Conference PaperPDF Available

Guidelines for eye tracker calibration using points of regard

Authors:

Abstract

Eye movement data may be used for many various purposes. In most cases it is utilized to estimate a gaze point - that is a place where a person is looking at. Most devices registering eye movements, called eye trackers, return information about relative position of an eye, without information about a gaze point. To obtain this information, it is necessary to build a function that maps output from an eye tracker to horizontal and vertical coordinates of a gaze point. Usually eye movement is recorded when a user tracks a group of stimuli being a set of points displayed on a screen. The paper analyzes possible scenarios of such stimulus presentation and discuses an influence of usage of five different regression functions and two different head mounted eye trackers on the results.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
1
2
3
4
5
6
7
8
10
11
12
13
9
0
0.2
0.4
0.6
0.8
1 0 0.2 0.4 0.6 0.8 1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
xs=f(xe, ye)
ys=f(xe, ye)
xeyexsys
xs=Axxe+Bxye+Cx
ys=Ayxe+Byye+Cy
xs=Axx2
e+Bxy2
e+Cxxe+Dxye+Ex
ys=Ayx2
e+Byy2
e+Cyxe+Dyye+Ey
xs=Axx3
e+Bxy3
e+Cxx2
eye+Dxxey2
e+Exxeye
+Fxx2
e+Gxy2
e+Hxye+Ixye+Jx
ys=Ayx3
e+Byy3
e+Cyx2
eye+Dyxey2
e+Eyxeye
+Fyx2
e+Gyy2
e+Hyye+Iyye+Jy
C= 10 γ= 8
Edeg
R2
Edeg =1
nX
i
p(xibxi)2+ (yibyi)2
R2= 1 Pi(yibyi)2
Pi(yi¯y)2
yi, xibyi,bxi
¯y, ¯x
R2
R2
xR2
y
x1x2x3
x3
Edeg
Edeg Edeg
x2x2
x1
x1x3
x3
x2x2
p= 0.07
x2x1x2
x1
x3x1
Edeg Edeg
x1x2
x2x1
x3
x3
x1
Edeg Edeg
x3
x3x2
x2x1
x1
x2
Edeg Edeg
Edeg Edeg
x2
Edeg
Edeg
Edeg Edeg
x3x3
x2x3
x1x2
x2x2
x2x2
x2x3
x2x2
x2x3
x2x2
x2x2
Edeg
x3
Edeg
Edeg
Edeg
Edeg Edeg Edeg
x1x2x3
x2x1
... Measurements are mapped to computer coordinates by using different techniques (Martinez et al., 2012); however, regression using least squares is widely because of its low computational cost and acceptable accuracy. Different regression models (functions) can modify the performance of the calibration; hence, a reliable and simple model is required (Kasprowski et al., 2014). ...
... Their results show that second-degree models can perform similar to more complex models, with insignificant differences in the range of 1-2 pixels on a screen with resolution of 1024×768. Kasprowski et al. (2014) use linear, quadratic, and cubic polynomials with 3, 5, and 18 calibration points, respectively. They proposed discarding models based on the ratio of the sum of squares residuals to the sample variance. ...
... They proposed discarding models based on the ratio of the sum of squares residuals to the sample variance. Kasprowski et al. (2014) compare calibration results for two different head mounted eye trackers by using cross validation, root mean error, coefficient R, and square and absolute error. Their results show that the optimal degree for the polynomial calibration function depends on the specific ET device. ...
Article
The straightforward approach to eye-tracker calibration considers that the calibration data do not have erroneous associations, and the calibration function is defined. The violation of the non-erroneous assumption could cause an arbitrarily large bias. The MMransac algorithm proposed in this paper is a modified version of the Random Sample Consensus. that achieves robust calibrations. On the other hand, polynomials in two variables (i.e., with terms in the form κxayb) are commonly used to map eye-tracker measurements to points on the screen. High-degree polynomials tend to be more accurate; however, as the order is increased, the function becomes more complex and less smooth, which could cause over-fitting. In this sense, this paper proposes an algorithmic approach that enables model selection criteria even in the presence of outliers. This approach was tested using different model selection criteria. Results show that more accurate calibrations are obtained with the combined robust fitting and model selection approach using the Akaike information criterion (AIC) and the Kullback information criterion (KIC).
... -Very often, electric components like sensors may not work as expected from the onset and are required to be calibrated for several times to get finer results [139]. For example, visual saliency-based ASD detection approaches use eye trackers that are often required to be calibrated several times to include information about shapes, light reflection, and refraction properties of the different parts of the eyes, and during calibration, participants are asked to look at different positions on the screen [140]. -An autistic individual is very unlikely to cooperate in the time of calibration. ...
Article
Full-text available
Abstract Background/ Introduction: Autism Spectrum Disorder (ASD) is a neuro-developmental disorder that limits social and cognitive abilities. ASD has no cure so early diagnosis is important for reducing its impact. The current behavioural observation-based subjective-diagnosis systems (e.g., DSM-5 or ICD-10) frequently misdiagnose subjects. Therefore, researchers are attempting to develop automated diagnosis-systems with minimal human intervention, quicker screening time, and better outreach. Method: This paper is a PRISMA-based systematic review examining the potential of automated autism detection system with Human Activity Analysis (HAA) to look for distinctive ASD-characteristics such as repetitive behaviour, abnormal gait and visual saliency. The literature from 2011 onward is qualitatively and quantitatively analysed to investigate whether HAA can identify the features of ASD, the level of its classification accuracy, the degree of human intervention, and screening time. Based on these findings we discuss the approaches, challenges, resources, and future directions in this area. Result: According to our quantitative assessment of the dataset[1], Inception v3 and LSTM [1] give the highest accuracy (89%) for repetitive behavior. For abnormal gait-based approach, the Multilayer Perceptron gives 98% accuracy based on 18 features from dataset [2]. For gaze pattern, a saliency-metric feature-based learning [3] gives 99% accuracy on dataset[4], while an algorithm involving statistical features and Decision Trees yields an accuracy of 76% on dataset [5]. Conclusion: In terms of the state-of-the-art, fully automated HAA systems for ASD diagnosis show promise but are still in developmental stages. However, this is an active research field, and HAA has good prospects for helping to diagnose ASD objectively in less time with better accuracy.
... The polynomial of the given order approximates the underlying signal within the window. The polynomial order was based on previously conducted research [25]. ...
Chapter
Full-text available
Motion analysis is widely used to monitor the dynamics of the gaze movement signal. The performance analysis in the spectral domain shows a variation of the characteristics due to the development of changes in dynamic behavior. Each dynamic reaps characteristics in its spectral domain. After examining the spectrum of gaze movement, one can provide valuable information about defects that develop with visible artifacts. The evaluation of the gaze movement is based on the presence of characteristic frequencies and its harmonics in the acquired movement signal. In this work, a signal processing analysis of the movement of the gaze in the domain of time and frequency was performed. The empirical mode decomposition has been proposed to plot the five first intrinsic mode functions in the polar form to detect the alteration of the gaze movement dynamics. The results show that the evaluation of the performance of the gaze movement dynamics can be done by observing the polar representation only.
... The polynomial of the given order approximates the underlying signal within the window. The polynomial order was based on previously conducted research [25]. ...
Chapter
Full-text available
Biological movement analysis describes specific characteristics of the person’s health status, which motivates the analysis of the dynamic aspects related to the rhythms and movements of our body. Natural, free movements can be periodic or irregular in time and space, and each type of dynamic behavior can be related to efficient or altered movements. This research describes an overview of nonlinear dynamics and concepts of chaos applied to bio-motion paths as a way to describe and analyze a bio-movement, for example, the eye movement presents different rhythms depending on the demand for exploration or the state of health of the person. For this research, 20 subjects with normal vision were involved. Their eye movements and head movements were registered using an eye tracker and head-mounted tracker, measuring the position and rotation of the eye and head. The results demonstrate that nonlinear analysis can be applied to evaluate alteration in the biological system showing more sensitivity than traditional spectral analysis. This type of evaluation can provide a tool to find the relationship between movement dynamics and physiological phenomena, which might be useful to describe any alteration in the biological system.
... Practically, this would entail that the very stimulus positions used in the actual task are also used for calibration. Furthermore, increasing the number of calibration targets or using smooth pursuit eye data can improve accuracy (Hassoumi et al., 2019;Kasprowski et al., 2014;Pfeuffer et al., 2013). ...
Preprint
Full-text available
The marmoset has emerged as a promising primate model system, in particular for visual neuroscience. Many common experimental paradigms rely on head fixation and a sustained period of eye fixation during the presentation of salient visual stimuli. Both of these behavioral requirements can be challenging for marmosets. Here, we present two methodological developments to overcome these difficulties. First, we show that it is possible to use a standard eye tracking system without head fixation to assess visual behavior in the marmoset. Eye tracking quality from head-free animals is sufficient to measure precise psychometric functions from a visual acuity task. Secondly, we introduce a novel method for efficient receptive field mapping that does not rely on moving stimuli but uses fast flashing annuli and wedges. We present data recorded in areas V1 and V6 and show that receptive field locations are readily obtained within a short period of recording time. Thus, the methodological advancements presented in this work will contribute to establish the marmoset as a valuable model in neuroscience.
... -Very often, electric components like sensors may not work as expected from the onset and are required to be calibrated for several times to get finer results [139]. For example, visual saliency-based ASD detection approaches use eye trackers that are often required to be calibrated several times to include information about shapes, light reflection, and refraction properties of the different parts of the eyes, and during calibration, participants are asked to look at different positions on the screen [140]. -An autistic individual is very unlikely to cooperate in the time of calibration. ...
Preprint
Autism Spectrum Disorder (ASD) is a neuro-developmental disorder that limits social interactions, cognitive skills, and abilities. Since ASD can last during an affected person's entire life cycle, the diagnosis at the early onset can yield a significant positive impact. The current medical diagnostic systems (e.g., DSM-5/ICD-10) are somewhat subjective; rely purely on the behavioral observation of symptoms, and hence, some individuals often go misdiagnosed or late-diagnosed. Therefore, researchers have focused on developing data-driven automated diagnosis systems with less screening time, low cost, and improved accuracy while significantly reducing professional intervention. Human Activity Analysis (HAA) is considered one of the most promising niches in computer vision research. This paper aims to analyze its potentialities in the automated detection of autism by tracking the exclusive characteristics of autistic individuals such as repetitive behavior, atypical walking style, and unusual visual saliency. This review provides a detailed inspection of HAA-based autism detection literature published in 2011 on-wards depicting core approaches, challenges, probable solutions, available resources, and scopes of future exploration in this arena. According to our study, deep learning outperforms machine learning in ASD detection with a classification accuracy of 76\% to 95\% on different datasets comprise of video, image, or skeleton data that recorded participants performing a large number of actions. However, machine learning provides satisfactory results on datasets with a small number of action classes and has a range of 60\% to 93\% accuracy among numerous studies. We hope this extensive review will provide a comprehensive guideline for researchers in this field.
Article
Full-text available
This paper, the last in the series "Advanced tools of optometry and visual psychophysics," is devoted to optical tracking of eye movements. The optical methods include: photo-oculography, video-oculography, video-based corneal and combined pupil/corneal reflection oculography, and photosensor oculography. OPTYKA issue 2(75)2022, pp. 42-43
Article
Full-text available
The marmoset has emerged as a promising primate model system, in particular for visual neuroscience. Many common experimental paradigms rely on head fixation and an extended period of eye fixation during the presentation of salient visual stimuli. Both of these behavioral requirements can be challenging for marmosets. Here, we present two methodological developments, each addressing one of these difficulties. First, we show that it is possible to use a standard eye tracking system without head fixation to assess visual behavior in the marmoset. Eye tracking quality from head-free animals is sufficient to obtain precise psychometric functions from a visual acuity task. Secondly, we introduce a novel method for efficient receptive field mapping that does not rely on moving stimuli but uses fast flashing annuli and wedges. We present data recorded during head-fixation in areas V1 and V6 and show that receptive field locations are readily obtained within a short period of recording time. Thus, the methodological advancements presented in this work will contribute to establish the marmoset as a valuable model in neuroscience.Significance StatementThe marmoset monkey is becoming an increasingly relevant model for biological and medical research. Here, we present two methodological advancements for visual neuroscience that are adapted to the marmoset. First, we present a head-free eye tracking protocol that is sufficiently accurate for a large variety of visual experiments. Second, we introduce an efficient technique for mapping visual receptive fields (RFs) and apply it to map RFs of neurons from the visual cortex of head-fixed marmosets. The concepts presented in this work can be easily transferred to other species. Together, this will promote diversification of the animal model landscape and solidify the contribution of marmoset research.
Chapter
Efficiency and ease of use are essential for practical applications of camera based eye/gaze-tracking. Gaze tracking involves estimating where a person is looking on a screen based on face images from a computer-facing camera. In this paper we investigate two complementary forms of efficiency in gaze tracking: 1. The computational efficiency of the system which is dominated by the inference speed of a CNN predicting gaze-vectors; 2. The usability efficiency which is determined by the tediousness of the mandatory calibration of the gaze-vector to a computer screen. To do so, we evaluate the computational speed/accuracy trade-off for the CNN and the calibration effort/accuracy trade-off for screen calibration. For the CNN, we evaluate the full face, two-eyes, and single eye input. For screen calibration, we measure the number of calibration points needed and evaluate three types of calibration: 1. pure geometry, 2. pure machine learning, and 3. hybrid geometric regression. Results suggest that a single eye input and geometric regression calibration achieve the best trade-off.
Conference Paper
Full-text available
In a video-based eye tracker the pupil-glint vector changes as the eyes move. Using an appropriate model, the pupil-glint vector can be mapped to coordinates of the point of regard (PoR). Using a simple hardware configuration with one camera and one infrared source, the accuracy that can be achieved with various mapping models is compared with one another. No single model proved to be the best for all participants. It was also found that the arrangement and number of calibration targets has a significant effect on the accuracy that can be achieved with the said hardware configuration. A mapping model is proposed that provides reasonably good results for all participants provided that a calibration set with at least 8 targets is used. It was shown that although a large number of calibration targets (18) provide slightly better accuracy than a smaller number of targets (8), the improvement might not be worth the extra effort during a calibration session.
Article
Full-text available
Recording eye movement data with high quality is often a prerequisite for producing valid and replicable results and for drawing well-founded conclusions about the oculomotor system. Today, many aspects of data quality are often informally discussed among researchers but are very seldom measured, quantified, and reported. Here we systematically investigated how the calibration method, aspects of participants' eye physiologies, the influences of recording time and gaze direction, and the experience of operators affect the quality of data recorded with a common tower-mounted, video-based eyetracker. We quantified accuracy, precision, and the amount of valid data, and found an increase in data quality when the participant indicated that he or she was looking at a calibration target, as compared to leaving this decision to the operator or the eyetracker software. Moreover, our results provide statistical evidence of how factors such as glasses, contact lenses, eye color, eyelashes, and mascara influence data quality. This method and the results provide eye movement researchers with an understanding of what is required to record high-quality data, as well as providing manufacturers with the knowledge to build better eyetrackers.
Conference Paper
In this paper, we present an appearance-based gaze estimation method for a head-mounted eye tracker. The idea is to extract discriminative image descriptors with respect to gaze before applying a regression scheme. We employ multilevel Histograms of Oriented Gradients (HOG) features as our appearance descriptor. To learn the mapping between eye appearance and gaze coordinates, two learning-based approaches are evaluated : Support Vector Regression (SVR) and Relevance Vector Regression (RVR). Experimental results demonstrate that, despite the high dimensionality, our method works well and RVR provides a more efficient and generalized solution than SVR by retaining a low number of basis functions.
Article
The nonlinear least-squares minimization problem is considered. Algorithms for the numerical solution of this problem have been proposed in the past, notably by Levenberg (Quart. Appl. Math., 2, 164-168 (1944)) and Marquardt (SIAM J. Appl. Math., 11, 431-441 (1963)). The present work discusses a robust and efficient implementation of a version of the Levenberg--Marquardt algorithm and shows that it has strong convergence properties. In addition to robustness, the main features of this implementation are the proper use of implicitly scaled variables and the choice of the Levenberg--Marquardt parameter by means of a scheme due to Hebden (AERE Report TP515). Numerical results illustrating the behavior of this implementation are included. 1 table. (RWR)