-based method, the multiple-camera-based methods have been studied. Existing studies [18,19] have proposed systems that check for driver inattention through two cameras installed in the vehicle. In particular, in [18], a faceLab eye-tracking system was used in the images of the driver obtained through the cameras installed on the dashboard and steering wheel to detect the user's eyes. ...
... Existing studies [18,19] have proposed systems that check for driver inattention through two cameras installed in the vehicle. In particular, in [18], a faceLab eye-tracking system was used in the images of the driver obtained through the cameras installed on the dashboard and steering wheel to detect the user's eyes. Through eye movement, a support vector machine (SVM) model was used to determine whether the driver was distracted. ...
Studies are being actively conducted on camera-based driver gaze tracking in a vehicle environment for vehicle interfaces and analyzing forward attention for judging driver inattention. In existing studies on the single-camera-based method, there are frequent situations in which the eye information necessary for gaze tracking cannot be observed well in the camera input image owing to the turning of the driver’s head during driving. To solve this problem, existing studies have used multiple-camera-based methods to obtain images to track the driver’s gaze. However, this method has the drawback of an excessive computation process and processing time, as it involves detecting the eyes and extracting the features of all images obtained from multiple cameras. This makes it difficult to implement it in an actual vehicle environment. To solve these limitations of existing studies, this study proposes a method that uses a shallow convolutional neural network (CNN) for the images of the driver’s face acquired from two cameras to adaptively select camera images more suitable for detecting eye position; faster R-CNN is applied to the selected driver images, and after the driver’s eyes are detected, the eye positions of the camera image of the other side are mapped through a geometric transformation matrix. Experiments were conducted using the self-built Dongguk Dual Camera-based Driver Database (DDCD-DB1) including the images of 26 participants acquired from inside a vehicle and the Columbia Gaze Data Set (CAVE-DB) open database. The results confirmed that the performance of the proposed method is superior to those of the existing methods.
... In past research, multiple camera-based methods were mostly used for the outdoor vehicle environment [21][22][23][24]. When dealing with the challenges of more peripheral gaze directions or large gaze coverage, multiple cameras may be the most suitable solution. ...
... Liang et al. observed driver distraction by using eye motion data in a support vector machine (SVM) model [22]. They compared it with a logistic regression model and found the SVM model performed better in identifying distraction. ...
... Multiple camera-based [21][22][23][24] Multiple cameras are used to classify driver's gaze region -Possibility of invisible eye region is reduced -The processing time is increased by the images of multiple cameras -Reliability is higher when information from multiple cameras is combined -Difficulties in applying to actual vehicular environment due to complicated, time-consuming calibration [21,22] Single camera-based ...
A paradigm shift is required to prevent the increasing automobile accident deaths that are mostly due to the inattentive behavior of drivers. Knowledge of gaze region can provide valuable information regarding a driver's point of attention. Accurate and inexpensive gaze classification systems in cars can improve safe driving. However, monitoring real-time driving behaviors and conditions presents some challenges: dizziness due to long drives, extreme lighting variations, glasses reflections, and occlusions. Past studies on gaze detection in cars have been chiefly based on head movements. The margin of error in gaze detection increases when drivers gaze at objects by moving their eyes without moving their heads. To solve this problem, a pupil center corneal reflection (PCCR)-based method has been considered. However, the error of accurately detecting the pupil center and corneal reflection center is increased in a car environment due to various environment light changes, reflections on glasses surface, and motion and optical blurring of captured eye image. In addition, existing PCCR-based methods require initial user calibration, which is difficult to perform in a car environment. To address this issue, we propose a deep learning-based gaze detection method using a near-infrared (NIR) camera sensor considering driver head and eye movement that does not require any initial user calibration. The proposed system is evaluated on our self-constructed database as well as on open Columbia gaze dataset (CAVE-DB). The proposed method demonstrated greater accuracy than the previous gaze classification methods.
... Substantial research has been done to define input variables for impairment detection algorithms based on eye movements, head position, and even facial expressions. Many algorithms have focused solely on eye tracking [2], and facial image analysis [3], whereas others combine multimodal features or include vehicle-based measures [4][6]. In a review of detection systems, Dong et al. [7] suggested that hybrid measures involving multiple modes perform best. ...
... Kaber et al. [17] examined the effects of visual and cognitive distraction on steering smoothness and headway time and found that drivers increased their headway time when visually distracted. Liang and Lee have mixed driver-based and vehicle-based signals to train cognitive distraction algorithms [6], [18]. ...
... Similar to the variety of distraction definitions and sensors, many different algorithms have been employed to detect impairment. These include traditional machine learning algorithms such as support vector machines (SVM) [6], [22], Neural Networks [27], graph based models such as Hidden Markov Models (HMM) [28], temporal graph based models [29], [30] and deep learning approaches [31]. All of these methods have been demonstrated with some degree of success and offer several promising directions for further development. ...
... Selected features and their types are shown inFig. 8. Eye movement features show high sensitivity to the driver cognitive distraction in both driving scenarios which agrees with previous study [30]. The cognitive distraction could be detected which indicated that the drivers changed in the both aspects of driving performance . ...
... Lee et al. [29] have used eye movements and vehicle parameters as inputs to the Bayesian Networks model for cognitive distraction detection. Liang et al. [30] have applied fixation, saccade, smoothing pursuit of the eye, steering-wheel angle, lane position, and steering error as inputs to real time SVM classifiers to detect the driver cognitive distraction caused by interacting with in-vehicle information systems (IVISs). The SVM classifier could detect driver distraction with average accuracy of 81.1%. ...
... As for the most important features for cognitive distraction detection, the top ten for the two driving scenarios are shown in Table VII. Selected features and their types are shown in Fig. 8. Eye movement features show high sensitivity to the driver cognitive distraction in both driving scenarios which agrees with previous study [30]. The cognitive distraction could be detected which indicated that the drivers changed in the both aspects of driving performance . ...
Driver distraction has been identified as one major cause of unsafe driving. The existing studies on cognitive distraction detection mainly focused on high-speed driving situations, but less on low-speed traffic in urban driving. This paper presents a method for the detection of driver cognitive distraction at stop-controlled intersections and compares its feature subsets and classification accuracy with that on a speed-limited highway. In the simulator study, 27 subjects were recruited to participate. Driver cognitive distraction is induced by the clock task that taxes visuospatial working memory. The support vector machine (SVM) recursive feature elimination algorithm is used to extract an optimal featuresubsetout of featuresconstructed from driving performance and eye movement. After feature extraction, the SVM classifier is trained and cross-validated within subjects. On average, the classifier based on the fusion of driving performance and eye movement yields the best correct rate and F-measure (correct rate = 95.8±4.4%; for stop-controlled intersections and correct rate = 93.7±5.0%; for a speed-limited highway) among four types of the SVM model based on different candidate features. The comparisons of extracted optimal feature subsets and the SVM performance between two typical driving scenarios are presented.
... For example, Zhang et al. [17] introduced a data-driven method and demonstrated promising capability of a simple classifier, i.e., decision tree, for cognitive workload estimation. Liang's group developed Support Vector Machines (SVM) based [18] and Bayesian Network based [19] detection systems for cognitive distraction detection using eye and driving behavior parameters. Long Short-Term Memory (LSTM) recurrent neural networks was adopted [20] because of the unique ability to learn the temporal contextual information. ...
... Human experts supervise the training by creating labels, which are the true distraction states corresponding to training data. In empirical studies, these labels are obtained in a costly way, e.g., additional subjective ratings by the driver [20], post-processing by the experimentalists [16], [21], [23], or additional computation based on data from other sources [18]. In a recent study [16], labeling the drivers' distraction state involves the development of Graphical User Interface (GUI), the training of external evaluators, and the actual labeling time, which is approximately 21.5 hours of manpower (43 minutes per evaluator × 30 evaluators) to label the entire 480 10-second video segments. ...
Monitoring drivers’ visual behavior using machine learning techniques has been identified as an effective approach to detect and mitigate driver distraction to enhance road safety. In our previous work, detection system based on supervised Extreme Learning Machine (ELM) was developed and tested with satisfactory performance. However, supervised ELM requires all training data to be labeled, which can be costly and time-consuming. This paper proposed and evaluated a semi-supervised distraction detection system based on Semi-Supervised Extreme Learning Machine (SS-ELM). The experimental results show that SS-ELM outperformed supervised ELM in both accuracy (95.5% for SS-ELM vs. 93.0% for ELM) and model sensitivity (97.6% for SS-ELM and 95.5% for ELM), suggesting that the proposed semi-supervised detection system can extract information from unlabeled data effectively to improve the performance. SS-ELM based detection system has the potential of improving accuracy and alleviating the cost of adapting distraction detection systems to new drivers, and thus is more promising for real world applications.
... pointing to a desired parking spot. Because of its high importance, the problem of gesture recognition for autonomous vehicles is not new to the community [1], [2], [3]. The current state-of-the-art for indoor [4] and outdoor [5] gesture recognition builds on deep neural networks. ...
Gesture recognition is essential for the interaction of autonomous vehicles with humans. While the current approaches focus on combining several modalities like image features, keypoints and bone vectors, we present neural network architecture that delivers state-of-the-art results only with body skeleton input data. We propose the spatio-temporal multilayer perceptron for gesture recognition in the context of autonomous vehicles. Given 3D body poses over time, we define temporal and spatial mixing operations to extract features in both domains. Additionally, the importance of each time step is re-weighted with Squeeze-and-Excitation layers. An extensive evaluation of the TCG and Drive&Act datasets is provided to showcase the promising performance of our approach. Furthermore, we deploy our model to our autonomous vehicle to show its real-time capability and stable execution.
... Until recently, the majority of driving observation frameworks comprised a manual feature extraction step followed by a classification module (for a thorough overview see [21]). The constructed feature vectors are often derived from hand-and body-pose [2], [3], [6], [7], [38], [39], facial expressions and eye-based input [40], [41], and head pose [42], [43], but also foot dynamics [44], detected objects [6], and physiological signals [45] have been considered. Classification approaches are fairly similar to the ones used in standard video classification, with LSTMs [3], [4], SVMs [2], [46], random forests [47] or HMMs [4], and graph neural networks [7], [48] being popular choices. ...
Driver observation models are rarely deployed under perfect conditions. In practice, illumination, camera placement and type differ from the ones present during training and unforeseen behaviours may occur at any time. While observing the human behind the steering wheel leads to more intuitive human-vehicle-interaction and safer driving, it requires recognition algorithms which do not only predict the correct driver state, but also determine their prediction quality through realistic and interpretable confidence measures. Reliable uncertainty estimates are crucial for building trust and are a serious obstacle for deploying activity recognition networks in real driving systems. In this work, we for the first time examine how well the confidence values of modern driver observation models indeed match the probability of the correct outcome and show that raw neural network-based approaches tend to significantly overestimate their prediction quality. To correct this misalignment between the confidence values and the actual uncertainty, we consider two strategies. First, we enhance two activity recognition models often used for driver observation with temperature scaling-an off-the-shelf method for confidence calibration in image classification. Then, we introduce Calibrated Action Recognition with Input Guidance (CARING)-a novel approach leveraging an additional neural network to learn scaling the confidences depending on the video representation. Extensive experiments on the Drive&Act dataset demonstrate that both strategies drastically improve the quality of model confidences, while our CARING model out-performs both, the original architectures and their temperature scaling enhancement, leading to best uncertainty estimates. Index Terms-Driver activity recognition, model confidence reliability, uncertainty in deep learning.
... Also, the related research is summarized in [96], [97]. Eye movements can be used to analyze whether the driver is distracted in real time [98]. Besides, a fatigue degree is a vital factor in evaluating the reliability of the driver. ...
With the continuous development of Artificial Intelligence (AI), autonomous driving has become a popular research area. AI enables the autonomous driving system to make a judgment, which makes studies on autonomous driving reaches a period of booming development. However, due to the defects of AI, it is not easy to realize a general intelligence, which also limits the research on autonomous driving. In this paper, we summarize the existing architectures of autonomous driving and make a taxonomy. Then we introduce the concept of hybrid human-artificial intelligence (H-AI) into a semi-autonomous driving system. For making better use of H-AI, we propose a theoretical architecture based on it. Given our architecture, we classify and overview the possible technologies and illustrate H-AI's improvements, which provides a new perspective for the future development. Finally, we have identified several open research challenges to attract the researchers for presenting reliable solutions in this area of research.
... There are limitations to this assumption in that a person's cognitive processing of an information icon can still be ongoing after the fixation has moved [53], [54]. However, driving is an inherently visual task and the majority of a driver's information is acquired visually [55], [56]consequently, the eye-mind assumption has been used and assumed to be valid in simulator studies before [57], [58]. ...
While partially automated vehicles can provide a range of benefits, they also bring about new Human Machine Interface (HMI) challenges around ensuring the driver remains alert and is able to take control of the vehicle when required. While humans are poor monitors of automated processes, specifically during ‘steady state’ operation, presenting the appropriate information to the driver can help. But to date, interfaces of partially automated vehicles have shown evidence of causing cognitive overload. Adaptive HMIs that automatically change the information presented (for example, based on workload, time or physiologically), have been previously proposed as a solution, but little is known about how information should adapt during steady-state driving. This study aimed to classify information usage based on driver experience to inform the design of a future adaptive HMI in partially automated vehicles. The unique feature of this study over existing literature is that each participant attended for five consecutive days; enabling a first look at how information usage changes with increasing familiarity and providing a methodological contribution to future HMI user trial study design. Seventeen participants experienced a steady-state automated driving simulation for twenty-six minutes per day in a driving simulator, replicating a regularly driven route, such as a work commute. Nine information icons, representative of future partially automated vehicle HMIs, were displayed on a tablet and eye tracking was used to record the information that the participants fixated on. The results found that information usage did change with increased exposure, with significant differences in what information participants looked at between the first and last trial days. With increasing experience, participants tended to view information as confirming technical competence rather than the future state of the vehicle. On this basis, interface design recommendations are made, particularly around the design of adaptive interfaces for future partially automated vehicles.
... Studies indicated that the use of cell phones among all drivers increases the risk of a crash by a factor of four [67], [76]. Similarly, another study using a simulator involving adolescent drivers showed that texting while driving increases the frequency of deviations in a lane in relation to the position from the centerline [77]. ...
... By merging the information obtained from vehicle's parameters (e.g., turning speed, and acceleration) and driver's physical and biological parameters, more accurate and reliable results are reported. For example, the authors of [194] reported the distraction detection accuracy to be 81.1% by fusing the data of saccades, eye fixation, lateral control, and steering wheel through a support vector machine algorithm. The authors of [195] detected driver's distraction by processing the information obtained from physical (blink frequency, location, and eye-fixation duration) and driving performance parameters (steering wheel and lateral control). ...
Tracking drivers’ eyes and gazes is a topic of great interest in the research of advanced driving assistance systems (ADAS). It is especially a matter of serious discussion among the road safety researchers’ community, as visual distraction is considered among the major causes of road accidents. In this paper, techniques for eye and gaze tracking are first comprehensively reviewed while discussing their major categories. The advantages and limitations of each category are explained with respect to their requirements and practical uses. In another section of the paper, the applications of eyes and gaze tracking systems in ADAS are discussed. The process of acquisition of driver’s eyes and gaze data and the algorithms used to process this data are explained. It is explained how the data related to a driver’s eyes and gaze can be used in ADAS to reduce the losses associated with road accidents occurring due to visual distraction of the driver. A discussion on the required features of current and future eye and gaze trackers is also presented.
... In addition, artificial intelligence methods were also successfully applied to ensure pedestrians [15] or environmental [16], [17] safety jeopardize by vehicle, and also to improve driver safety threatened by distracted driving [18], [19]. Amditis et al. developed a simple DVE conditions prediction model, "Basic Indicators of Driver Operational Navigation" [20]. ...
Transportation systems are invariably burdened with dynamically changing environmental conditions and ill–defined human factor. To raise ground vehicle safety on a new supreme level and to boost autonomous vehicles development driver–vehicle–environment cooperation is inevitable. In this paper, an overview of several existing driver–vehicle–environment integration methods with purpose of vehicle safety enhancement are stressed. Five unique and fundamentally different solutions are proposed, which have common similarity: the solutions are accomplished with machine learning algorithms. The methods aim at modelling drivers’ or vehicles’ behaviour with reasonable prediction accuracy under various complex scenarios. All five solutions are developed in individual projects in a framework of a continuous interdisciplinary European network ITEAM. The aim of the paper is to underline significant benefit of man–machine–environment integration in vehicle safety systems by exploiting fairly received tremendous attention machine learning methods
... Por conta da invasividade na aquisição de sinais elétricos de motoristas, métodos de identificação de sonolência baseados em informações do veículo foram criados para avaliar o estado do condutor com base em parâmetros como velocidade [11], aceleração e frenagem [12], posição na faixa [13] e angulação do volante [14]. Embora essa abordagem não seja desconfortável, ela é considerada lenta para prever o estado de sonolência [15]. ...
Vehicular accidents caused by driver drowsinessinvolve about 7,000 people/year in Brazil, only on federalhighways, and cause psychic damage and traumatic stressesfor both the victims involved and their families. Drowsiness ischaracterized by reduced level of vigilance and concentration,which are essential during driving activity. Due to this adversity,many applications of drowsiness detection had been continuouslydeveloped through electrical body signals to alert the driverat the time when sleepiness is identified, such as heart ratevariability (HRV) and electroencephalogram (EEG). Althoughthese methods work, the use of electrodes in the driver’s bodyis highly invasive. Therefore, we propose a drowsiness detectionsystem based on driver’s real time video capture, by estimatingthe percentage of eyelid closure over a period, without anycontact device. Since the use of smartphones has been growingin the last decade, the system has been implemented in a mobilephone even with memory and processing limitations. Processingreduction procedures were developed to improve the applicationperformance, such as the reduction of the region of interest andthe limitation of the search window, which increased by 93.09%the number of frames per second and allowed the application tooperate smoothly.
... Trusted autonomous operations place significant emphasis on humanmachine teaming, requiring human operators to work collaboratively with intelligent and oftentimes autonomous systems in order to process more complex and larger quantities of information, thereby supporting more precise and timely decisionmaking [1][2][3]. As part of this ongoing evolution, elements of artificial intelligence and machine learning are being incorporated to augment the data collection, information processing and decision-making functions [4][5][6][7]. These enhancements are expected to particularly impact the evolution of Human-Machine Interfaces and Interactions (HMI2). ...
Adaptive Human-Machine Interfaces and Interactions (HMI²) are closed-loop cyber-physical systems comprising a network of sensors measuring human, environmental and mission parameters, in conjunction with suitable software for adapting the HMI² (command, control and display functions) in response to these real-time measurements. Cognitive HMI² are a particular subclass of these systems, which support dynamic HMI² adaptations based on the user's cognitive states These states are estimated in real-time using various neuro-physiological parameters including gaze, cardiorespiratory and brain signals, which are processed by an Adaptive Neuro-Fuzzy Inference System (ANFIS). However, the accuracy and precision of biometric measurements are affected by a variety of environmental factors and therefore need to be accurately characterised prior to operational use. This paper describes the characterisation activities performed on two types of eye tracking devices used in the Aerospace Intelligent and Autonomous Systems (AIAS) laboratory of RMIT University to support the development of cognitive human-machine systems. The uncertainty associated with the ANFIS outputs is quantified by propagating the uncertainties of the input data (determined experimentally), through the inference engine. This process is of growing relevance because similar machine learning techniques are now being developed for an increasing number of applications including aerospace, transport, biomedical and defence cyber-physical systems.
... In situations where safety criteria are not met, they must be re-evaluated 504 until they are met, represented in the model by the conditional loop. Ways to automatically evaluate these safety 505 criteria exist, such as in systems that monitor on-coming traffic (Curry et al., 2010), label nearby road users 506 (Ashraf et al., 2016), detect weather conditions (Green, 2004), and systems that attempt to estimate driver state 507 (Ferreira et al., 2014;Liang, Reyes, & Lee, 2007;van Gent, Melman, et al., 2018b). ...
Persuasive in-vehicle systems aim to intuitively influence the attitudes and/or behaviour of a driver (i.e. without forcing them). However, the challenge in using these systems in a driving setting, is to maximise the persuasive effect without infringing upon the driver’s safety.
This paper proposes a conceptual model for driver persuasion at the tactical level (i.e., driver manoeuvring level, such as lane-changing and car-following). The main focus of the conceptual model is to describe how to safely persuade a driver to change his or her behaviour, and how persuasive systems may affect driver behaviour.
First, existing conceptual and theoretical models that describe behaviour are discussed, along with their applicability to the driving task. Next, we investigate the persuasive methods used with a focus on the traffic domain. Based on this we develop a conceptual model that incorporates behavioural theories and persuasive methods, and which describes how effective and safe driver persuasion functions. Finally, we apply the model to a case study of a lane-specific advice system that aims to reduce travel time delay and traffic congestion by advising some drivers to change lanes in order to achieve a better distribution of traffic over the motorway lanes.
... In the field of driving behavior, many studies apply SVM to analyze drowsiness state [24][25][26]. In addition, the study of Chen [27] shows that SVM can distinguish abnormal driving behaviors from normal driving behaviors. ...
Aggressive driving, amongst inappropriate driving behaviors, is largely responsible for leading to traffic accidents, which threatens both safety and property of human beings. With the objective to reduce traffic accidents and improve road safety, effective and reliable aggressive driving recognition methods, which enables the development of driving behavior analysis and early warning systems, are urgently needed. Most recently, the research focus of aggressive recognition has shifted to the use of vehicle motion data, which has emerged as a new tool for traffic phenomenon explanation. As aggressive driving corresponds to sudden variations in data, they can be recognized based on the recorded vehicle motion data. In this paper, several kinds of anomaly recognition algorithms are studied and compared, using the motion data collected by the accelerometer and gyroscope of a smartphone mounted on vehicle. Gaussian mixture model (GMM), Partial Least Squares Regression (PLSR), Wavelet Transformation (WT), Support Vector Regression (SVR) are considered as the representative algorithms of statistical regression, time series analysis, and machine learning, respectively. These algorithms are evaluated by the three widely used validation metrics, including F1-score, Precision, and Recall. Empirical results show that GMM, PLSR and SVR are promising methods for aggressive driving recognition. GMM and SVR outperform PLSR when only single-source dataset is used. The improvement of F1-score is almost 0.1. PLSR performs the best when multi-source datasets are used, and the F1-score is 0.77. GMM and SVR are more robust to hyperparameter. In addition, incorporating multi-source datasets helps improve the accuracy of aggressive driving behavior recognition.
... Focusing on the methodological framework of the research, a key remark, concerns the measures used to express driving performance in driver distraction studies and in general. The parameters for assessing driving performance vary significantly, and the driving-related outcomes have been analyzed in several studies as presented below: speed (Beede & Kas, 2006;Collet, Guillot, & Petit, 2010;Yannis et al., 2010), accident probability ( Caird, Johnston, Willness, & Asbridge, 2014), lane position (Engstr€ om, Johansson, & € Ostlund, 2005;Horrey & Wickens, 2006;Liang & Lee, 2010), number of eye glances (Liang, Reyes, & Lee, 2007), headway (Ranney, Harbluk, & Noy, 2005;Strayer, Drews, & Johnston, 2003), reaction time (Hancock, Lesch, & Simmons, 2003;Horrey & Wickens, 2006;Ishigami & Klein, 2009). Certainly, a more holistic approach would be beneficial, whereby many independent variables used in concert will describe the overall performance capturing the effect of many variables together with their interrelationships. ...
Considering that unexpected events are a major contributory factor of road accidents the main objective of this article is to investigate the effect of several parameters including overall driving performance, distraction sources, driver characteristics, as well as road and traffic environment on accident probability at unexpected incidents. For this purpose, a driving simulator experiment was carried out, in which 95 participants from all age groups were asked to drive under different types of distraction (no distraction, conversation with passenger, cell phone use) in different road and traffic conditions. Then, in the framework of the statistical analysis, driving performance is estimated as a new unobserved (latent) variable based on several individual driving simulator parameters while a structural equation model is developed investigating which factors lead to increased accident probability at unexpected incidents. Regarding driver distraction, results indicate that cell phone use has a negative effect on accident risk confirming the initial hypothesis that when talking on the cell phone drivers find it difficult to handle an unexpected incident and as a result are more likely to commit an accident. Overall, a risky driving profile is developed, completing the puzzle of the effect of driver distraction on driver behavior and road safety. © 2018 Taylor & Francis Group, LLC and The University of Tennessee
... The SVM algorithm generates a linear hyperplane and divides the two classes with the maximal margin between the two categories within this hyperplane. As mentioned in [20], the SVM method has two main advantages. First, only few samples are needed for training of SVM in high-dimensional spaces. ...
In this contribution, a combination of two methods for the evaluation of the best behavior including situated decisions based on the assessment of human actions or decisions related to actions is presented. The suggestion of the behavior is based on a Situation-OperatorModeling approach in combination with a stochastic state machine. While the Situation- Operator-Modeling (SOM) allows the concrete, individualized, and situated modeling of human interaction in formal context, an approach is used to work with a stochastic state automata. First based on observations (training), the observed transition probability between predefined states. The combination of both practical and theoretical approaches allows the combined prediction of human behavior.
... In the last decade, using infotainment system and digital devices during driving increased significantly. They exacerbated the problem by adding a new source of distraction [1,3]. A recent study [4] estimates the effect of distraction on accidents among teenage drivers at a much higher rate of 58%. ...
... As shown with the DRT, cognitive load and its effect on attention can be successfully measured by observing secondary tasks performance indicators, such as response times (time completion times) and hit rates (performance success). Driving performance is also observed as an indicator of changes in cognitive load [40][41][42]. One such method that uses degradation of driving performance is the Lane Change Task. ...
The Detection-Response Task is a method for assessing the attentional effects of cognitive load in a driving environment. Drivers are presented with a sensory stimulus every 3-5 s, and are asked to respond to it by pressing a button attached to their finger. Response times and hit rates are interpreted as indicators of the attentional effect of cognitive load. The stimuli can be visual, tactile and auditory, and are chosen based on the type of in-vehicle system or device that is being evaluated. Its biggest disadvantage is that the method itself also affects the driver's performance and secondary task completion times. Nevertheless, this is an easy to use and implement method, which allows relevant assessment and evaluation of in-vehicle systems. By following the recommendations and taking into account its limitations, researchers can obtain reliable and valuable results on the attentional effects of cognitive load on drivers.
... Moreover, there is no direct link between all these features and the "operational state", which is why methods such as machine learning or statistical models are used, combining the different measures. The different algorithms used include k-nearest neighbors (Chauhan et al., 2015), decision trees (Lee et al., 2010;Sukanesh and Vijayprasath, 2013), Bayesian classifiers (Lee and Chung, 2012;Yang et al., 2010), Support Vector Machines (Bhowmick and Chidanand Kumar, 2009;Krajewski et al., 2009a,b;Liang et al., 2007;Yeo et al., 2009), artificial neural networks (ANN) (Bundele and Banerjee, 2009;Eskandarian et al., 2007;Sayed and Eskandarian, 2001;Samiee et al., 2014), ensemble methods like random forest (Krajewski et al., 2009a,b;McDonald et al., 2013;Torkkola et al., 2008;Zhang et al., 2004) and, more recently, deep learning (Hajinoroozi et al., 2015). Most studies consider the problem of estimating the driver's impaired operational state as a classification problem. ...
Not just detecting but also predicting impairment of a car driver's operational state is a challenge. This study aims to determine whether the standard sources of information used to detect drowsiness can also be used to predict when a given drowsiness level will be reached. Moreover, we explore whether adding data such as driving time and participant information improves the accuracy of detection and prediction of drowsiness. Twenty-one participants drove a car simulator for 110min under conditions optimized to induce drowsiness. We measured physiological and behavioral indicators such as heart rate and variability, respiration rate, head and eyelid movements (blink duration, frequency and PERCLOS) and recorded driving behavior such as time-to-lane-crossing, speed, steering wheel angle, position on the lane. Different combinations of this information were tested against the real state of the driver, namely the ground truth, as defined from video recordings via the Trained Observer Rating. Two models using artificial neural networks were developed, one to detect the degree of drowsiness every minute, and the other to predict every minute the time required to reach a particular drowsiness level (moderately drowsy). The best performance in both detection and prediction is obtained with behavioral indicators and additional information. The model can detect the drowsiness level with a mean square error of 0.22 and can predict when a given drowsiness level will be reached with a mean square error of 4.18min. This study shows that, on a controlled and very monotonous environment conducive to drowsiness in a driving simulator, the dynamics of driver impairment can be predicted.
... In another study, Weller and Schlag [78] used longitudinal deceleration, lateral acceleration, and speed as driving performance measures, and the same physical measures as Liang and Lee [77]. A support vector machine with an accuracy of 81.1% was used by Liang and colleagues to detect distraction using information from the steering wheel, lateral control, eye fixations, and saccades [79]. A study by Miyaji et al. [80] was the only study that used biological measures such as heart rate, and physical measures such as eye gaze, head orientation, and pupil diameter. ...
Driving distraction is a topic of great interest in the transport safety-research community because it is now a primary cause of road accidents. A recent report has revealed that distraction is more alarming than previously thought, and a suitable measurement to effectively detect distraction is required. Most agree that driving distraction actually comprises the simultaneous interaction of two or more types of distraction. The purpose of this paper is therefore to determine the promising method for measuring visual cognitive distraction. We discuss the five common measurement methods for visual and cognitive driving distraction, which include driving performance, driver physical measures, driver biological measures, subjective reports, and hybrid measures. Hybrid measurement of driver’s physical measures (e.g., eye movement) and driver’s biological measures (e.g., electroencephalogram) is better than other methods at detecting types of visual cognitive distraction. This new perspective on measurement methods will help the field of transport safety to determine the best means of detecting and measuring the effect of visual cognitive distraction.
... A typical method for measuring visual performance is to use eye tracking techniques ( Pradhan et al., 2005), though video recording is often used as well. In the field of traffic research, many studies on drivers' eye movements under different driving situations have already been conducted, such as drivers' characteristics of different experience or age (e.g., Dukic and Broberg, 2012;Crundall et al., 2004;Crundall and Underwood, 1998), different weather conditions (e.g., Konstantopoulos et al., 2010), drowsy and distracted driving (e.g., Zeng et al., 2010;Hu and Zheng, 2009;Miyaji et al., 2009;Hong et al., 2007;Liang et al., 2007;Hayami et al., 2002), etc. In the ISO 15007-1 and 15007-2 standards, the eye movement behavior can be analyzed based on either by the glancebased measures or fixations-based measures when supported by an eye-tracker with sufficient accuracy and precision. ...
... Vehiclebased approaches collect signal data from sensors in vehicles to evaluate driver's performance. These methods monitor the variations of steering wheel angle, lane position, speed, acceleration , and braking to predict the driver fatigue [17], [18], [19], [20], [21]. It is convenient to collect vehicle signals. ...
Driver's fatigue is one of the major causes of traffic accidents, particularly for drivers of large vehicles (such as buses and heavy trucks) due to prolonged driving periods and boredom in working conditions. In this paper, we propose a vision-based fatigue detection system for bus driver monitoring, which is easy and flexible for deployment in buses and large vehicles. The system consists of modules of head-shoulder detection, face detection, eye detection, eye openness estimation, fusion, drowsiness measure percentage of eyelid closure (PERCLOS) estimation, and fatigue level classification. The core innovative techniques are as follows: 1) an approach to estimate the continuous level of eye openness based on spectral regression; and 2) a fusion algorithm to estimate the eye state based on adaptive integration on the multimodel detections of both eyes. A robust measure of PERCLOS on the continuous level of eye openness is defined, and the driver states are classified on it. In experiments, systematic evaluations and analysis of proposed algorithms, as well as comparison with ground truth on PERCLOS measurements, are performed. The experimental results show the advantages of the system on accuracy and robustness for the challenging situations when a camera of an oblique viewing angle to the driver's face is used for driving state monitoring.
... have been adopted to solve many problems in the transportation domain, such as real-time detection of 160 driver cognitive distraction (Liang et al. 2007), lane detection and tracking (Kim 2008), transportation 161 mode recognition (Jahangiri and Rakha), traffic sign detection (Balali and Golparvar-Fard), and incident 162 detection (Yuan and Cheu 2003). These studies illustrate that applying AI methods can lead to promising 163 results. ...
In the United States, 683 people were killed and an estimated 133,000 were injured in crashes due to running red lights in 2012. To help prevent/mitigate crashes caused by running red lights, these violations need to be identified before they occur, so both the road users (i.e., drivers, pedestrians, etc.) in potential danger and the infrastructure can be notified and actions can be taken accordingly. Two different data sets were used to assess the feasibility of developing red-light running (RLR) violation prediction models: (1) observational data and (2) driver simulator data. Both data sets included common factors, such as time to intersection ( ), distance to intersection ( ), and velocity at the onset of the yellow indication. However, the observational data set provided additional factors that the simulator data set did not, and vice versa. The observational data included vehicle information (e.g., speed, acceleration, etc.) for several different time frames. For each vehicle approaching an intersection in the observational data set, required data were extracted from several time frames as the vehicle drew closer to the intersection. However, since the observational data were inherently anonymous, driver factors such as age and gender were unavailable in the observational data set. Conversely, the simulator data set contained age and gender. In addition, the simulator data included a secondary (non-driving) task factor and a treatment factor (i.e., incoming/outgoing calls while driving). The simulator data only included vehicle information for certain time frames (e.g., yellow onset); the data did not provide vehicle information for several different time frames while vehicles were approaching an intersection. In this study, the random forest (RF) machine-learning technique was adopted to develop RLR violation prediction models. Factor importance was obtained for different models and different data sets to show how differently the factors influence the performance of each model. A sensitivity analysis showed that the factor importance to identify RLR violations changed when data from different time frames were used to develop the prediction models. , , the required deceleration parameter ( ), and velocity at the onset of a yellow indication were among the most important factors identified by both models constructed using observational data and simulator data. Furthermore, in addition to the factors obtained from a point in time (i.e., yellow onset), valuable information suitable for RLR violation prediction was obtained from defined monitoring periods. It was found that period lengths of 2–6 m contributed to the best model performance.
... For example, electrical heart activity and blood pressure can be used to objectively measure levels of mental workload [16], stress [9], and mental fatigue [23]. Eyelid activity tracking was used to estimate level of mental fatigue [21] and detect visually distracted driving [24]. [25] Indeed, the driver state is truly multidimensional. ...
The purpose of this paper is to bring together multiple literature sources which present innovative methodologies for the assessment of driver state, driving context and performance by means of technology within a vehicle and consumer electronic devices. It also provides an overview of ongoing research and trends in the area of driver state monitoring. As part of this review a model of a hybrid driver state monitoring system is proposed. The model incorporates technology within a vehicle and multiple brought- in devices for enhanced validity and reliability of recorded data. Additionally, the model draws upon requirement of data fusion in order to generate unified driver state indicator(-s) that could be used to modify in-vehicle information and safety systems hence, make them driver state adaptable. Such modification could help to reach optimal driving performance in a particular driving situation. To conclude, we discuss the advantages of integrating hybrid driver state monitoring system into a vehicle and suggest future areas of research.
... The cellular systems introduced a decade ago ( Bolla and Davoli, 20 0 0; Ygnace and Drane, 2001; Zhao, 2000 ) offer a resolution to the cost and coverage problems ( ). Nevertheless , their application is prohibited or discouraged in many countries, because the use of cell phones while driving disrupts the drivers' attention ( Liang et al., 2007 ). Global positioning systems (GPSs) are another promising means of collecting traffic data from almost the entire network at a relatively low cost ( Miwa et al., 2013 ). ...
Reliable transport models calibrated from accurate traffic data are crucial for predicating transportation system performance and ensuring better traffic planning. However, due to the impracticability of collecting data from an entire population, methods of data inference such as the linear data projection are commonly adopted. A recent study has shown that systematic bias may be embedded in the parameters calibrated due to linearly projected data that do not account for scaling factor variability. Adjustment factors for reducing such biases in the calibrated parameters have been proposed for a generalized multivariate polynomial model. However, the effects of linear data projection on the dispersion of and confidence in the adjusted parameters have not been explored. Without appropriate statistics examining the statistical significance of the adjusted model, their validity in applications remains unknown and dubious. This study reveals that heteroscedasticity is inherently introduced by data projection with a varying scaling factor. Parameter standard errors that are estimated by linearly projected data without any appropriate treatments for non-homoscedasticity are definitely biased, and possibly above or below their true values. To ensure valid statistical tests of significance and prevent exposure to uninformed and unnecessary risk in applications, a generic analytical distribution-free (ADF) method and an equivalent scaling factor (ESF) method are proposed to adjust the parameter standard errors for a generalized multivariate polynomial model, based on the reported residual sum of squares. The ESF method transforms a transport model into a linear function of the scaling factor before calibration, which provides an alternative solution path for achieving unbiased parameter estimations. Simulation results demonstrate the robustness of the ESF method compared with the ADF method at high model nonlinearity. Case studies are conducted to illustrate the applicability of the ESF method for the parameter standard error estimations of six Macroscopic Bureau of Public Road functions, which are calibrated using real-world global positioning system data obtained from Hong Kong.
... One of the key points in ADAS applications are automatic and intelligent algorithms and strategies such as machine learning. Among the multiple tasks that might take advantage of these useful techniques are: Detection and identification of road signs [1], vehicle [2] and pedestrian [3] detection, estimation of driver distraction [4], and environment interpretation and understanding [5]. The present work is focus on the extraction and detection of road lane delimitation lines. ...
According to the Department for Transport statistics in UK, around 100.000 accidents were reported in 2013 [13], and almost 25% of them were related to impairment or distraction factors. Advanced Driver Assistance Systems (ADAS) are a powerful tool for road safety that can help to mitigate this problem. This paper presents a robust road lane detection and classification algorithm, one of the most important tasks in ADAS. This paper describes a road line detection algorithm based on a segmentation algorithm designed according to the constraints defined in the legal regulation for road marks. Later, pairs of lines, separated a fixed distance, are searched in the bird view of the road image. The bird view transformation is applied to the captured images, using the extrinsic parameters estimation algorithm reported in [10]. After the extraction of the road lines profiles, they are characterized using a specifically designed descriptor based on both space and frequency values. The descriptors are used in the supervised training of a Support Vector Machines classifier, whose performance is compared against the previous version of the module, a heuristic based approach. The performed tests showed a considerable increase of the system performance using the SVM approach, in comparison with the previous heuristic approach.
... The available machine learning algorithms, computation power and data sets from fixed detectors or data probes and intelligent transportation systems (ITSs) encourages Transportation engineers to apply machine learning in their field. Recently, some machine learning algorithms were used in the transportation field, including: classifying and counting vehicles detected by multiple inductive loop detectors [13], identifying motorway rear-end crash risks using disaggregate data [14], real-time detection of driver distraction [15, 16], and transportation mode recognition using smartphone sensor data [17, 18]. Modeling driver stop/run behavior at signalized intersections is very important and is ideal for applying machine learning techniques [19]. ...
The ability to classify driver stop/run behavior at signalized intersections considering the vehicle type and roadway surface conditions is critical in the design of advanced driver assistance systems. Such systems can reduce intersection crashes and fatalities by predicting driver stop/run behavior. The research presented in this paper uses data collected from three controlled field experiments and one data set collected using truck simulator. The field experiments are done on the Smart Road at the Virginia Tech Transportation Institute (VTTI) to model driver stop/run behavior at the onset of a yellow indication for different roadway surface conditions and different vehicle type. The paper offers two contributions. First, it introduces a new predictor related to driver aggressiveness and demonstrates that this measure enhances the modeling of driver stop/run behavior. Second, it applies well-known Artificial Intelligence techniques including: adaptive boosting (adaboost), artificial neural networks (ANN), and Support Vector Machine (SVM) algorithms on the data in order to develop a model that can be used by traffic signal controllers to predict driver stop/run decisions in a connected vehicle environment. The research demonstrates that by adding the driver aggressiveness predictor to the model, the increase in the model accuracy is significant for all models except SVM. However, the reduction in the false alarm rate was not statistically significant when using any of the approaches.
... In the recent literature, information from multiple eye-activity variables has been used to make reliable predictions of a given cognitive state or human factor (Tsai et al. 2007; Regis et al. 2012; Li et al. 2013; Jang et al. 2014). Specifically, multivariate machine learning techniques have been successfully applied to combine a variety of measures from eye-activity data (Liang et al. 2007; Marshall 2007; Miyaji et al. 2009; Jang et al. 2014; Jaques et al. 2014). Examples of algorithms that have been employed in this aim include linear discriminant analysis (LDA) (Richstone et al. 2010; Jang et al. 2014), support vector machines (SVM) (Jang et al. 2014; Jaques et al. 2014), Bayesian inference systems (Diard et al. 2013; Jaques et al. 2014), Neural ...
Recent advances in the reliability of the eye-tracking methodology as well as the increasing availability of affordable non-intrusive technology have opened the door to new research opportunities in a variety of areas and applications. This has raised increasing interest within disciplines such as medicine, business and education for analysing human perceptual and psychological processes based on eye-tracking data. However, most of the currently available software requires programming skills and focuses on the analysis of a limited set of eye-movement measures (e.g., saccades and fixations), thus excluding other measures of interest to the classification of a determined state or condition. This paper describes 'EALab', a MATLAB toolbox aimed at easing the extraction, multivariate analysis and classification stages of eye-activity data collected from commercial and independent eye trackers. The processing implemented in this toolbox enables to evaluate variables extracted from a wide range of measures including saccades, fixations, blinks, pupil diameter and glissades. Using EALab does not require any programming and the analysis can be performed through a user-friendly graphical user interface (GUI) consisting of three processing modules: 1) eye-activity measure extraction interface, 2) variable selection and analysis interface, and 3) classification interface.
... Consider a vehicle approaching a signalized intersection, our goal is to build a model that 100 [14], real-time detection of driver distraction [15, 16], transportation mode recognition 122 using smartphone sensor data [17], and video-based highway asset segmentation and recognition 123 [18]. Modeling driver stop/run behavior at signalized intersections is very important and is ideal 124 for applying machine learning techniques [19]. ...
The ability to model driver stop/run behavior at signalized intersections considering the roadway surface condition is critical in the design of advanced driver assistance systems. Such systems can reduce intersection crashes and fatalities by predicting driver stop/run behavior. The research presented in this paper uses data collected from two controlled field experiments on the Smart Road at the Virginia Tech Transportation Institute (VTTI) to model driver stop/run behavior at the onset of a yellow indication for different roadway surface conditions. The paper offers two contributions. First, it introduces a new predictor related to driver aggressiveness and demonstrates that this measure enhances the modeling of driver stop/run behavior. Second, it applies well-known artificial intelligence techniques including: adaptive boosting (AdaBoost), random forest, and support vector machine (SVM) algorithms as well as traditional logistic regression techniques on the data in order to develop a model that can be used by traffic signal controllers to predict driver stop/run decisions in a connected vehicle environment. The research demonstrates that by adding the proposed driver aggressiveness predictor to the model, there is a statistically significant increase in the model accuracy. Moreover the false alarm rate is significantly reduced but this reduction is not statistically significant. The study demonstrates that, for the subject data, the SVM machine learning algorithm performs the best in terms of optimum classification accuracy and false positive rates. However, the SVM model produces the best performance in terms of the classification accuracy only.
Copyright © 2015 Elsevier Ltd. All rights reserved.
... SVM has been successfully applied to several applications including text categorization, bioinformatics and database marketing [22]. It has also been used recently in the active safety research, including lane departure warning systems [17] and driver distraction detection algorithms [23]. The reader is encouraged to refer to reference [24] for a detailed description of SVM. ...
The ability to classify driver behavior lays the foundation for more advanced driver assistance systems. Improving safety at intersections has also been identified as high priority due to the large number of intersection related fatalities. This paper focuses on developing algorithms for estimating driver behavior at road intersections. It introduces two classes of algorithms that can classify drivers as compliant or violating. They are based on 1) Support Vector Machines (SVM) and 2) Hidden Markov Models (HMM), two very popular machine learning approaches that have been used extensively for classification in multiple disciplines. The algorithms are successfully validated using naturalistic intersection data collected in Christiansburg, VA, through the US Department of Transportation Cooperative Intersection Collision Avoidance System for Violations (CICAS-V) initiative.
... Driver's gaze was estimated using head motion and eye states to detect distracted driving. As another example, Liang et al. (2007) collected driver's eye movements and driving performance data through driving simulators to detect distracted driving. Miyaji et al. (2009) designed a topic conversation experiment and carried it out on a driving simulator to study the distracted driving. ...
Drowsy/distracted driving has become one of the leading causes of traffic crash. Only certain particular drowsy/distracted driving behaviors have been studied by previous studies, which are mainly based on dedicated sensor devices such as bio and visual sensors. The objective of this study is to extract the common features for identifying drowsy/distracted driving through a set of common vehicle motion parameters. An intelligent vehicle was used to collect vehicle motion parameters. Fifty licensed drivers (37 males and 13 females, M=32.5 years, SD=6.2) were recruited to carry out road experiments in Wuhan, China and collecting vehicle motion data under four driving scenarios including talking, watching roadside, drinking and under the influence of drowsiness. For the first scenario, the drivers were exposed to a set of questions and asked to repeat a few sentences that had been proved valid in inducing driving distraction. Watching roadside, drinking and driving under drowsiness were assessed by an observer and self-reporting from the drivers. The common features of vehicle motions under four types of drowsy/distracted driving were analyzed using descriptive statistics and then Wilcoxon rank sum test. The results indicated that there was a significant difference of lateral acceleration rates and yaw rate acceleration between "normal driving" and drowsy/distracted driving. Study results also shown that, under drowsy/distracted driving, the lateral acceleration rates and yaw rate acceleration were significantly larger from the normal driving. The lateral acceleration rates were shown to suddenly increase or decrease by more than 2.0m/s(3) and the yaw rate acceleration by more than 2.5°/s(2). The standard deviation of acceleration rate (SDA) and standard deviation of yaw rate acceleration (SDY) were identified to as the common features of vehicle motion for distinguishing the drowsy/distracted driving from the normal driving. In order to identify a time window for effectively extracting the two common features, a double-window method was used and the optimized "Parent Window" and "Child Window" were found to be 55s and 6s, respectively. The study results can be used to develop a driving assistant system, which can warn drivers when any one of the four types of drowsy/distracted driving is detected.
Copyright © 2015. Published by Elsevier Ltd.
Driver activity engagement while driving plays a vital role that leads to negative outcomes of driving safety. To reduce traffic accidents and ensure driving safety, real-time driver activity recognition architecture is proposed in this study. Specifically, a total of eight kinds of common driving-related activities are identified, which include the normal driving, left or right checking, texting, answering the phone, using media, drinking, and picking up objects. Raw experiment videos are collected via onboard monocular cameras, which are used for the upper body skeleton information extraction of the driver. Then, the graph convolu-tional networks (GCN) are constructed for spatial structure feature reasoning in a single frame, which is consecutively followed by long short-term memory (LSTM) networks for temporal motion feature learning within the sequence. Moreover, the attention mechanism is further utilised to emphasise the keyframes to select discriminative sequential information. Finally, a large-scale driver activity dataset, consisting of both naturalistic driving data and simulative driving data, is collected for model training and evaluations. Experimental results show that the general recall ratio of those eight driving-related activities reaches up to 88.8% and the real-time recognition efficiency can reach up to 24 fps, which would satisfy the real-time requirements of engineering applications.
This paper aims at two problems in fatigue expression recognition. First, texture features extracted by LBP (Local Binary Patterns) descriptor are limited and can not effectively describe the edge and direction information of image. Second, structural features extracted by HOG (Histogram of Oriented Gradient) descriptor are redundant and its computational complexity is high. To fill the gaps of these two problems, we proposed a reconstructed LBPHOG (LBP-RHOG) algorithm which extracted texture spectrum features and edge features from LBP operator and reconstructed HOG operator respectively and obtain fusion information by fusing these two features. To better evaluate the recognition performance, we complete simulation under a self-built fatigue expression database. The results show that our method has low computational complexity and high recognition rate, and can identify fatigue state well.
http://hdl.handle.net/11285/629753
Most recent efforts made in the industry to the path for a full autonomous vehicle have been focusing on the automation of the vehicles’ maneuvers, and the understanding of the surroundings. While a great advance has been achieved, the most advanced implementations of such systems may be only at the scale defined by the US National Highway Traffic Safety Administration as a level 2: “automation of at least two primary control functions”. Such systems require the driver to keep their hands on the steering wheel at all the time. One popular example for this is Tesla Motors ́ “Autopilot” feature, which is in fact just a diver-assistant feature rather than a fully autonomous driving system.
Given an NHTSA level 2 can still drive on itself by hundreds of miles in the highway, it is easy for driver to misinterpret the real capabilities of current systems and get comfortable on letting the machine take its decisions alone, wandering around visually or mentally, believing they are using a “limited self-diving” NHTSA level 3 of automation, or even a fully autonomous level 4. Current systems have evolved to process a great amount of information coming from the environment, but they may be leaving out the most important character involved in the vehicle: the driver.
This project focuses on that forgotten element in the vehicle framework and intends to stablish a robust yet flexible representation for such a concept system as a driving environment, considering the driver itself, the internal mechanics of the vehicle, and external elements such as the driveway, other vehicles or pedestrians and traffic signals both passive and possible active signals with intelligent capabilities of Intelligent Transportations Systems (ITS).
Vehicle technologies have advanced significantly over the past 20 years, especially with respect to novel in-vehicle systems for route navigation, information access, infotainment, and connected vehicle advancements for vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) connectivity and communications. While there is great interest in migrating to fully automated, self-driving vehicles, factors such as technology performance, cost barriers, public safety, insurance issues, legal implications, and government regulations suggest it is more likely that the first step in the progression will be multifunctional vehicles. Today, embedded controllers as well as a variety of sensors and high-performance computing in present-day cars allow for a smooth transition from complete human control toward semisupervised or assisted control, then to fully automated vehicles. Next-generation vehicles will need to be more active in assessing driver awareness, vehicle capabilities, and traffic and environmental settings, plus how these factors come together to determine a collaborative safe and effective driver-vehicle engagement for vehicle operation. This article reviews a range of issues pertaining to driver modeling for the detection and assessment of distraction. Examples from the UTDrive project are used whenever possible, along with a comparison to existing research programs. The areas addressed include 1) understanding driver behavior and distraction, 2) maneuver recognition and distraction analysis, 3) glance behavior and visual tracking, and 4) mobile platform advancements for in-vehicle data collection and human-machine interface. This article highlights challenges in achieving effective modeling, detection, and assessment of driver distraction using both UTDrive instrumented vehicle data and naturalistic driving data
Data mining techniques support numerous applications of intelligent transportation systems (ITSs). This paper critically reviews various data mining techniques for achieving trip planning in ITSs. The literature review starts with the discussion on the contributions of descriptive and predictive mining techniques in ITSs, and later continues on the contributions of the clustering techniques. Being the largely used approach, the use of cluster analysis in ITSs is assessed. However, big data analysis is risky with clustering methods. Thus, evolutionary computational algorithms are used for data mining. Though unsupervised clustering models are widely used, drawbacks such as selection of optimal number of clustering points, defining termination criterion, and lack of objective function also occur. Eventually, various drawbacks of evolutionary computational algorithm are also addressed in this paper.
Driver distraction has been identified as one major cause of unsafe driving. Distraction is highly demanding on the real-time detection compared with fatigue detection. It is unclear about the common features of different distraction types under different driving contexts. To solve those problems above, the research on driver distraction indicated by driving performance and eye movement was conducted; from feature extraction to real-time detection. The experiments were based on driving simulation, two typical driving scenarios were reconstructed; the stop-controlled intersection (urban scenario) and the speed limited highway (highway scenario). There are two types of distraction in the study; visual distraction and cognitive distraction. Distracted driving is defined as driving with specifically selected secondary tasks; therefore the distraction detection is a 0-1 question. The statistical analysis of candidate features from driving performance and eye movement was applied first. The optimal feature subsets were extracted by applying Support Vector Machine – Recursive Feature Elimination (SVM-RFE). The significant features were further extracted for distraction detection among different combinations of distraction type and driving scenario. By testing the performance of SVM classification based on the optimal feature subsets, the advantages of multi-source information fusion were quantified and the Yerkes-Dodson Law was therefore partially validated. Finally based on the SVM classification model, the extracted features as input, the real-time distraction detection algorithm was designed and cross validatd. The optimal combination of the algorithm parameters was determined. The performance of this real-time distraction detection algorithm is good on both correct rate and rapidity.
Stop-controlled intersection and speed limited highway were reconstructed as the driving scenarios. Two types of cognitive secondary tasks and two types of visual secondary tasks were applied to generate the specific kind of distraction. The Peripheral Detection Task device (PDT) used for quantifying the cognitive workload during driving was modified into a new version called Detection Response Task device (DRT) with improved portability. Measured by DRT, the cognitive workload is regarded as the important reference for distraction detection. Driving and eye movement data were logged under two conditions; normal driving and distracted driving. Twenty-two subjects participated in the experiment of urban scenario and 16 subjects for the experiment of highway scenario. The candidate features of driving performance and eye movement are calculated and statistically analyzed. The results indicate the relationship between these features impacted by the distracted driving and driver characteristics like age and gender.
To cope with high-dimension and small sample size of collected data, SVM-RFE is adopted to extract the optimal feature subset for best classification performance. Based on the analysis of the structure and importance of extracted optimal feature subsets, the common and significant features across different distraction types and driving scenarios are determined. By testing the performance of SVM classification based on the optimal feature subsets, the advantages of multi-source information fusion are quantified. Seen from the measured workload under two driving scenarios, it is found that under high-workload driving scenario, the fusion of driving performance and eye movement features yield significantly improved correct rates of distraction recognition. Through this finding, the Yerkes-Dodson Law is validated and the importance of driving context is proved from the perspective of driving workload. These results provide method and data support for the feature extraction of distraction detection.
A SVM-based real-time distraction detection algorithm with both driving performance and eye movement features as inputs is designed and cross validated. The algorithm performs well in both correct rate and detection rapidity which is also adaptive to different driving scenarios and distractions. The length of calculation time window and the overlap rate of it are two parameters of the algorithm. The optimal combination of these parameters, 5-second-length calculation window with 75% overlap rate, is determined by the best comprehensive performance of distraction detection. The algorithm gets correct rates on average between 86.6% and 98.9% while the degree of decision in advance (the indicator of the detection rapidity) reaches 88.7% to 95.0%. More specific, with 30s as the length of extracted event sample, the distracted status of the driver can be recognized within 6.5s to 9.0s which indicates the good performance of detection rapidity as well as accuracy.
In two experiments we explored the influence of individual differences in working memory capacity (WMC) on hazard perception performance in a simulated driving task. In Experiment 1, we examined the relationship between WMC and hazard perception performance under control and dual task conditions, and self-reported driving behavior. Results revealed significant relationships between WMC, hazard perception performance and self-reported driving behavior. Participants lower in WMC performed poorer in dual task conditions and reported more instances of inattention when driving. In Experiment 2 we explored the gaze behavior of low and high WMC individuals whilst completing the hazard perception test under control and dual task conditions. Results revealed that low-WMC individuals had poorer hazard perception performance under dual task conditions and these performance decrements were mirrored in reductions in mean fixation durations on the hazard. Interestingly, pupillary dilation appears to discriminate between low- and high-WMC individuals and might be a useful index of attention for future research.
Although integration of environment and driver information can be achieved at both micro- and macro- levels with different benefits towards driving safety, most studies focus only on the micro-level integration by coupling individual external environment events and driver responses. In the macro level, however, it is more important to understand overall effects of environment features on driver behavior, and their combined effects on driving safety. Based on some previous findings on the significant effects of driver glance behavior on crash risk and the prominent moderating effects of traffic density levels on this relationship, this paper tries to use surrounding vehicle patterns to classify traffic density levels and study the direct effects of traffic density levels on driver glance behavior. The datasets used for analysis are based on VTTI 100-car study. After proposing the measures of surrounding vehicle patterns, the classification of traffic density is completed using support vector machine (SVM). The results show that, although it is difficult to classify four traffic density levels based on the proposed surrounding vehicle patterns, the predication accuracy for two or three traffic density levels is good. Also, significant effects of traffic density on driver glance behaviors to the vehicle mirrors are identified.
Purpose
Argos is a long-term program started in 1990 to promote experimental research about driver behavior under realistic driving conditions. The development of the latest Argos platform has recently been completed and some experiments have been carried out based on this new complex and powerful tool.
System Description
Argos allows recording multiple car parameters (such as speed), driver variables (such as the point of gaze), and environmental parameters (some of which are obtained by means of real-time signal processing, such as the distance to lateral road marks). All of these data can be used to replay driving sessions in the laboratory and to extract data associated with different driving session segments.
Paper contents
This paper is focused on the description of the ”Roadmap Program Interpreter”, a tool embedded in the Argos acquisition and control system which not only interacts with the equipment, but also with the driver and with the experiment supervisor. This tool is based on a new language, ESL –Experiment Specification Language– designed for the Argos project to help the experiment designer to formally specify complex, controlled, and repeatable experiments in a very efficient way. Argos-ESL has allowed the Spanish Traffic Agency (DGT, a governmental agency) as well as other local agencies to develop complex experiments in a fraction of the time previously needed by the DGT for designing similar driving experiments.
In transportation and traffic planning studies, accurate traffic data are required for reliable model calibration to accurately predict transportation system performance and ensure better traffic planning. However, it is impractical to gather data from an entire population for such estimations because the widely used loop detectors and other more advanced wireless sensors may be limited by various factors. Thus, making data inferences based on smaller populations is generally inevitable. Linear data projection is a commonly and intuitively adopted method for inferring population traffic characteristics. It projects a sample of observable traffic quantities such as traffic count based on a set of scaling factors. However, scaling factors are subject to different types of variability such as spatial variability. Models calibrated based on linearly projected data that do not account for variability may introduce a systematic bias into their parameters. Such a bias is surprisingly often ignored. This paper reveals the existence of a systematic bias in model calibration caused by variability in the linear data projection. A generalized multivariate polynomial model is applied to examine the effect of this variability on model parameters. Adjustment factors are derived and methods are proposed for detecting and removing the embedded systematic bias. A simulation is used to demonstrate the effectiveness of the proposed method. To illustrate the applicability of the method, case studies are conducted using real-world global positioning system data obtained from taxis. These data calibrate the Macroscopic Bureau of Public Road function for six 1 × 1 km regions in Hong Kong.
In this paper, we presented a novel quantification tool for the mental engagement of patients during rehabilitation. This method is based on eye-movement analysis, which can perform automatic, objective, and real-time quantification with a simple setup. Using this quantification tool, the therapists can monitor the patients' engagement in real time and adjust the rehabilitation exercises to maintain the patients' engagement at a certain level for the best rehabilitation outcome. A preliminary study was performed, in which the presented method successfully distinguished the different engagement levels in different rehabilitation games.
The consequences of performing verbal and spatial-imagery tasks on visual search when driving were studied. Twelve participants drove 84 km on 2 highways and 2 roads. On each route, they performed 2 verbal tasks and 2 spatial-imagery tasks while their eye movements were recorded. The same results were repeated on all routes. Pupillary dilation indicated similar effort for each task. Visual functional-field size decreased horizontally and vertically, particularly for spatial-imagery tasks. Compared with ordinary driving, fixations were longer during the spatial-imagery task. With regard to driving performance, glance frequency at mirrors and speedometer decreased during the spatial-imagery task. Results are interpreted in terms of multiple attention-resource theories; implications of internal distractions on road safety are discussed in terms of possible impairment in relevant information processing.
Mental workload is known to reduce the area of one's visual field, but little is known about its effects on the shape of the visual field. Considering this, the visual fields of 13 subjects were measured concurrently under three levels of mental workload using a Goldmann visual perimeter. Tone counting tasks were employed to induce mental workload, avoiding interference with visual performance. Various methods of shape measurement and analysis were used to investigate the variation of the shape of the visual field as a function of mental load. As expected, the mean area of visual fields reduced to 92.2% in the medium workload condition and to 86.41% under heavy workload, compared to light load condition. This tunnelling effect was not uniform, but resulted in statistically significant shape distortion as well, as measured by the majority of the 12 shape indices used here. These results have visual performance implications in many tasks that are susceptible to changes in visual fields and peripheral vision. Knowledge of the dynamics of the visual field as a function of mental workload can offer significant advantages also in mathematical modelling of visual search.
Signal detection theory (SDT) may be applied to any area of psychology in which two different types of stimuli must be discriminated. We describe several of these areas and the advantages that can be realized through the application of SDT. Three of the most popular tasks used to study discriminability are then discussed, together with the measures that SDT prescribes for quantifying performance in these tasks. Mathematical formulae for the measures are presented, as are methods for calculating the measures with lookup tables, computer software specifically developed for SDT applications, and general purpose computer software (including spreadsheets and statistical analysis software).
This research examined the effects of hands-free cell phone conversations on simulated driving. The authors found that these conversations impaired driver's reactions to vehicles braking in front of them. The authors assessed whether this impairment could be attributed to a withdrawal of attention from the visual scene, yielding a form of inattention blindness. Cell phone conversations impaired explicit recognition memory for roadside billboards. Eye-tracking data indicated that this was due to reduced attention to foveal information. This interpretation was bolstered by data showing that cell phone conversations impaired implicit perceptual memory for items presented at fixation. The data suggest that the impairment of driving performance produced by cell phone conversations is mediated, at least in part, by reduced attention to visual inputs.
The effects of mental workload on visual search and decision making were studied in real traffic conditions with 12 participants who drove an instrumented car. Mental workload was manipulated by having participants perform several mental tasks while driving. A simultaneous visual-detection and discrimination test was used as performance criteria. Mental tasks produced spatial gaze concentration and visual-detection impairment, although no tunnel vision occurred. According to ocular behavior analysis, this impairment was due to late detection and poor identification more than to response selection. Verbal acquisition tasks were innocuous compared with production tasks, and complex conversations, whether by phone or with a passenger, are dangerous for road safety.
The move towards testing computers, phones, and cars that sense when one is busy and spare him/her from distraction is discussed. It is observed that when one is unexpectedly interrupted, he/she not only work less efficiently but also make more mistakes. It is suggested that if computers and phones could be given with understandinng of limits of human attention and memory, it would make them seem a lot more thoughtful and courteous. The way out as perceived, lies in considerate computing.
Driver workload estimation (DWE) refers to the activities of monitoring a driver and the driving environment in real-time and acquiring the knowledge of the driver's workload continuously. With this knowledge of the driver's workload, the in-vehicle information systems (IVIS) can provide information on when the driver has the spare capacity to receive and comprehend it, which is both effective and efficient. However, after years of study, it is still difficult to build a robust DWE system. In this paper, we analyze the difficulties facing the existing methodology of developing DWE systems and propose a machine-learning-based DWE development process. Some preliminary but promising results are reported using a popular machine-learning method, the decision tree.
This paper presents a nonintrusive prototype computer vision system for monitoring a driver's vigilance in real time. It is based on a hardware system for the real-time acquisition of a driver's images using an active IR illuminator and the software implementation for monitoring some visual behaviors that characterize a driver's level of vigilance. Six parameters are calculated: Percent eye closure (PERCLOS), eye closure duration, blink frequency, nodding frequency, face position, and fixed gaze. These parameters are combined using a fuzzy classifier to infer the level of inattentiveness of the driver. The use of multiple visual parameters and the fusion of these parameters yield a more robust and accurate inattention characterization than by using a single parameter. The system has been tested with different sequences recorded in night and day driving conditions in a motorway and with different users. Some experimental results and conclusions about the performance of the system are presented
The Transportation Systems Center, in support of research carried out by the National Highway Traffic Safety Administration's Crash Avoidance Division, has reviewed research into driver attentional processes to assess the potential for the development of methods and techniques for reducing the number of accidents related to attentional lapses. Contained in this paper is a summary of the results of the review with regard to the (a) safety implications of inattention, (b) psychological and physiological indices of inattention, and (c) in-vehicle instrumentation for detecting inattention. Areas of research are suggested that could be valuable in the development of practical attention monitors for in-vehicle use.
Measurement of eye movements is a powerful tool for investigating perceptual and cognitive function in both infants and adults. Straightforwardly, eye movements provide a multifaceted measure of performance. For example, the location of fixations, their duration, time of occurrence, and accuracy all are potentially revealing and often allow stronger inferences than measures such as percentage correct or reaction time. Another advantage is that eye movements are an implicit measure of performance and do not necessarily involve conscious processes. Indeed, they are often a more revealing measure than conscious report (Hayhoe, Bensinger, & Ballard, 1998). Although the mere presence of gaze at a particular location in the visual field does not reveal the variety of brain computations that might be operating at that moment, the experimental context within which the fixation occurs often provides critical information that allows powerful inferences. The articles in this thematic collection are excellent examples of this.
We propose a method of modifying a kernel function to improve the performance of a support vector machine classifier. This is based on the structure of the Riemannian geometry induced by the kernel function. The idea is to enlarge the spatial resolution around the separating boundary surface, by a conformal mapping, such that the separability between classes is increased. Examples are given specifically for modifying Gaussian Radial Basis Function kernels. Simulation results for both artificial and real data show remarkable improvement of generalization errors, supporting our idea.
In this paper, we present a comprehensive survey on applications of Support Vector Machines (SVMs) for pattern recognition. Since SVMs show good generalization performance on many real-life data and the approach is properly motivated theoretically, it has been applied to wide range of applications. This paper describes a brief introduction of SVMs and summarizes its numerous applications.
From the publisher: This is the first comprehensive introduction to Support Vector Machines (SVMs), a new generation learning system based on recent advances in statistical learning theory. SVMs deliver state-of-the-art performance in real-world applications such as text categorisation, hand-written character recognition, image classification, biosequences analysis, etc., and are now established as one of the standard tools for machine learning and data mining. Students will find the book both stimulating and accessible, while practitioners will be guided smoothly through the material required for a good grasp of the theory and its applications. The concepts are introduced gradually in accessible and self-contained stages, while the presentation is rigorous and thorough. Pointers to relevant literature and web sites containing software ensure that it forms an ideal starting point for further study. Equally, the book and its associated web site will guide practitioners to updated literature, new applications, and on-line software.
Four investigations were carried out to assess the feasibility of using eye movement measures as indices of mental workload. In the first experiment, saccadic extent was measured during free viewing while subjects performed low, moderate and high complexity, auditory tone counting as the workload tasks. The range of saccadic extent decreased significantly as tone counting complexity (workload) was increased. In the second experiment the range of spontaneous saccades was measured under three levels of counting complexity with a visual task that did not require fixation or tracking. The results indicated that the extent of saccadic eye movements was significantly restricted as counting complexity increased. In the third experiment, the effects of practice were examined and decreased saccadic range under high tone counting complexity was observed even when significant increases in performance occurred with practice. Finally, in experiment 4, the first experiment was repeated with additional optokinetic stimulation and the saccadic range was again observed to decrease with tone counting complexity. The results indicated that the extent of spontaneous and elicited eye movements was significantly restricted as counting complexity increased. We conclude that this measure may provide a valuable index of mental workload.
In this study, 150 subjects observed a 25-minute video driving sequence containing 45 highway traffic situations to which they were expected to respond by manipulation of simulated vehicle controls. Each situation occurred under five conditions of distraction: placing a cellular phone call, carrying on a causal cellular phone conversation, carrying on an intense cellular phone conversation, tuning a radio, and no distraction. All of the distractions led to significant increases in the proportion of situations to which subjects failed to respond. However, significant age differences of nonresponse appeared. Among subjects over age 50, nonresponses increased by about one-third under all of the telephone distractions. The response rate of younger subjects increased by a lesser degree except under intense conversation. Results were not influenced by gender or prior experience with cellular phones. The authors conclude that older drivers might reduce their accident risk during attention-demanding traffic conditions by avoiding use of cellular phones and that other drivers might do so by refraining from calls involving intense conversation.
Studies have examined possible effects of concurrent mobile phone use on driving performance. Although interference is often apparent, determining the implications of such findings for 'real world' driving is problematic. This paper considers some relevant methodological issues including the definition of procedures and terms, operationalization of task elements, sampling of task components, and the provision of experimental controls. Suggestions are made about how methodological rigor could be improved.
Dual-task studies assessed the effects of cellular-phone conversations on performance of a simulated driving task. Performance was not disrupted by listening to radio broadcasts or listening to a book on tape. Nor was it disrupted by a continuous shadowing task using a handheld phone, ruling out, in this case, dual-task interpretations associated with holding the phone, listening, or speaking, However significant interference was observed in a word-generation variant of the shadowing task, and this deficit increased with the difficulty of driving. Moreover unconstrained conversations using either a handheld or a hands-free cell phone resulted in a twofold increase in the failure to detect simulated traffic signals and slower reactions to those signals that were detected. We suggest that cellular-phone use disrupts performance by diverting attention to an engaging cognitive context other than the one immediately associated with driving.
As computer applications for cars emerge, a speech-based interface offers an appealing alternative to the visually demanding direct manipulation interface. However, speech-based systems may pose cognitive demands that could undermine driving safety. This study used a car-following task to evaluate how a speech-based e-mail system affects drivers' response to the periodic braking of a lead vehicle. The study included 24 drivers between the ages of 18 and 24 years. A baseline condition with no e-mail system was compared with a simple and a complex e-mail system in both simple and complex driving environments. The results show a 30% (310 ms) increase in reaction time when the speech-based system is used. Subjective workload ratings and probe questions also indicate that speech-based interaction introduces a significant cognitive load, which was highest for the complex e-mail system. These data show that a speech-based interface is not a panacea that eliminates the potential distraction of in-vehicle computers. Actual or potential applications of this research include design of in-vehicle information systems and evaluation of their contributions to driver distraction.
Driver-assistance systems that monitor driver intent, warn drivers of lane departures, or assist in vehicle guidance are all being actively considered. It is therefore important to take a critical look at key aspects of these systems, one of which is lane-position tracking. It is for these driver-assistance objectives that motivate the development of the novel "video-based lane estimation and tracking" (VioLET) system. The system is designed using steerable filters for robust and accurate lane-marking detection. Steerable filters provide an efficient method for detecting circular-reflector markings, solid-line markings, and segmented-line markings under varying lighting and road conditions. They help in providing robustness to complex shadowing, lighting changes from overpasses and tunnels, and road-surface variations. They are efficient for lane-marking extraction because by computing only three separable convolutions, we can extract a wide variety of lane markings. Curvature detection is made more robust by incorporating both visual cues (lane markings and lane texture) and vehicle-state information. The experiment design and evaluation of the VioLET system is shown using multiple quantitative metrics over a wide variety of test conditions on a large test path using a unique instrumented vehicle. A justification for the choice of metrics based on a previous study with human-factors applications as well as extensive ground-truth testing from different times of day, road conditions, weather, and driving scenarios is also presented. In order to design the VioLET system, an up-to-date and comprehensive analysis of the current state of the art in lane-detection research was first performed. In doing so, a comparison of a wide variety of methods, pointing out the similarities and differences between methods as well as when and where various methods are most useful, is presented
This paper presents methods for collecting and analyzing physiological data during real-world driving tasks to determine a driver's relative stress level. Electrocardiogram, electromyogram, skin conductance, and respiration were recorded continuously while drivers followed a set route through open roads in the greater Boston area. Data from 24 drives of at least 50-min duration were collected for analysis. The data were analyzed in two ways. Analysis I used features from 5-min intervals of data during the rest, highway, and city driving conditions to distinguish three levels of driver stress with an accuracy of over 97% across multiple drivers and driving days. Analysis II compared continuous features, calculated at 1-s intervals throughout the entire drive, with a metric of observable stressors created by independent coders from videotapes. The results show that for most drivers studied, skin conductivity and heart rate metrics are most closely correlated with driver stress level. These findings indicate that physiological signals can provide a metric of driver stress in future cars capable of physiological monitoring. Such a metric could be used to help manage noncritical in-vehicle information systems and could also provide a continuous measure of how different road and traffic conditions affect drivers.
This paper presents a system for analyzing human driver visual attention. The system relies on estimation of global motion and color statistics to robustly track a person's head and facial features. The system is fully automatic, it can initialize automatically, and reinitialize when necessary. The system classifies rotation in all viewing directions, detects eye/mouth occlusion, detects eye blinking and eye closure, and recovers the three dimensional gaze of the eyes. In addition, the system is able to track both through occlusion due to eye blinking, and eye closure, large mouth movement, and also through occlusion due to rotation. Even when the face is fully occluded due to rotation, the system does not break down. Further the system is able to track through yawning, which is a large local mouth motion. Finally, results are presented, and future work on how this system can be used for more advanced driver visual attention monitoring is discussed.
INTRODUCTION The problem of human-computer interaction can be viewed as two powerful information processors (human and computer) attempting to communicate with each other via a narrowbandwidth, highly constrained interface [23]. To address it, we seek faster, more natural, and more convenient means for users and computers to exchange information. The user's side is constrained by the nature of human communication organs and abilities; the computer's is constrained only by input/output devices and interaction techniques that we can invent. Current technology has been stronger in the computer-to-user direction than user-to-computer, hence today's user-computer dialogues are rather one-sided, with the bandwidth from the computer to the user far greater than that from user to computer. Using eye movements as a user-tocomputer communication medium can help redress this imbalance. This chapter describes the relevant characteristics of the human eye, eye tracking technology, how to design in
Taxonomy of mitigation strategies for driver distraction
- B Donmez
- L Boyle
- J D Lee
B. Donmez, L. Boyle, and J. D. Lee, " Taxonomy of mitigation strategies
for driver distraction, " in Proc. Hum. Factors Ergonom. Soc. 47th Annu.
Meeting, Denver, CO, 2003, pp. 1865–1869.
Lane change intent analysis using robust operators and sparse Bayesian learning
- J Mccall
- D Mipf
- M M Trivedi
- B Rao
J. McCall, D. Mipf, M. M. Trivedi, and B. Rao, " Lane change intent analysis using robust operators and sparse Bayesian learning, " IEEE Trans.
Intell. Transp. Syst., to be published.
Using support vector machines for lane-change detection
- H M Mandalia
- D D Salvucci
H. M. Mandalia and D. D. Salvucci, " Using support vector machines for
lane-change detection, " in Proc. Hum. Factors and Ergonomics Soc. 49th
Annu. Meeting, Orlando, FL, 2005, pp. 1965–1969.