No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Pedestrians at the kerb – Recognising the action intentions of humans
Abstract
Aim of the presented research is the development of a cognitive driver assistance system, which can capture the traffic situation, analyse it, and warn the driver in case a pedestrian is a potential hazard. Hence parameters have to be identified by which the intention of the pedestrian can be unambiguously predicted. Two approaches to the topic are addressed. First, the pedestrian’s perspective was taken. The question was how crossing decisions were influenced by the parameters distance and velocity of the car. Following a signal, participants had to choose to cross the road in front of or behind the car. The data analysis showed that pedestrians relied on the distance of the car rather than the time to collision for their decision. In the second experiment the observer’s perspective raised the question what parameters humans use to predict pedestrians’ intentions. Videos of natural traffic scenes were presented. Participants had to make statements about whether the shown pedestrian would cross the street during the next moment. In a baseline and four experimental conditions, certain information was masked in the videos. Just the condition in which only the trajectory information of the pedestrian was available produced a higher error rate.
... However, the approaching vehicle with a slower speed gives pedestrians a bigger collision threat under the same TTC. The results tie well with the previous experimental observations on pedestrian crossing behavior [21], [34], [35]. ...
... The SW-PRD model predictions of crossing gap acceptance for each speed and time gap condition are compared with the observed data in Fig. 6a. According to the empirical data, crossing gap acceptance increased with vehicle speed and traffic gap, aligning well with previous studies [21], [35]. The SW-PRD model reproduces these behavioral patterns very well (R 2 = 0.890, RM SE = 0.050), suggesting that pedestrians might adapt their crossing decisions based on the changes in collision cues. ...
... In our previous study [22], we showed that the visual collision cue,θ, could capture the effects of vehicle kinematics on pedestrian crossing decisions in single gaps and explain why pedestrians tended to rely on distance from vehicles to make crossing decisions [21], [35]. In this study, this finding is formally applied to model crossing decisions and extended to a more complicated traffic scenario, i.e., a continuous flow of traffic. ...
As safe and comfortable interactions with pedestrians could contribute to automated vehicles' (AVs) social acceptance and scale, increasing attention has been drawn to computational pedestrian behavior models. However, very limited studies characterize pedestrian crossing behavior based on specific behavioral mechanisms, as those mechanisms underpinning pedestrian road behavior are not yet clear. Here, we reinterpret pedestrian crossing behavior based on a deconstructed crossing decision process at uncontrolled intersections with continuous traffic. Notably, we explain and model pedestrian crossing behavior as they wait for crossing opportunities, optimizing crossing decisions by comparing the visual collision risk of approaching vehicles around them. A collision risk-based crossing initiation model is proposed to characterize the time-dynamic nature of pedestrian crossing decisions. A simulation tool is established to reproduce pedestrian behavior by employing the proposed model and a social force model. Two datasets collected in a CAVE-based immersive pedestrian simulator are applied to calibrate and validate the model. The model predicts pedestrian crossing decisions across all traffic scenarios well. In particular, by considering the decision strategy that pedestrians compare the collision risk of surrounding traffic gaps, model performance is significantly improved. Moreover, the collision risk-based crossing initiation model accurately captures the timing of pedestrian crossing initiations within each gap. This work concisely demonstrates how pedestrians dynamically adapt their crossings in continuous traffic based on perceived collision risk, potentially providing insights into modeling coupled human-AV interactions or serving as a tool to realize human-like pedestrian road behavior in virtual AVs test platforms.
... • Trajectory: pedestrians' ability to estimate speed is also affected by their walking direction. When pedestrians walk in the same direction as vehicles, they are more likely to make risky decisions about whether or not to cross [Schmidt and Faerber, 2009]. ...
... Variations in social norms exist, obviously between different countries but also within the same country [Björklund and Åberg, 2005]. For instance, each culture could assign different levels of importance to traffic issues (e.g speeding and jaywalking between swedish and chinese drivers [Lindgren et al., 2008]), could have a different gap acceptance times 2 (e.g indians cross on average between 2s to 8s whereas germans cross between 3s to 7s time to collision [Schmidt and Faerber, 2009]) or could perceive and analyze a situation differently (e.g americans judge traffic behavior based on pedestrian features, but indians place more emphasis on contextual elements such as traffic circumstances, road structure... [Clay, 1995]) ...
... • Traffic density affects both pedestrians and drivers [Schmidt and Faerber, 2009]. To put it in a nutshell, the higher the density of traffic, the lower the chance of pedestrians to cross against the signal [Ishaque and Noland, 2008]. ...
The autonomous vehicle (AV) is a major challenge for the mobility of tomorrow. Progress is being made every day to achieve it; however, many problems remain to be solved to achieve a safe outcome for the most vulnerable road users (VRUs). One of the major challenge faced by AVs is the ability to efficiently drive in urban environments. Such a task requires interactions between autonomous vehicles and VRUs to resolve traffic ambiguities. In order to interact with VRUs, AVs must be able to understand their intentions and predict their incoming actions. In this dissertation, our work revolves around machine learning technology as a way to understand and predict human behaviour from visual signals and more specifically pose kinematics. Our goal is to propose an assistance system to the AV that is lightweight, scene-agnostic that could be easily implemented in any embedded devices with real-time constraints. Firstly, in the gesture and action recognition domain, we study and introduce different representations for pose kinematics, based on deep learning models as a way to efficiently leverage their spatial and temporal components while staying in an euclidean grid-space. Secondly, in the autonomous driving domain, we show that it is possible to link the posture, the walking attitude and the future behaviours of the protagonists of a scene without using the contextual information of the scene (zebra crossing, traffic light...). This allowed us to divide by a factor of 20 the inference speed of existing approaches for pedestrian intention prediction while keeping the same prediction robustness. Finally, we assess the generalization capabilities of pedestrian crossing predictors and show that the classical train-test sets evaluation for pedestrian crossing prediction, i.e., models being trained and tested on the same dataset, is not sufficient to efficiently compare nor conclude anything about their applicability in a real-world scenario. To make the research field more sustainable and representative of the real advances to come. We propose new protocols and metrics based on uncertainty estimates under domain-shift in order to reach the end-goal of pedestrian crossing behavior predictors: vehicle implementation.
... Not only does it affect pedestrians, but also it affects drivers. Schmidt and Färber 2009 suggested that drivers driving at high speed will tend to receive more dangerous crossings from pedestrians, potentially resulting in more accidents. However, few studies have studied the potential decision-making mechanism of this unsafe crossing behaviour specifically. ...
... These can be roughly categorised as external and internal attributes. External attributes which may affect pedestrian gap acceptance behaviour include vehicle speed (Schmidt and Färber, 2009), time to arrival (TTA) (Avinash et al., 2019;Pawar and Patil, 2016), distance (Lobjois and Cavallo, 2007;Schmidt and Färber, 2009), number of lanes (Chandra et al., 2014), and vehicle size (Beggiato et al., 2017;Lee and Sheppard, 2017). Internal attributes which may have an impact include gender, age (Hulse et al., 2018;Kalatian and Farooq, 2021) and group size (Pawar and Patil, 2015;Avinash et al., 2019). ...
... These can be roughly categorised as external and internal attributes. External attributes which may affect pedestrian gap acceptance behaviour include vehicle speed (Schmidt and Färber, 2009), time to arrival (TTA) (Avinash et al., 2019;Pawar and Patil, 2016), distance (Lobjois and Cavallo, 2007;Schmidt and Färber, 2009), number of lanes (Chandra et al., 2014), and vehicle size (Beggiato et al., 2017;Lee and Sheppard, 2017). Internal attributes which may have an impact include gender, age (Hulse et al., 2018;Kalatian and Farooq, 2021) and group size (Pawar and Patil, 2015;Avinash et al., 2019). ...
Accidents involving pedestrians are particularly common at unsignalised intersections and mid-block crosswalks,
where vehicles often do not yield to them. Analysing and understanding pedestrian crossing behaviour at such
locations is vital for improving road safety. Previous studies have repeatedly shown that pedestrians tend to
accept smaller time gaps in conditions with higher vehicle speeds and thus potentially less safe. This has prompted the hypothesis that pedestrians rely on spatial distance to make crossing decisions. However, few studies have investigated the mechanism underpinning this phenomenon. We propose a novel approach to characterise pedestrian crossing behaviour: a psychophysics-based gap acceptance (PGA) model based on visual looming cues and binary choice logit method. Road crossing data collected in a simulated experiment were used to analyse pedestrian behaviour and test the model. Our analysis indicates that, in line with previous studies, higher vehicle speed increased the tendency of gap acceptance, leading to a higher rate of unsafe crossings. Crucially, the PGA model could accurately account for these crossing decisions across experimental scenarios, more parsimoniously than a conventional model. These results explain the speed-induced unsafe behaviour by suggesting that pedestrians apply visual looming, which depends on vehicle speed and distance, to make crossing decisions. This study reinforces the notion that for two vehicles with the same time gap, the one with higher speed can elicit more risky crossing behaviour from pedestrians, potentially resulting in more severe accidents. The practical implications of the results for traffic safety management, modelling and development of automated
vehicles are discussed.
... Not only does it affect pedestrians, but also it affects drivers. Schmidt and Färber 2009 suggested that drivers driving at high speed will tend to receive more dangerous crossings from pedestrians, potentially resulting in more accidents. However, few studies have studied the potential decision-making mechanism of this unsafe crossing behaviour specifically. ...
... These can be roughly categorised as external and internal attributes. External attributes which may affect pedestrian gap acceptance behaviour include vehicle speed (Schmidt and Färber, 2009), time to arrival (TTA) (Avinash et al., 2019;Pawar and Patil, 2016), distance (Lobjois and Cavallo, 2007;Schmidt and Färber, 2009), number of lanes (Chandra et al., 2014), and vehicle size (Beggiato et al., 2017;Lee and Sheppard, 2017). Internal attributes which may have an impact include gender, age (Hulse et al., 2018;Kalatian and Farooq, 2021) and group size (Pawar and Patil, 2015;Avinash et al., 2019). ...
... These can be roughly categorised as external and internal attributes. External attributes which may affect pedestrian gap acceptance behaviour include vehicle speed (Schmidt and Färber, 2009), time to arrival (TTA) (Avinash et al., 2019;Pawar and Patil, 2016), distance (Lobjois and Cavallo, 2007;Schmidt and Färber, 2009), number of lanes (Chandra et al., 2014), and vehicle size (Beggiato et al., 2017;Lee and Sheppard, 2017). Internal attributes which may have an impact include gender, age (Hulse et al., 2018;Kalatian and Farooq, 2021) and group size (Pawar and Patil, 2015;Avinash et al., 2019). ...
Accidents involving pedestrians are particularly common at unsignalised intersections and mid-block crosswalks, where vehicles often do not yield to them. Analysing and understanding pedestrian crossing behaviour at such locations is vital for improving road safety. Previous studies have repeatedly shown that pedestrians tend to accept smaller time gaps in conditions with higher vehicle speeds and thus potentially less safe. This has prompted the hypothesis that pedestrians rely on spatial distance to make crossing decisions. However, few studies have investigated the mechanism underpinning this phenomenon. We propose a novel approach to characterise pedestrian crossing behaviour: a psychophysics-based gap acceptance (PGA) model based on visual looming cues and binary choice logit method. Road crossing data collected in a simulated experiment were used to analyse pedestrian behaviour and test the model. Our analysis indicates that, in line with previous studies, higher vehicle speed increased the tendency of gap acceptance, leading to a higher rate of unsafe crossings. Crucially, the PGA model could accurately account for these crossing decisions across experimental scenarios, more parsimoniously than a conventional model. These results explain the speed-induced unsafe behaviour by suggesting that pedestrians apply visual looming, which depends on vehicle speed and distance, to make crossing decisions. This study reinforces the notion that for two vehicles with the same time gap, the one with higher speed can elicit more risky crossing behaviour from pedestrians, potentially resulting in more severe accidents. The practical implications of the results for traffic safety management, modelling and development of automated vehicles are discussed.
... To the best of our knowledge, there are no comparative studies to date between Japan and Germany that systematically examine pedestrians' subjective risk perception. In different studies, it was reported that Japanese pedestrians [44] require a longer time-gap (16s) to consider a crossing safe compared to German pedestrians (5-6s) [10], [56] These findings are consistent with the higher risk aversion in Japan and correspond to the gap acceptance (see section IV-C). The frequency and duration of uncertainty behavior (freezing, abandoning, accelerating) during crossing at red is also more pronounced in Japan compared to France (10%, 1.5s vs. 5%, 1.22s) [57]. ...
... To decide whether a gap in traffic to the next car is large enough to cross the street safely, distance and velocity cues can be used. Slower vehicles can increase the time gap due to their larger distance [56], [59]. In addition, there are indications that with reduced cognitive resources (e.g. ...
... In Japan, female pedestrians required larger gaps (19s) compared to male pedestrians (11s) [44]. In Germany, different studies show that pedestrians require a time gap of 5-6s to decide for a safe crossing, both in a real environment [10], [56], as well as in a simulated environment [59] with a speed limit of 50 km/h. ...
The prediction of pedestrian behavior remains a
major objective for the development of autonomous vehicles.
Pedestrians do not merely represent the most vulnerable traffic
participants, but are also a challenge in the prediction process,
since their behavior entails a large number of options for
possible paths, velocities, and motions. In addition, autonomous
vehicles should be able to operate safely in different countries, and thus the incorporation of cultural differences in
the training and evaluation of the relevant AI systems is
required. This paper provides the first review of Japanese and
German pedestrians’ behavior in urban traffic. In particular,
cultural behavior differences of pedestrians in risk avoidance,
compliance, gap acceptance, and walking velocity together with
different environmental factors like pedestrian facilities in both
countries are addressed.
... In general, the time-to-collision (TTC) value is a key indicator [9]. A typical limit is a TTC value of less than 3 s, which makes it unlikely for pedestrians to attempt a crossing [10]. While this property is often used to model pedestrian behaviour, it is necessary to also consider the social aspects of AD. ...
... B. Pedestrian Models 1) Level-1: To resemble a basic but rational human crossing behavior [10], we define a pedestrian policy which evaluates the TTC value at each time step t according to ...
... See Section III-B.2 for a description of these components. The AV's action space U AV is equivalent to (10). Similar to (8), the AV's reward function R AV with reward r AV t+1 is described by ...
Reliable pedestrian crash avoidance mitigation (PCAM) systems are crucial components of safe autonomous vehicles (AVs). The nature of the vehicle-pedestrian interaction where decisions of one agent directly affect the other agent's optimal behavior, and vice versa, is a challenging yet often neglected aspect of such systems. We address this issue by defining a Markov decision process (MDP) for a simulated driving scenario in which an AV driving along an urban street faces a pedestrian trying to cross an unmarked crosswalk. The AV's PCAM decision policy is learned through deep reinforcement learning (DRL). Since modeling pedestrians realistically is challenging, we compare two levels of intelligent pedestrian behavior. While the baseline model follows a predefined strategy, our advanced pedestrian model is defined as a second DRL agent. This model captures continuous learning and the uncertainty inherent in human behavior, making the vehicle-pedestrian interaction a deep multi-agent reinforcement learning (DMARL) problem. We benchmark the developed PCAM systems according to the agents' collision rate and the resulting traffic flow efficiency with a focus on the influence of observation uncertainty or noise on the decision-making of the agents. The results show that the AV is able to completely mitigate collisions under the majority of the investigated conditions, and that the DRL pedestrian model indeed learns a more intelligent crossing behavior.
... [67] and [68] showed that pedestrians used mainly the car's speed to decide to cross or not. However, an experiment by [69] came to a different conclusion. The authors observed that when the car was approaching faster, pedestrians accepted smaller TTC when crossing. ...
... The authors observed that when the car was approaching faster, pedestrians accepted smaller TTC when crossing. They concluded that the distance rather than the speed of the car or the TTC was perceived and used in pedestrian decision-making [69]. That conclusion was refuted by studies [70] and [71] which have shown that when the speed of the approaching vehicle was high, pedestrians were more likely to underestimate its speed. ...
... TTC. Although there is no one-fits-all TTC for gentle car braking initiation according to [117], [21], [69], and [82] found similar results, with accepted TTC to cross from 3 s to 7 s. The TTC pedestrians accept to cross is influenced by the car's speed. ...
Pedestrians will increasingly have to share their space with autonomous vehicles (AVs), at pedestrian crossings, and in urban shared spaces where segregation between pedestrians and vehicles is minimized. This article proposes an integrative framework to analyze pedestrian behavior in shared spaces with AVs. Following the "perception-cognition-action" cycle, the proposed framework breaks down pedestrian behavior into 3 parts: 1. Sensation and Perception; 2. Emotion and Cognition; 3. Action and Communication. The framework is used to review and synthesize current knowledge on pedestrian behavior in urban shared spaces. Studies involving pedestrians in shared environments with AVs are limited. Since an AV can be seen as a hybrid of a conventional car and a mobile robot, this review includes studies on pedestrian behavior with conventional cars and with mobile robots as well as with AVs. Examining pedestrian behavior in these three situations of interaction allows us to make assumptions about how humans will behave in sharing their urban spaces with AVs. The reviewed interactions reveal that pedestrians have diverse and imperfect behaviors. AVs must consider this variety of behaviors and follow socially compliant rules in order to be understood and accepted by pedestrians. Perspectives for AVs in shared spaces and research directions are also identified.
... 2) Behavior Features: As proposed by Schmidt and Färber [89], using only trajectory information for intention prediction is insufficient. The behavioral features, especially the appearance and posture, usually indicate a pedestrian's intention, and are used by many intention prediction and joint prediction works. ...
... 2) Heterogeneous -Interaction With Other Road Users (ORUs): The future behavior of pedestrians is influenced by the interaction with ORUs such as vehicles according to Shirazi et al. [9]. a) Hand-crafted features: In Schmidt and Färber's research [89], parameters such as the distance and velocity of the vehicles can influence the crossing intention. For the intention prediction, many researchers used hand-crafted features as inputs, such as in studies [7], [64], [67], [79], including vehicle's velocity or speed, relative velocity and distance between the pedestrian and vehicle, or time to collision (TTC). ...
The prediction of pedestrian behavior is essential for automated driving in urban traffic and has attracted increasing attention in the vehicle industry. This task is challenging because pedestrian behavior is driven by various factors, including their individual properties, the interactions with other road users, and the interactions with the environment. Deep learning approaches have become increasingly popular because of their superior performance in complex scenarios compared to traditional approaches such as the social force or constant velocity models. In this paper, we provide a comprehensive review of deep learning-based approaches for pedestrian behavior prediction. We review and categorize a large selection of scientific contributions covering both trajectory and intention prediction from the last five years. We categorize existing works by prediction tasks, input data, model features, and network structures. Besides, we provide an overview of existing datasets and the evaluation metrics. We analyze, compare, and discuss the performance of existing work. Finally, we point out the research gaps and outline possible directions for future research.
... The basic approach followed by these studies is to predict pedestrians' 15 crossing intention using their historical trajectory. Later studies suggested that using historical trajectory 16 data is insufficient alone (18). A variety of studies tried to predict pedestrians crossing intention using 17 their body language. ...
... The dependent variable, i.e. future state of the pedestrian, is a categorical variable 15 that has two classes: walking and waiting. 16 17 Figure 5 Description of input variables 18 All the independent and dependent variables with their descriptions are summarised in Table 1 The variables discussed are extracted at every 0.5s interval separately for each jaywalking pedestrian. The 4 ...
In this study, a framework is proposed to predict jaywalkers' future state by employing machine learning algorithms. Different variables such as pedestrian pose, walking speed, location in the road environment, count and direction of approaching traffic, speed and type of closest approaching vehicle in upstream, etc., are used as input variables. The dataset for this study consists of 47588 samples gathered by analysing 1753 jaywalkers under non-lane-based heterogeneous traffic situations. Keypoint detection on the pedestrian body is made using MediaPipe. YOLOv4 and DeepSORT are used to detect and track road users to get trajectory data. By testing the performance of several machine learning models based on evaluation metrics, the best model is determined. Training and testing datasets are prepared for different prediction horizons to test the proposed models’ applicability for roads of varying design speeds. Four machine learning models based on ensemble techniques such as Random Forest (RF), AdaBoost, Gradient Boosting, and Extreme Gradient Boosting are trained and tested for different prediction horizons from 0.5s to 4s. Up to the prediction horizon of 1s, all models have performed equally well with AUC values above 0.95. At higher prediction horizons, Random Forest is found to be outperforming other models. All models, except AdaBoost, maintained an AUC value of greater than 0.9 when predicting the future state up to a maximum of 2.5s. The outcomes of this study can be utilised to assist Connected and Autonomous Vehicles (CAV), infrastructure-to-vehicle (I2V) connectivity, and driver assistance technologies in empowering vehicles to navigate through jaywalkers safely, enhancing pedestrian safety.
... Although pedestrian intentions can be studied via self-reported roadside observations or interviews (Sucha, Dostal, & Risser, 2017) and experiments with recruited actors (Schmidt & Faerber, 2009;Palmeiro, et al., 2018), these methods are usually intrusive, non-naturalistic, or limited with qualitative metrics. They are insufficient to study the dynamic changes of the lower-level intentions in different situations. ...
... Instead, prior studies have tried to study pedestrian intention from the driver's perspective with video experiments. Schmidt et al. (Schmidt & Faerber, 2009) had participants watch pedestrian crossing videos and estimate intent to cross the street at predetermined critical frames. By blocking different video parts, the experiments try to detect the influence of pedestrian features on the driver's decision-making. ...
... The importance of such communication to ensure roadway safety has been widely documented in the field of traffic psychology. For instance, Schmidt and Färber (2009) discussed the role of eye contact between pedestrians and drivers. They showed that pedestrians willing to cross a street usually look at the approaching vehicle, to make sure the driver sees them. ...
... Moreover, beyond the stress experienced, a new set of road safety issues also arise according to the behavioral difference between automated versus conventional vehicles piloted by real humans, specifically regarding older people. Indeed, the literature shows that elderly people have more difficulties than younger pedestrians to identify time gap and make a safe decision to cross, particularly in the frame of complex traffic situations or when facing a continuous flow of approaching vehicles (Lobjois and Cavallo, 2007;Schmidt and Färber, 2009;Dommes, 2019;Núñez Velasco et al., 2019). Speeds of approaching vehicles were also identified as particularly important risk factors for elderly pedestrians, who may have difficulties to perceive and to adequately estimate them before implementing their roadcrossing (Cavallo et al., 2009;Beggiato et al., 2017). ...
This study focuses on Automated Vehicles (AVs) interactions with pedestrians during road crossing situations. A dual-phase experiment was designed: one from the pedestrian’s perspective and the other one from the AV passenger’s point of view. Eight AV behaviors to yield were investigated. Participants’ task was to assess the safety of each one of these yielding behaviors. Moreover, an external HMI (eHMI) was designed to support them in these interactions. 40 participants were involved in this experiment (50% females, 20 young versus 20 elderly). Results obtained show significant differences between old and young participants: elderly people have not the same way to perceive and assess the safety of the yielding behaviors from “the inside” and from “the outside” of the car. Conversely, young participants assessed AV behaviors similarly whether as pedestrians or as AV passengers. When considering benefits introduced by the eHMI, it significantly reduces differences between old and young participants and tends to harmonize their safety assessments: with to the eHMI, elderly people are more able to adequately perceive and assess the safety/dangerousness of the AV braking manoeuvers, and their safety judgments become at last quite similar to those of young participants. Moreover, the eHMI increases participants’ Acceptance of AV and reduces their concerns about their future interactions with AV as a pedestrian, especially for elderly people.
... To simplify the probabilistic prediction model of pedestrian action, it is necessary to find out the most important factors that influence pedestrian action. In Schmidt and Farber [26], the data analysis showed that pedestrians relied on the distance of the car. Besides, in Rasouli et al. [22], pedestrian behavior in complex traffic scenes is investigated by using information such as weather, types of roads, time-to-collision, pedestrian's characteristics (age, gender). ...
... The goal is to analyze how the above information influences the pedestrian when a vehicle is approaching, whether the pedestrian will pass through the crosswalk or stop. Stimulated by Schmidt and Farber [26] and Rasouli et al. [22], this paper addresses the near-accident pedestrian action modeling with simple models, intending to provide a practical method with a simple model to predict the near-accident event and pedestrian action. The near-accident data set has been pre-processed for this study. ...
This paper proposes an innovative framework of modeling the statistical properties of the near-accident event and pedestrian behavior at non-signalized intersections based on Poisson process and logistic regression. The first contribution of this study is that the predictive intensity model of the near-accident event is established by regarding the near-accident event as a Poisson process on space of the vehicle velocity, distance to the intersection and lateral distance to the pedestrian at the time when pedestrian appears. Besides, logistic regression is used to build the model which can predict the probability of pedestrian behavior. The two proposed models are validated in a generative simulation. The simulated pedestrian behavior data is generated by the proposed models and compared with the real data. The real data set is from the drive recorder data base of Smart Mobility Research Center (SMRC) at Tokyo University of Agriculture and Technology. Accident and near-accident data has been collected in the city streets with an image-captured drive recorder mounted on a taxi since 2006. The findings in this study are expected to be useful for constructions of traffic simulators or safety control design which considers the pedestrian-vehicle interaction.
... Prior work has shown that pedestrian non-verbal actions (e.g., body posture) play a key role in influencing driving behavior [11,43]. Many studies investigated different aspects of pedestrian actions at crosswalks such as eye contact before crossing [10,40,45], including how often and when they typically occurred [5]. Pedestrian conditions like standing or walking before crossings were also studied [33,41]. ...
... Pedestrian conditions like standing or walking before crossings were also studied [33,41]. Researchers demonstrated that body language like hand, leg, and head movements are important cues of pedestrian actions [18,40,42] and often help drivers to yield [10,36], thereby generating a positive or negative interaction [9,16]. ...
... As can be seen from the previous work, traditional interactions between motorized vehicles and pedestrians play a fundamental role in infuencing the pedestrian's crossing decision, which relies on the vehicle's speed and distance to estimate both the awareness and the intent of the driver [48,53,55]. Although pedestrians and cyclists receive additional information about what a car intends to do with an introduction of on-vehicle displays, they still heavily rely on the drivers' behavior inside the vehicles. ...
... To safely cross a street, people usually rely on a distance to a vehicle and its speed [8,48,53,55]. These factors play an essential role (2) Countdown -depicts the remaining time the intersection is safe to cross (right). ...
Cycling has become increasingly popular as a means of transportation. However, cyclists remain a highly vulnerable group of road users. According to accident reports, one of the most dangerous situations for cyclists are uncontrolled intersections, where cars approach from both directions. To address this issue and assist cyclists in crossing decision-making at uncontrolled intersections, we designed two visualizations that: (1) highlight occluded cars through an X-ray vision and (2) depict the remaining time the intersection is safe to cross via a Countdown. To investigate the efficiency of these visualizations, we proposed an Augmented Reality simulation as a novel evaluation method, in which the above visualizations are represented as AR, and conducted a controlled experiment with 24 participants indoors. We found that the X-ray ensures a fast selection of shorter gaps between cars, while the Countdown facilitates a feeling of safety and provides a better intersection overview.
... For instance, it was found that pedestrians who want to cross the street without a crossing aid often use eye contact to ensure that an approaching driver sees them [36]. If the driver returns the eye contact, pedestrians assume that they have been seen and that the driver will act accordingly [87]. This assumption is supported by studies showing that a pedestrian's direct stare towards an oncoming driver invokes more compliant and yielding behavior [40,41]. ...
... Variations in traffic culture exist not only between different countries but also within the same country, such as between urban and rural areas or between different cities [9]. Culture has been shown to have an effect on gap acceptance [87], perceptions of traffic problems such as speeding or jaywalking [60], and the assessment of traffic situations [16]. Besides culture, past experience may also play an important role in crossing behavior. ...
As the control of driving tasks is increasingly transferred from the human driver to the on-board sensor and computer systems, a potential gap in communication is created between the car as an entity, and other road users. How can pedestrians, cyclists, and other drivers be certain that an automated vehicle is aware of the different road users in its environment and will do the ‘right thing’? In the need for creating a sense of optimal trust in automated vehicles, particularly in the nascent stages of their development and introduction to traffic, this communication gap needs to be filled. This chapter looks at the state of the art of the research that tries to answer this question, and lays out some common considerations and recommendations for the design of such systems.
... Furthermore, the pedestrian's behavior is more flexible than other road user's behavior, as they can quickly switch from one state to another (walking, running, stepping back, stopping, etc.). Also, pedestrians can change suddenly their direction (Schmidt and Faerber, 2009), and they do not have a flashing lights or horn to warn other road users. This diversity may lead to unexpected behavior, as viewed from automated vehicle, and thus to accidents. ...
... This may help the experimenter implementing virtual reality experiments with non-normative pedestrians in order to study drivers or passengers of automated vehicles facing unexpected pedestrian crossings. Furthermore, by improving our model, it might help to improve driver assistants (Schmidt and Faerber, 2009) for a better anticipation of the pedestrian crossing decision. ...
Automated vehicle driving raises the challenge of producing vehicle behaviors similar to those produced by human drivers. Automated vehicles could then be accepted more easily. Among the many different behaviors to be simulated, one is particularly important from a road safety point of view: the interaction between automated vehicles and pedestrians during a street crossing. More particularly, pedestrians non-normative and unexpected behaviors should be taken into account by automated vehicles. Studying the behaviors of drivers or passengers in these situations becomes a major stake for the acceptance of automated vehicles. Thus, in this contribution, we aim to develop a new model of pedestrian street crossing at a red light for virtual pedestrians in order to facilitate the scenario development. This model will allow virtual pedestrians to cross the street illegally, and to influence other pedestrians to follow them. Thus, some among the waiting pedestrians will also illegally cross the street. The model is able to manage in a very simple manner a group of pedestrians with only few parameters. These parameters allow to coordinate the pedestrians in the group for street crossing, and experimenters will keep the control on their studies in ceteris paribus conditions. The model is based on three hypotheses. The first one concerns the time that each pedestrian accepts to wait at the red light. The second and third hypotheses are related to social influence: the actions made by others influence the pedestrian street crossing decisions. Based on the works of Rosenbloom (2009), we assume that: i) waiting pedestrians encourage the others to wait too, and: ii) crossing pedestrians at a red light shows the others there is an opportunity to cross, and thus encourages them to cross. The simulation results illustrate our method and show different pedestrians group behaviour of street crossings. Our method allows experimenters to coordinate the pedestrians in the group with only few parameters.
... However, AVs will likely share traffic space with conventional traffic users, including pedestrians. Established communication cues between pedestrians and drivers, such as eye contact (Sucha et al. 2017) or body movements (Schmidt and Färber 2009), may become less reliable as AV drivers may be distracted or even absent (Mahadevan et al. 2018). Therefore, it is vital to understand pedestrian crossing behavior in response to oncoming AVs. ...
With the development of autonomous vehicle (AV) technology, understanding how pedestrians interact with AVs is of increasing importance. In most field studies on pedestrian crossing behavior when encountering AVs, pedestrians were not permitted to physically cross the street due to safety restrictions. Instead, the physical crossing experience was replaced with indirect methods (e.g., by signalizing with gestures). We hypothesized that this lack of a physical crossing experience could influence the participants’ crossing behavior. To test this hypothesis, we adapted a reference study and constructed a crossing facility using a virtual reality (VR) simulation. In a controlled experiment, the participants encountered iterations of oncoming AVs. For each interaction, they were asked to either cross the street or signify their crossing decisions by taking steps at the edge of the street without crossing.
Our study reveals that the lack of a physical crossing can lead to a significantly lower measured critical gap and perceived stress levels, thus indicating the need for detailed analysis when indirect methods are applied for future field studies.
Practical Relevance: Due to safety requirements, experiments will continue to measure participants’ crossing behavior without permitting them to physically walk in front of an oncoming vehicle. Our study was the first attempt to reveal how this lack of crossing could potentially affect pedestrians’ behavior, and we obtained empirical evidence in support of our hypothesis, thus providing insights for future studies.
... Second, as shown in Fig. 5, this could be because the drivers of this group, on average, drove faster replicating the findings in the literature that pedestrians are more likely to accept smaller gaps when the speed of approaching vehicle is higher (S. Schmidt & Faerber, 2009;K. Tian et al., 2022;Velasco et al., 2019). ...
One of the current challenges of automation is to have highly automated vehicles (HAVs) that communicate effectively with pedestrians and react to changes in pedestrian behaviour, to promote more trustable HAVs. However, the details of how human drivers and pedestrians interact at unsignalised crossings remain poorly understood. We addressed some aspects of this challenge by replicating vehicle-pedestrian interactions in a safe and controlled virtual environment by connecting a high fidelity motion-based driving simulator to a CAVE-based pedestrian lab in which 64 participants (32 pairs of one driver and one pedestrian) interacted with each other under different scenarios. The controlled setting helped us study the causal role of kinematics and priority rules on interaction outcome and behaviour, something that is not possible in naturalistic studies. We also found that kinematic cues played a stronger role than psychological traits like sensation seeking and social value orientation in determining whether the pedestrian or driver passed first at unmarked crossings. One main contribution of this study is our experimental paradigm, which permitted repeated observation of crossing interactions by each driver-pedestrian participant pair, yielding behaviours which were qualitatively in line with observations from naturalistic studies.
... In the present study, it is imperative to identify a gap that is neither too large nor too small and will be deemed acceptable by pedestrians. Prior research examining midblock crossings has reported that no one crossed below 3 seconds, and all crossed over 8 seconds (Schmidt & Färber, 2009). Research also indicated that the near-side gap is considered the critical gap when pedestrians cross at midblock locations (Wang et al., 2010). ...
One of the main challenges autonomous vehicles (AVs) will face is interacting with pedestrians, especially at unmarked midblock locations where the right-of-way is unspecified. This study investigates pedestrian crossing behavior given different roadway centerline features (i.e., undivided, two-way left-turn lane (TWLTL), and median) and various AV operational schemes portrayed to pedestrians through on-vehicle signals (i.e., no signal, yellow negotiating indication, and yellow/blue negotiating/no-yield indications). This study employs virtual reality to simulate an urban unmarked midblock environment where pedestrians interact with AVs. Results demonstrate that both roadway centerline design features and AV operations and signaling significantly impact pedestrian unmarked midblock crossing behavior, including the waiting time at the curb, waiting time in the middle of the road, and the total crossing time. But only the roadway centerline features significantly impact the walking time. Participants in the undivided scene spent a longer time waiting at the curb and walking on the road than in the median and TWLTL scenes, but they spent a shorter time waiting in the middle. Compared to the AV without a signal, the design of yellow signal significantly reduced pedestrian waiting time at the curb and in the middle. But yellow/blue significantly increased the pedestrian waiting time. Interaction effects between roadway centerline design features and AV operations and signaling are significant only for waiting time in the middle. For middle waiting time, yellow/blue signals had the most impact on the median road type and the least on the undivided road. Demographics, past behaviors, and walking exposure are also explored. Older individuals tend to wait longer, and pedestrian past crossing behaviors and past walking exposures do not significantly impact pedestrian walking behavior.
... Foregoing in view, the decisions of pedestrians were also influenced by road width, availability of the median (refuse island), as well as width of the median. Besides accepting shorter gap sizes (Schmidt & Färber, 2009), pedestrians tended to violate more on narrow roads . In addition, the compliance rate of pedestrians was decreased for the existence of a median . ...
To ensure the safety of pedestrians, it is essential to have a comprehensive understanding of their road crossing behaviors, including the factors that influence the decisions they make regarding crossing. One of the crucial crossing behaviors of pedestrians is the crossing pattern, which refers to whether a pedestrian crosses the road by walking or running. Safety of the pedestrians often depends on it as running crossing pattern is one of the riskiest crossing behaviors. However, there is a lack of inclusive studies that explore pedestrians’ decision regarding their crossing pattern. Therefore, this study aimed to identify the significant factors influencing pedestrians' decision regarding their crossing patterns (walk or run) at intersections in Dhaka, Bangladesh, using the chi-square test, and to examine the association between the identified contributory factors and crossing pattern using the association rules mining technique. Pedestrian road crossing behaviors, their characteristics, and traffic characteristics related data were collected from six busy intersections in Dhaka using the videography survey method. Findings of the study showed that walking crossing pattern was strongly associated with the factors such as controlled intersection, narrow road, wide median, female pedestrian, older pedestrian, using two-stage strategy, group crossing, accepting larger gap, using crosswalk, and crossing in front of slower vehicles. Besides, running crossing pattern was strongly associated with uncontrolled intersection, wide road, narrow median, male pedestrian, younger pedestrian, using rolling gap strategy, crossing alone, accepting shorter gap, crossing through conflict zone, and crossing in front of light and faster vehicles. The findings of this study would aid policymakers to develop effective solutions to improve pedestrian safety as well as to design future technologies like automated driving systems.
... Currently, drivers typically communicate their driving intentions to pedestrians via signals from the vehicle, eye contact, gestures, etc. [3][4][5][6][7]. These communication methods (facial expression, eye contact, gesture, vehicle movement, and the sound from the vehicle) allow pedestrians to clearly understand the intentions of car drivers and to be aware of upcoming vehicles [8][9][10][11][12]. While crossing a road, pedestrians can assess whether they can safely cross an intersection based on the speed and acceleration of vehicles as well as the distance between the vehicles and themselves [13,14]. ...
In this study, a method is devised that allows the intentions of autonomous vehicles to be effectively communicated to pedestrians and passengers via an efficient interactive interface. Visual and auditory factors are used as variables to investigate the effects of different autonomous vehicle signal factors on the judgment of pedestrians and to determine the main factors such that the best combination can be proposed. Two visual dimensions (i.e., color and flashing) and three auditory dimensions (i.e., rhythm, frequency, and melody) are used as the experimental signal variables. In addition, deceleration and waiting-to-restart scenarios are investigated. Multiple-choice questions and a subjective cognition scale are used for evaluation. The results show that the combination of green and slow rhythm can be used for the road-user-first case, whereas the combination of red and fast rhythm can be used for the vehicle-first case. Under the same intention, factors of color, flashing, rhythm, and melody are highly similar in terms of the combination mode, except for the frequency. In the deceleration and waiting-to-restart scenarios, the frequencies of the best signal are high and low frequencies, respectively. The results of this study can be used as a reference for the signal design of autonomous vehicles in the future and provide ideas for the interactions between autonomous vehicles and pedestrians.
... For the case of Z Car , [SF09] estimated that the visual angle threshold for the perception of the time-to-contact with an oncoming vehicle is about 0, 17°/s. [YS21] found that for an average male pedestrian around 1.54 meters, a distance around 62.7 meters is needed to perceive an oncoming vehicle moving at 80 km/h. ...
This thesis aims to increase the heterogeneity of pedestrian interactions in virtual urban environments. The interactions focus on collision avoidance and street crossing. My first contribution consists in allowing the pedestrian agents to avoid the collisions by adapting his behavior according to the neighbor agents with which they interact. Simulations show the expected adaptation during the avoidance interactions by increasing or decreasing the avoidance effort depending on the behavior of the perceived neighbors. Then, in the cases of unidirectional flows, the pedestrian agent will seek to overtake its slower neighbors. Based on the position, the speed, and the size of the neighbors, the pedestrian agent will take into account particularly the neighbor in front of him (called the leader), who constraints him the most. Therefore, my second contribution combines a collision avoidance model with a queueing model by considering the leader differently from the rest of the perceived neighbors. These two contributions rely on the perception of the perceived neighbors' physical characteristics (size, position, speed). However, the temporal dimension (how long the agent has been in a situation) and the social influence from the neighbors behaviors are important factors in decision-making. My main contribution is the development of a decision model for street crossing, which considers the waiting time on the sidewalk of the pedestrian agent and the actions of his neighbors (cross or wait). The model is assessed under two scenarios, 1) a street crossing with a pedestrian light, without road traffic, and 2) a street crossing with traffic without a traffic light. The model is based on three assumptions. The first one is that waiting time is an important factor in the decision to cross. On one hand, we assume that a pedestrian agent is willing to wait a certain amount of time before crossing (a sort of patience) and beyond that, the pedestrian will want to cross even at a red light. On the other hand, we assume that the waiting time plays a role in the perception of the vehicle in front of which the pedestrian plans to cross. Indeed, literature shows that the time-to-contact with an oncoming vehicle can be overestimated or underestimated before the crossing decision.The next two hypotheses are based on Rosenbloom's work, who suggests that a pedestrian waiting at a red light may be influenced by those crossing and those waiting. Combining these two hypotheses forms what we call the social influence, which will modulate both the patience of the pedestrian agent and his perception of the time-to-contact of the oncoming vehicles.In a first study, the simulations show that pedestrians supposed to cross at the red light may wait for the green light influenced by waiting neighbors. Conversely, neighbors crossing at the red light will encourage the agent to cross during the red light, leading to a different decision compared to being alone in this situation. Moreover, pedestrians with similar characteristics (patience, speed) and perceiving the same situation (color of the traffic light and number of neighbors) but arriving at the crossing location at different times will make different decisions. In the second study, crossing in front of road traffic, the combination of social influence and waiting time will induce a bias in the perception that may lead the pedestrian agent to overestimate or underestimate the time-to-contact with the vehicle. The pedestrian agent may thus decide to wait when he could cross, missing an opportunity; he also may decide to cross when the situation does not allow it, which leads to an inadequate decision
... Several other concepts that simply utilize light ramps to indicate speeding up or down have also been proposed but they seem to lack the same empirical support [28,29]. Cues such as posture and head movement seem to be more than enough to indicate pedestrians' intentions [30], suggesting that body language is at least as important as eye contact in conveying intent in traffic. One could assume also that recognizing a vehicle accelerating or decelerating is key information when pedestrians assess their situation. ...
The Operator 4.0 typology depicts the collaborative operator as one of eight operator working scenarios of operators in Industry 4.0. It signifies collaborative robot applications and the interaction between humans and robots working collaboratively or cooperatively towards a common goal. For this collaboration to run seamlessly and effortlessly, human-robot communication is essential. We briefly discuss what trust, predictability, and intentions are, before investigating the communicative features of both self-driving cars and collaborative robots. We found that although communicative external HMIs could arguably provide some benefits in both domains, an abundance of clues to what an autonomous car or a robot is about to do are easily accessible through the environment or could be created simply by understanding and designing legible motions.
... Mean pedestrian walking speeds for different pedestrian types are set based on the field study by Knoblauch et al. [23]. Since pedestrian gap acceptance can differ widely depending on the country and environment [3], we set the default gap acceptance range following the study conducted by Schmidt et al. [24] in Europe, which is in agreement with an earlier European study by Ashworth [25] and North American one by Brewer et al. [16]. ...
... Data-driven approaches had a major share in the used methods as they can combine our knowledge about human behavior with other visual and spatial information, such as the works in [Has+15; Völ+16; Ras+19; CMF19; AA20]. According to [SF09] a visual information of the pedestrian such as gaze or body movement detection is imperative to predict the intention. While a more recent study in [DT17b] argues that the detection of explicit body or gaze communication between the pedestrian and the vehicle is not significant in the intention prediction process. ...
Application of deep learning to geometric 3D data poses various challenges for researchers. The complex nature of geometric 3D data allows to represent it in different forms: occupancy grids, point clouds, meshes, implicit functions, etc. Each of those representations has already spawned streams of deep neural network models, capable of processing and predicting according data samples for further use in various data recognition, generation, and modification tasks.Modern deep learning models force researchers to make various design choices, associated with their architectures, learning algorithms and other specific aspects of the chosen applications. Often, these choices are made with the help of various heuristics and best practice methods discovered through numerous costly experimental evaluations. Probabilistic modeling provides an alternative to these methods that allows to formalize machine learning tasks in a meaningful manner and develop probability-based training objectives. This thesis explores combinations of deep learning based methods and probabilistic modeling in application to geometric 3D data.The first contribution explores how probabilistic modeling could be applied in the context of single-view 3D shape inference task. We propose a family of probabilistic models, Probabilistic Reconstruction Networks (PRNs),which treats the task as image conditioned generation and introduces a global latent variable, encoding shape geometry information. We explore different image conditioning options, and two different training objectives based on Monte Carlo and variational approximations of the model likelihood. Parameters of every distribution are predicted by multi-layered convolutional and fully-connected neural networks from the input images. All the options in the family of models are evaluated in the single-view 3D occupancy grid inference task on synthetic shapes and according image renderings from randomized viewpoints. We show that conditioning the latent variable prior on the input images is sufficient to achieve competitive and state-of-the-art single-view 3D shape inference performance for point cloud based and voxel based metrics, respectively. We additionally demonstrate that probabilistic objective based on variational approximation of the likelihood allows the model to obtain better results compared to Monte Carlo based approximation.The second contribution proposes a probabilistic model for 3D point cloud generation. It treats point clouds as distributions over exchangeable variables and use de Finetti’s representation theorem to define a global latent variable model with conditionally independent distributions for coordinates of each point. To model these point distributions a novel type of conditional normalizing flows is proposed, based on discrete coupling of point coordinate dimensions. These flows update the coordinates of each point sample multiple times by dividing them in two groups and inferring the updates for one group of coordinates from another group and, additionally, global latent variable sample by the means of multi-layered fully-connected neural networks with parameters shared for all the points. We also extend our Discrete Point Flow Networks (DPFNs) from generation to single-view inference task by conditioning the global latent variable prior in a manner similar to PRNs from the first contribution. Resulting generative performance demonstrates that DPFNs produce sets of samples of similar quality and diversity compared to state of the art based on continuous normalizing flows, but are approximately 30 times faster both in training and sampling. Results in autoencoding and single-view inference tasks show competitive and state-of-the-art performance for Chamfer distance, F-score and earth mover’s distance similarity metrics for point clouds.
... Along with the progress, the role of human drivers will be transformed into passengers, who no longer participate in traffic interactions. Due to this transformation, the conventional communication modalities between the driver and the surrounding road users, such as eye contact, facial expression, and hand gestures, will be impaired or dropped [2], [3]. Since these communication modalities are frequently used in ambiguous traffic situations for signaling intentions or negotiating rights of way, their ineffectiveness may bring interaction problems for autonomous vehicles (AVs) in future road systems [4]. ...
In order to facilitate safe and efficient interactions between autonomous vehicles (AVs) and pedestrians, lighting communication functions through extended signal forms such as light patterns and pictograms are of particular interest to express AV intentions. Currently, there is no consensus on which signal form should be used for Autonomous Vehicle-to-Pedestrian (AV2P) communications. Considering that the understandability of signal forms is crucial for the effectiveness of AV2P communications and thereby for choosing the appropriate signal form, we conducted a controlled investigation, consisting of designing forty-nine signals in the forms of light patterns and pictograms to express the vehicle yielding intention, and evaluating their performance in terms of understandability through monitor-based tests with more than five hundred online participants. It is found that, with the blank background where no external influence existed, the light patterns performed badly for correctly expressing the vehicle yielding intention but they were able to convey a warning message. The pictograms performed relatively well for expressing the yielding intention, but the diverse types of confusion were observed for those which were not well designed. With the vehicle-pedestrian encounter scenario where the vehicle yielding behaviors were explicit, the light patterns and the pictograms all performed well and the difference between them became less evident. These findings add new insights on the difference of understandability between light patterns and pictograms as available signal forms for AV2P communication functions, and are discussed in terms of usage recommendations in future road systems.
... Mean pedestrian walking speeds for different pedestrian types are set based on the field study by Knoblauch et al. [23]. Since pedestrian gap acceptance can differ widely depending on the country and environment [3], we set the default gap acceptance range following the study conducted by Schmidt et al. [24] in Europe, which is in agreement with an earlier European study by Ashworth [25] and North American one by Brewer et al. [16]. ...
In this paper, we present a microscopic agent-based pedestrian behavior model Intend-Wait-Cross. The model is comprised of rules representing behaviors of pedestrians as a series of decisions that depend on their individual characteristics (e.g. demographics, walking speed, law obedience) and environmental conditions (e.g. traffic flow, road structure). The model's main focus is on generating realistic crossing decision-model, which incorporates an improved formulation of time-to-collision (TTC) computation accounting for context, vehicle dynamics, and perceptual noise. Our model generates a diverse population of agents acting in a highly configurable environment. All model components, including individual characteristics of pedestrians, types of decisions they make, and environmental factors, are motivated by studies on pedestrian traffic behavior. Model parameters are calibrated using a combination of naturalistic driving data and estimates from the literature to maximize the realism of the simulated behaviors. A number of experiments validate various aspects of the model, such as pedestrian crossing patterns, and individual characteristics of pedestrians.
... The label of a video sequence (transition or no-transition) is determined by the TTE tag of its last frame. As crossing-related cases are arguably the most critical for stop and go forecasting, we set the prediction time horizon λ to be 2 seconds, which is the minimum time within which pedestrians make crossing decisions [51]. We choose a relatively large sampling rate of 5fps in the hope of reducing overfitting and speeding up training. ...
Forecasting pedestrians' future motions is essential for autonomous driving systems to safely navigate in urban areas. However, existing prediction algorithms often overly rely on past observed trajectories and tend to fail around abrupt dynamic changes, such as when pedestrians suddenly start or stop walking. We suggest that predicting these highly non-linear transitions should form a core component to improve the robustness of motion prediction algorithms. In this paper, we introduce the new task of pedestrian stop and go forecasting. Considering the lack of suitable existing datasets for it, we release TRANS, a benchmark for explicitly studying the stop and go behaviors of pedestrians in urban traffic. We build it from several existing datasets annotated with pedestrians' walking motions, in order to have various scenarios and behaviors. We also propose a novel hybrid model that leverages pedestrian-specific and scene features from several modalities, both video sequences and high-level attributes, and gradually fuses them to integrate multiple levels of context. We evaluate our model and several baselines on TRANS, and set a new benchmark for the community to work on pedestrian stop and go forecasting.
... Data-driven approaches had a major share in the used methods as they can combine our knowledge about human behavior with other visual and spatial information, such as the works in [Has+15; Völ+16; Ras+19; CMF19; AA20]. According to [SF09] a visual information of the pedestrian such as gaze or body movement detection is imperative to predict the intention. While a more recent study in [DT17b] argues that the detection of explicit body or gaze communication between the pedestrian and the vehicle is not significant in the intention prediction process. ...
The current trend in electric autonomous vehicles design is based on pre-existing models of cities which have been built for cars. The carbon footprint of cities cannot be reduced until the overall requirement for vehicles is reduced and more green and pedestrianized zones are created for better livability. However, such green zones cannot be scaled without providing autonomous mobility solutions, accessible to people with reduced mobility. Such solutions need to be capable of operating in spaces shared with pedestrians, which makes this a much harder problem to solve as compared to traditional autonomous driving. This thesis serves as a starting point to develop such autonomous mobility solutions. The work is focused on developing a navigation system for autonomous vehicles operating around pedestrians. The suggested solution is a proactive framework capable of anticipating pedestrian reactions and exploiting their cooperation to optimize the performance while ensuring pedestrians safety andcomfort.A cooperation-based model for pedestrian behaviors around a vehicle is proposed. The model starts by evaluating the pedestrian tendency to cooperate with the vehicle by a time-varying factor. This factor is then used in combination with the space measurements to predict the future trajectory. The model is based on social rules and cognitive studies by using the concept of the social zones and then applying the deformable virtual zone concept (DVZ) to measure the resulting influence in each zone. Both parts of the model are learnt using a data-set of pedestrians to vehicle interactions by manually annotating the behaviors in the data-set.Moreover, the model is exploited in the navigation system to control both the velocity and the local steering of the vehicle. Firstly, the longitudinal velocity is proactively controlled. Two criteria are considered to control the longitudinal velocity. The first is a safety criterion using the minimum distance between an agent and the vehicle’s body. The second is proactive criterion using the cooperation measure of the surroundingagents. The latter is essential to exploit any cooperative behavior and avoid the freezing of the vehicle in dense scenarios. Finally, the optimal control is derived using the gradient of a cost function combining the two previous criteria. This is possible thanks to a suggested formulation of the cooperation model using a non-central chi distribution for the distance between the vehicle and an agent.A smooth steering is derived using a proactive dynamic channel method for the space exploration. The method depends on evaluating the navigation cost in a channel (sub-space) using a fuzzy cost model. The channel with the minimum cost is selected, and a human-like steering is affected using a Quintic spline candidate path between channels. Finally, the local steering is derived using a sliding mode path follower.The navigation is evaluated using PedSim simulator under ROS in pedestrian-vehicle interaction scenarios. The navigation is tested with different pedestrian density and sparsity. The proactive framework managed to navigate the vehicle producing smooth trajectories while maintaining the pedestrians’ safety and reducing the travel time in comparison with traditional reactive methods (Risk-RRT).
... In order to obtain the cognitive annotations, a video experiment was conducted for all the recruited 24 video annotators. Previous studies have used video experiments to study human neural and cognitive mechanisms in the activity segmentation process [40], investigate the driver's decision-making process in predicting pedestrian's behaviors [41], and explain the driving behaviors with text descriptions [17]. Furthermore, a recent study [11] adopted a similar video experiment process to annotate pedestrians' desire to cross the street. ...
Prediction of pedestrian behavior is critical for fully autonomous vehicles to drive in busy city streets safely and efficiently. The future autonomous cars need to fit into mixed conditions with not only technical but also social capabilities. As more algorithms and datasets have been developed to predict pedestrian behaviors, these efforts lack the benchmark labels and the capability to estimate the temporal-dynamic intent changes of the pedestrians, provide explanations of the interaction scenes, and support algorithms with social intelligence. This paper proposes and shares another benchmark dataset called the IUPUI-CSRC Pedestrian Situated Intent (PSI) data with two innovative labels besides comprehensive computer vision labels. The first novel label is the dynamic intent changes for the pedestrians to cross in front of the ego-vehicle, achieved from 24 drivers with diverse backgrounds. The second one is the text-based explanations of the driver reasoning process when estimating pedestrian intents and predicting their behaviors during the interaction period. These innovative labels can enable several computer vision tasks, including pedestrian intent/behavior prediction, vehicle-pedestrian interaction segmentation, and video-to-language mapping for explainable algorithms. The released dataset can fundamentally improve the development of pedestrian behavior prediction models and develop socially intelligent autonomous cars to interact with pedestrians efficiently. The dataset has been evaluated with different tasks and is released to the public to access.
... Beside using static environment for predictions, also the interaction of traffic participants is modeled in literature. In Schmidt and Faerber (2009), Schmidt analyzed the gap acceptance of pedestrians to cross the street, based on time to collision which is an important information for predicting the crossing behavior. Rasouli Rasouli et al. (2017) analyzed the typical behavior of traffic participants before crossing a road on zebra crossings or other situations. ...
Ensuring a safe journey with an autonomous vehicle, the surrounding has to be sensed and understood. Especially human intuition about the plans and intentions of traffic participants is hard to model for machines. In literature, there are already several prediction techniques existing for pedestrians, which are based on different features. Some models are very complex, whereas others only rely on the considered person’s motion. The goal of this work is to analyze the importance of different classes of context knowledge for the prediction performance, derive features to remove this lack of information and prove this by an improved prediction algorithm. In order to judge the lack of context knowledge, we analyze the prediction performance and error cases of a long short-term memory (LSTM) Neural Network as State-of-the-Art prediction algorithm, only based on motion data. The Network is trained and evaluated on a benchmark dataset, to make the results comparable to other approaches. Analyzing the most error-prone predictions, the missing context shall be identified, which could improve the prediction results. Since the data was generated by video, we can evaluate the whole scenario and identify the influencing factors. The found influences were classified in categories and their importance for the prediction model estimated. We prove the necessity of additional context knowledge by retraining a neural network with additional context knowledge. In a literature research we compare our found results to existing approaches.
Pose estimation has been a critical aspect for the recent improvements made in the field of pedestrian intent prediction. Current pose estimators are capable of providing highly accurate posture and head orientation information. In our previous work, we utilised posture information for predicting the crossing behaviour of pedestrians in urban environments. We referred to this as the multi-scale pedestrian intent prediction (MS-PIP) architecture. This technique yielded state-of-the-art results of 94% accuracy. It has been suggested from some previous works that head orientation information provides insight into the pedestrian’s behaviours and intentions. Therefore, in this study, we investigate the benefits of implementing head orientation on top of the existing MS-PIP architecture. We found that the addition of head orientation information in fact decreases accuracy when compared to our previous works, in some cases by over 50%. Data augmentation and data generalisation techniques were also applied which slightly improved the accuracy. However, the accuracy was still lower than the original MS-PIP results.KeywordsPedestriansPose EstimationAction Recognition
In this paper, we propose a trajectory prediction method that takes into account pedestrian behavior. To realize safe automated driving in urban areas, it is necessary to predict the future movements of road users. Pedestrians entering the vehicle's direction are a target of interest, but they are difficult to predict because they change their behavior through significant interactions with the vehicle. In this study, we first predict whether pedestrians will yield the way to a vehicle. Next, the predictions are then used to predict the future trajectories of all road users in the scene. The proposed model consists of two neural network structures: the Yielding judgment module to predict pedestrian behavior and the Trajectory prediction module. The yielding judgment module provides intuitive and easily understandable indicators, which also helps to increase the interpretability of the overall model. We evaluated the usefulness of the proposed model using a publicly available dataset. The proposed method was found to reduce the average displacement error by 2.79% and the final displacement error by 2.70% compared to the case where the target pedestrian's behavior is not considered.
The task of designing unequivocal communication between autonomous vehicles and humans is important, considering the supposed daily interaction which might take place in the future. Using vehicles which are still in development for a study, can be a security risk and might even endanger the subjects. At the same time, it can be a waste of resources to integrate an interaction model which might not work well. It is therefore crucial to create a safe but realistic environment, in which studies can be held. In this paper we study the interaction between humans and an autonomous vehicle called CityBot, by using self-developed VR and web-applications. We describe two conducted studies, which were focused on speech, gestures and audio signals as means of communication between both parties. We discuss the problems, as well as differences between gestures and speech for controlling the CityBot. Furthermore, we propose a prefered usage of speech and how to handle ambigious gestures.KeywordsVRHuman-Robot-InteractionUsability-Testing
This study proposes a novel framework to predict jaywalkers’ future state in non-lane-based heterogeneous traffic conditionsby combining the effects of the surrounding dynamics with jaywalkers’ poses. Different variables, such as the pedestrian pose,walking speed, location in the road environment, count and direction of approaching traffic, speed and type of closestapproaching vehicle, and so forth, are used as input variables. The dataset for this study consists of 47,588 samples gatheredby analyzing 1753 jaywalkers under non-lane-based heterogeneous traffic situations. Keypoint detection on the pedestrianbody is made using MediaPipe. YOLOv4 and DeepSORT are used to detect and track road users to get trajectory data.Training and testing datasets are prepared for different prediction horizons to test the proposed models’ applicability forroads of varying design speeds. Four machine learning models based on ensemble techniques, namely random forest (RF),adaptive boosting (AdaBoost), gradient boosting, and extreme gradient boosting, are trained and tested for different predic-tion horizons from 0.5 to 4 s. Up to the prediction horizon of 1 s, all models performed equally well with Area under theROC curve (AUC) values above 0.95. At higher prediction horizons, the RF is found to outperform the other models. Allmodels, except AdaBoost, maintained an AUC value of greater than 0.9 when predicting future states up to a maximum of2.5 s. The proposed model performs well for both short-term and long-term predictions by combining the effect of sur-rounding dynamics with pedestrian stance and speed. The outcomes can be utilized to assist infrastructure-to-vehicle connec-tivity in empowering vehicles to navigate through jaywalkers safely, enhancing pedestrian safety.
To navigate safely in traffic environments, road users must correctly predict another road users’ intentions. Understanding how road users correctly predict the intent of other road users can help create possible countermeasures for collision avoidance. The aim of this paper is to examine what cues road users (drivers, bicyclists, and pedestrians) use to successfully predict other road user’s intentions and to highlight gaps and outline future research directions. A systematic literature search using the PRISMA method was conducted, and twenty-seven articles were included in the review. Overall, the results from these studies suggest that observers use body language, cues exhibited by the road user, and seek eye contact, when making predictions of intent about another road user. Future research should aim to understand how specific cues impact a road user’s decision-making process and what factors (e.g., point of view or eye contact) modulate a road user’s prediction performance.
Pedestrians’ red-light crossing can present a threat to traffic safety. Among all the existing work related to pedestrian’s red-light crossing, there are few studies using trajectory data in time sequence. This paper uses pose estimation (keypoint detection) to generate pedestrians’ variables from CCTV videos. Four machine learning models are used to predict pedestrians’ crossing intention at intersections’ red-light. The best model achieves an accuracy of 0.920 and AUC value of 0.849, with data from three intersections. Different prediction horizons (up to 4 sec) are used. With longer prediction horizons, the sample size gets smaller, which partially leads to worse model performance. However, the performance with prediction horizon up to 2 sec is still good (AUC value as 0.841). It is found that keypoint variables such as the angles between ankle and knee (left side) and elbow and shoulder (right side) are important. This model can be further implemented in the Infrastructure-to-Vehicle (I2V) applications and thus prevent accidents due to pedestrians’ red-light crossing by issuing warnings to drivers.
Recent evidence suggests that the assumed conflict-avoidant programming of autonomous vehicles will incentivize pedestrians to bully them. However, this frequent argument disregards the embedded nature of social interaction. Rule violations are socially sanctioned by different forms of social control, which could moderate the rational incentive to abuse risk-avoidant vehicles. Drawing on a gamified virtual reality (VR) experiment ( n = 36) of urban traffic scenarios, we tested how vehicle type, different forms of social control, and monetary benefit of rule violations affect pedestrians’ decision to jaywalk. In a second step, we also tested whether differences in those effects exist when controlling for the risk of crashes in conventional vehicles. We find that individuals do indeed jaywalk more frequently when faced with an automated vehicle (AV), and this effect largely depends on the associated risk and not their automated nature. We further show that social control, especially in the form of formal traffic rules and norm enforcement, can reduce jaywalking behavior for any vehicle. Our study sheds light on the interaction dynamics between humans and AVs and how this is influenced by different forms of social control. It also contributes to the small gamification literature in this human–computer interaction.
Automated driving is transforming the driving experience in the 21st-century vehicle. As a result, interacting with in-vehicle information systems, infotainment, in-car productivity or social interactions and real-life experiences with other passengers in the car, are slowly emerging as primary activities. UX researchers focus more and more on the users not only by developing products and services for them and enhancing their experiences but also actively involving them in co-designing for their own experience. Our research with designers inside the automotive industry suggests that the industry is exceptionally traditional regarding the methods and tools used to design and evaluate interactive experiences in comparison to other domains. In this chapter, we will report on the limitations of the industry in comparison to academia. Besides, we will report on the needs of the automotive UX practitioners and discuss the state of the art methods and tools that are most valued in the automotive industry.
This distributed simulator study investigated pedestrians’ head-turning behaviour during a series of road crossings in a CAVE-based pedestrian simulator. Pedestrians were required to cross the road in front of an approaching vehicle, the kinematic behaviour of which was either programmed by the simulation to depict an automated vehicle (AV) or controlled by a human driver (HD), via a connected (hidden) desktop driving simulator. A within-participant experimental design was used with twenty-five pairs of participants (a pedestrian and a driver). For each trial, pedestrians had to decide whether to cross in front of the HD/AV, which was instructed (or programmed) to yield (or not) to the pedestrian. For the AV trials, two braking patterns were included: a hard-braking AV (AVHB, deceleration rate = 3.2 m/s2, stopping distance = 12 m from pedestrian) and soft-braking (AVSB, deceleration rate = 2.5 m/s2, stopping distance = 4 m from pedestrian). Pedestrians’ head-turning frequency and the change in head-turning angle, were calculated for each condition, both before a crossing was initiated, and during the actual road crossing. Results showed a significant increase in head-turning behaviour in the last 2 seconds before a crossing initiation in the yielding trials, in line with a ‘last-second check’ reported in observations of real-world crossings (Hassan, Geruschat, & Turano, 2005). The vehicle’s braking behaviour and stopping distance were the most important factors affecting pedestrians’ head-turning patterns during the crossing, with the least head-turning behaviour seen in the AVSB condition, compared with AVHB and HDB trials. This suggests that a closer stopping distance for the AV was associated with less confusion for the pedestrian, although this condition was also associated with the longest crossing initiation time. In contrast, the highest number of head-turnings were seen for the human-driven vehicle, which, on average, yielded about 40 m away from the participants, enabling a much faster crossing initiation. Overall, the shortest crossing initiation time (~ 1 sec) and highest head-turning behaviour were seen in the non-braking conditions, where participants crossed as quickly as the circumstances allowed. These results provide new insights about the use of VR simulators for understanding pedestrians’ crossing behaviour in response to different vehicle kinematics. They also extend our knowledge of pedestrian cues for the development of suitable sensors in future automated vehicles, which should help with providing a more seamless interaction between AVs and other road users in mixed traffic settings.
The importance of investigating pedestrian safety has been evaluated repeatedly in safety studies. The present study attempts to evaluate the various methods used by previous researchers in a hierarchical process, to determine the characteristics, advantages, and limitations of each method. Two general analysis approaches (passive and active) were taken into account to categorize 169 previous types of research. In the passive approach, the studied methods were those based on crash databases, questionnaires, and post-crash field observation data; while, in the active approach, the studied methods were those based on driving simulations and videography. The result of the passive approach reveals that road users’ features and road characteristics (crash database studies), and error, lapses, intentional and unintentional violations (questionnaire studies) by them were among the most important causes of crashes and conflicts. Furthermore, road users’ distractions also reported a set of factors affecting the possibility of conflicts and crashes based on post-crash field observation studies. Also, results of the active approach showed that risky behaviors are the most important factor in threatening pedestrian safety such as unauthorized speeding, non-compliance with traffic law, unauthorized overtaking by drivers, and illegal crossing. Furthermore, risk perception and decision-making processes are the most important bond between the attitude and behavior of road users in dangerous driving situations. Examining studies through passive approach would lead to identifying the causes of crashes, recognizing the attitude of road users towards safety, and determining road users' behavioral patterns in certain situations, while the active approach has led to a more detailed understanding of behaviors and attitudes of road users. The inference of the findings obtained in this study will lead to a better understanding of the behavior of road users for studies on advanced driving assistance systems (ADAS).
Two time-to-contact (T c) experiments are reported that manipulated the manner in which a visually simulated target vehicle disappeared from the screen. In both experiments, one condition featured the traditional, spontaneous disappearance of the vehicle. A contrasting condition featured the occlusion of the vehicle behind a natural object. The available visual information was essentially equivalent in each condition. If T c is specified by information in the expanding optic array alone, the two conditions should produce equivalent estimates of T c. Results of each experiment, however, showed estimates with 14% and 12% greater accuracy in the occlusion condition compared to the disappearance condition. This implies that T c judgments depend on more than the rate of optical expansion. In addition to the occlusion manipulation, factors influencing the accuracy of T c estimates included both the sex and age of the participant. In an effort to compare T c estimates with time-judgment ability, participants also performed a time-production task with the same temporal structure as the T c task but with no graphic scene representation. A positive relation was found but further clarification is still needed between these two capabilities.
This paper describes a system for pedestrian detection in infrared im- ages implemented and tested on an experimental vehicle. A specific stabilization procedure is applied after image acquisition and before processing to cope with vehicle movements affecting the camera calibration. The localization of pedestri- ans is based on the search for warm symmetrical objects with specific size and aspect ratio. A set of filters is used to reduce false detections. The final validation process relies on the human shape's morphological characteristics.
Theories in motor control suggest that the parameters specified during the planning of goal-directed hand movements to a visual target are defined in spatial parameters like direction and amplitude. Recent findings in the visual attention literature, however, argue widely for early object-based selection processes. The present experiments were designed to examine the contributions of object-based and space-based selection processes to the preparation time of goal-directed pointing movements. Therefore, a cue was presented at a specific location. The question addressed was whether the initiation of responses to uncued target stimuli could benefit from being either within the same object (object based) or presented at the same direction (space based). Experiment 1 replicated earlier findings of object-based benefits for non-goal-directed responses. Experiment 2 confirmed earlier findings of space-based benefits for goal-directed hand pointing movements. In Experiments 3 and 4, space-based and object-based manipulations were combined while requiring goal-directed hand pointing movements. The results clearly favour the notion that the selection processes for goal-directed pointing movements are primarily object based. Implications for theories on selective attention and action planning are discussed.
Three gender-balanced groups of 16 school children (5-6 years, 8-9 years, 11-12 years) participated in individual pretests of vision, hearing, and time to walk across a 12-m wide urban street and back. Each child then completed 10 roadside trials requiring judgement of the threshold point at which they would no longer cross in front of traffic approaching from their right. The judgements were made from a site immediately in front of a parked car at a point 2 m from the kerb and 4 m from the centre of the road. Traffic speeds and distances were measured using a laser speed and distance detector. The results indicated that, overall, distance gap thresholds remained constant regardless of vehicle approach speeds. Analysis of the thresholds for distance gap judgements for the 4-m half-street crossing showed that some of the older children could be expected to make safe decisions, but this was not so for the 5-6- and 8-9-year-olds at vehicle approach speeds above 60 kph. Almost two-thirds of the children reported using distance to judge gaps, which proved the least adequate strategy in terms of proportion of resultant safe decisions. The findings are discussed in relation to developing effective child pedestrian safety strategies.
A category of stimuli of great importance for primates, humans in particular, is that formed by actions done by other individuals. If we want to survive, we must understand the actions of others. Furthermore, without action understanding, social organization is impossible. In the case of humans, there is another faculty that depends on the observation of others' actions: imitation learning. Unlike most species, we are able to learn by imitation, and this faculty is at the basis of human culture. In this review we present data on a neurophysiological mechanism--the mirror-neuron mechanism--that appears to play a fundamental role in both action understanding and imitation. We describe first the functional properties of mirror neurons in monkeys. We review next the characteristics of the mirror-neuron system in humans. We stress, in particular, those properties specific to the human mirror-neuron system that might explain the human capacity to learn by imitation. We conclude by discussing the relationship between the mirror-neuron system and language.
Understanding the intentions of others while watching their actions is a fundamental building block of social behavior. The neural and functional mechanisms underlying this ability are still poorly understood. To investigate these mechanisms we used functional magnetic resonance imaging. Twenty-three subjects watched three kinds of stimuli: grasping hand actions without a context, context only (scenes containing objects), and grasping hand actions performed in two different contexts. In the latter condition the context suggested the intention associated with the grasping action (either drinking or cleaning). Actions embedded in contexts, compared with the other two conditions, yielded a significant signal increase in the posterior part of the inferior frontal gyrus and the adjacent sector of the ventral premotor cortex where hand actions are represented. Thus, premotor mirror neuron areas-areas active during the execution and the observation of an action-previously thought to be involved only in action recognition are actually also involved in understanding the intentions of others. To ascribe an intention is to infer a forthcoming new goal, and this is an operation that the motor system does automatically.
This paper summarises a Presidential Address to the Division of Traffic and Transportation Psychology at the 2002 International Congress of Applied Psychology. It considers whether traffic psychology is a distinct area of psychology, and concludes that the range of psychological approaches that understanding drivers and traffic requires is too pervasive for it to be so. The difficulties and shortcomings of various attempts to apply cognitive psychology to driving and traffic are explored, with respect to perceptual, motor and skilled aspects of the driving task. Examples are given of how ‘understanding driving’ poses theoretical challenges to mainstream cognitive psychology that have yet to be satisfactorily resolved.
This paper reports the first phase of a research program on visual perception of motion patterns characteristic of living
organisms in locomotion. Such motion patterns in animals and men are termed here as biological motion. They are characterized
by a far higher degree of complexity than the patterns of simple mechanical motions usually studied in our laboratories. In
everyday perceptions, the visual information from biological motion and from the corresponding figurative contour patterns
(the shape of the body) are intermingled. A method for studying information from the motion pattern per se without interference
with the form aspect was devised. In short, the motion of the living body was represented by a few bright spots describing
the motions of the main joints. It is found that 10–12 such elements in adequate motion combinations in proximal stimulus
evoke a compelling impression of human walking, running, dancing, etc. The kinetic-geometric model for visual vector analysis
originally developed in the study of perception of motion combinations of the mechanical type was applied to these biological
motion patterns. The validity of this model in the present context was experimentally tested and the results turned out to
be highly positive.
A multinomial logit model is used to examine pedestrian and driver reaction to “encounters” occurring on pedestrian crossings. The probabilities of a driver braking or weaving, and of a pedestrian continuing to cross in response to an encounter are identified for a variety of pedestrian, environmental, and traffic conditions. Results indicate that the most important explanatory variables included pedestrian distance from kerb, city size, number of pedestrians simultaneously crossing, vehicle speed, and vehicle platoon size. It is felt that the model performed well, should be applied in further studies, and could be a useful technique for identifying the most hazardous situations and locations within an area, for planning relevant safety measures, and for national research for developing traffic legislation.
Four questions concerning the perceptual source of information about time to contact (tc) are addressed: (a) What conditions are required for the optic variable tau to play a role in the perception of tc? (b) When these conditions are met, does tau alone provide sufficient information for accurate timing of interceptive actions? (c) Does a distance divided by velocity account of tc perception provide a convincing alternative to an account that is based on tau? (d) Is there any empirical evidence that distinguishes the two accounts? A "global" type of tau variable and a "local" type of tau variable are distinguished, each with different limitations. The discussion is largely concerned with local tau variables, 2 versions of which are identified. It is concluded that tau alone cannot provide sufficient information for skilled timing. An extended tau-based account presented in an earlier article (Tresilan, 1990) is discussed. It is argued that no extant empirical data can distinguish the extended account from the distance divided by velocity account.
Subjects estimated time of vehicle arrival while viewing twenty-four film clips of an approaching vehicle, half with a constant viewing time of 4.2 s and half with a constant vehicle-movement distance of 40 m. The distances from the subject at which the film ended were 20, 60, and 100 m. Speeds of approach varied between 7.45 and 15.44 m s-1. Performance was strongly dependent on age of the subject. Subjects in the 5-6-year-old group made estimates based on the distance of the vehicle; at 7-8 years an interaction between the effects of distance and velocity appeared and for 9-10-year-olds there was a main effect of the vehicle velocity. Only for adults was the information from distance and velocity fully integrated. There was no significant difference between males and females for any of the age groups. Performance of adults was very similar to that reported by other authors in that subjects underestimated the time to arrival of the vehicle, with estimated times about 60% of the actual times. Standard deviations of the estimated times were such that a small percentage of subjects overestimated times and hence would have caused a collision if they had proceeded with a crossing. The mechanism of time estimation was strongly dependent on the angular velocity of the vehicle subtended at the eye of the observer. This must exceed a threshold value of about 0.002 rad s-1 (adults) if a linear relationship between estimated and actual times is to be obtained.
Drivers' estimates of time to collision were determined in a laboratory simulation using film segments made from a following vehicle approaching a lead vehicle, which was also in motion. Headway, approach speed, and viewing time were varied to make a total of 48 conditions. It was found that, provided the angular velocity subtended by the lead vehicle was above a threshold value of about .003 radians/sec, the driver was able to give reasonable estimates of time to collision. The standard deviation of the estimates varied linearly with the time to collision. Although drivers underestimated the time to collision when it was small, the large standard deviation shows the possibility of rear-end collisions due to poor estimation of time to collision, especially when the times for control action and vehicle deceleration are considered.
The over-representation of older pedestrians in serious injury and fatal crashes compared to younger adults may be due, in part, to age-related diminished ability to select gaps in oncoming traffic for safe road-crossing. Two experiments are described that examine age differences in gap selection decisions in a simulated road-crossing environment. Three groups of participants were tested, younger (30-45 years), young-old (60-69 years) and old-old (>75 years). The results showed that, for all age groups, gap selection was primarily based on vehicle distance and less so on time-of-arrival. Despite the apparent ability to process the distance and speed of oncoming traffic when given enough time to do so, many of the old-old adults appeared to select insufficiently large gaps. These results are discussed in terms of age-related physical, perceptual and cognitive limitations and the ability to compensate for these limitations. Practical implications for road safety countermeasures are also highlighted, particularly the provision of safe road environments and development of behavioural and training packages.
Two experiments were conducted to study how age affects street-crossing decisions in an estimation task, with particular emphasis on how oncoming vehicle speed and a time constraint influence the time gap deemed acceptable for crossing. Experiment 1 showed that when there was a time constraint, all age groups selected a shorter time gap for the higher speed. This was associated with a large number of missed opportunities for the low speed and many unsafe decisions for the high speed. In the second experiment, which had no time constraint, young pedestrians operated in a constant-time mode regardless of speed, whereas older pedestrians accepted shorter and shorter time gaps as speed increased. The results seem to indicate that the effect of speed is due to a mixed operating mode of participants, whose decisions may be based on either time or vehicle distance, depending on the task requirements and on the participant's own ability to meet those requirements.
We examine the behaviour of pedestrians wishing to cross a stream of traffic at signalized intersections. We model each pedestrian as making a discrete crossing choice by comparing the gaps between vehicles in traffic to an individual-specific 'critical gap' that characterizes the individual's minimal acceptable gap. We propose both parametric and nonparametric approaches to estimate the distribution of critical gaps in the population of pedestrians. To estimate the model, we gather field data on crossing decisions and vehicle flows at three intersections in New Delhi. The estimates provide information about heterogeneity in critical gaps across pedestrians and intersections, and permit simulation of the effect of changes in traffic light sequences on pedestrian crossing behaviour and waiting times. Copyright © 2005 John Wiley & Sons, Ltd.
This paper presents a method for pedestrian detection and tracking using a single night-vision video camera installed on the vehicle. To deal with the nonrigid nature of human appearance on the road, a two-step detection/tracking method is proposed. The detection phase is performed by a support vector machine (SVM) with size-normalized pedestrian candidates and the tracking phase is a combination of Kalman filter prediction and mean shift tracking. The detection phase is further strengthened by information obtained by a road-detection module that provides key information for pedestrian validation. Experimental comparisons (e.g., grayscale SVM recognition versus binary SVM recognition and entire-body detection versus upper-body detection) have been carried out to illustrate the feasibility of our approach.
Intelligent vehicles and unattended driving systems ofthefuturewillneedtheabilitytorecognizerelevant tra#c participants (such as other vehicles, pedestrians, bicyclists, etc.) and detect dangerous situations ahead of time. An important component of these systems is one that is able to distinguish pedestrians and track their motion to make intelligent driving decisions. The associated computer vision problem that needs to be solved is detection and tracking of pedestrians from a moving camera, which is extremely challenging. Robust pedestrian tracking performance can be achieved by temporal integration of the data in a probabilistic setting. Weemployashapemodelforpedestriansand an e#cient variant of the Condensation tracker to achieve these objectives. The tracking is performed in the high-dimensional space of shape model parameters which consists of Euclidean transformation parameters and deformation parameters. Our Condensation tracker employs sampling on quasirandom points, improving its asymptotic complexity and robustness, and making it amenable to realtime implementation.
Unfallgeschehen im Straßenverkehr
- Statistik Bundesamt Für
Bundesamt für Statistik (2007). Unfallgeschehen im Straßenverkehr 2006. Fachserie 8 Reihe 7, Wiesbaden.
Visual perception of biological motion and a model for its analysis
- Johannson
Johannson, G. (1973). Visual perception of biological motion and a model for its analysis. Perception and Psychophysics, 14(2), 201-211.
Pedestrian tracking from a moving vehicle
- V Philomin
- R Duraiswami
- L Davis
Philomin, V., Duraiswami, R. & Davis, L. (2000). Pedestrian tracking from a moving vehicle. In Proceedings of the IEEE intelligent vehicles symposium. Vol. IV,
pp. 350-355.
Development, analysis and testing of a pedestrian alert system (PAS)
- C Rodgers
- D Greenlee
- R Blomberg
Rodgers, C., Greenlee, D., & Blomberg, R. (2002). Development, analysis and testing of a pedestrian alert system (PAS). In Proceedings of the ION GPS,
September 2002.
System and method for providing pedestrian alerts
- C E Rodgers
- D F Greenlee
- R D Blomberg
Rodgers, C. E., Greenlee, D. F., & Blomberg, R. D. (2006). System and method for providing pedestrian alerts. US-Patent 7,095,336; 25 pp.
Geht er oder geht er nicht? - Ein FAS zur Vorhersage von Fußgängerabsichten
- S Schmidt
- B Färber
- A Grassi
Schmidt, S., Färber, B., & Pérez Grassi, A. (2008). Geht er oder geht er nicht? -Ein FAS zur Vorhersage von Fußgängerabsichten. In M. Maurer & C. Stiller
(Eds.), Workshop Fahrerassistenzsysteme, Freundeskreis Mess -und Regelungstechnik Karlsruhe e.V., Walting, 2-4 April 2008.
Fusion von MIR-Bildern und Lidardaten zur Klassifikation menschlicher Verkehrsteilnehmer
- M Thuy
- A Pérez Grassi
- V A Frolov
- F Puente León
Thuy, M., Pérez Grassi, A., Frolov, V. A., & Puente León, F. (2008). Fusion von MIR-Bildern und Lidardaten zur Klassifikation menschlicher
Verkehrsteilnehmer. In M. Maurer & C. Stiller (Eds.), Workshop Fahrerassistenzsysteme, Freundeskreis Mess -und Regelungstechnik Karlsruhe e.V.,
Walting, 2-4 April 2008.
Klassifikation menschlicher Verkehrsteilnehmer in MIR-Bildern: eine erste Annäherung
- Pérez Grassi
Grasping the intentions of others with one’s own mirror neuron system
- Iacobini