A Method for Standardizing Eye-Tracking and Behavioral Data

in Real and Virtual Environments

Maxime Dumonteil

1 a

, Marc J.-M. Mac

2 b

, Val

erie Gouranton

3 c

, Th

eophane Nicolas

4 d

and Ronan Gaugne

1 e

Univ. Rennes, Inria, CNRS, IRISA, France

CNRS, Univ. Rennes, Inria, IRISA, France

Univ. Rennes, INSA Rennes, Inria, CNRS, IRISA, France

Inrap, UMR Trajectoires, France

Keywords:

Methodology, Virtual Reality, Eye Tracking, Gaze Behavior, Immersive Room, HMD.

Abstract:

This paper introduces a methodology for generating standardized and comparable eye-tracking and behavioral

data across multiple modalities, in real and virtual environments. Our approach handles data collected using

different devices, thereby enabling a comprehensive comparison between different modalities: a real envi-

ronment, a virtual one using an immersive room setup, and another virtual environment using head-mounted

displays. The versatility of this methodology is illustrated through an archaeological case study, in which the

gaze patterns and behavioral responses of participants are analyzed while they interact with artifacts. How-

ever, this methodology is applicable to broader research areas involving eye tracking and behavior in mixed

environments. By explaining a workﬂow for the preparation, data acquisition, and post-processing of data,

our approach enables the generation of 3D eye-tracking and behavioral data. Subsequently, our presentation is

accompanied by examples of metrics and visualization that are relevant in such a comparison study, providing

insights into cross-modal behavioral and gaze pattern analysis.

1 INTRODUCTION

The study of the human gaze provides valuable in-

sights into human behavior, and recent advances in

virtual reality (VR) equipment, adding native eye

tracking (ET) capabilities, have opened up new and

exciting possibilities for research using eye-tracking

data. While the majority of current tools and methods

for studying gaze are based on a 2D context, there is

a growing interest in extending these analyses to a 3D

context. This move to 3D allows for a more complete

understanding of user behavior, especially when in-

teracting with virtual environments and objects.

Compared to the same case in a real environment,

object exploration and manipulation could differ in

a virtual context, notably eye and head coordination,

which could be inﬂuenced by the differences between

the two environments (Pfeil et al., 2018; Kollenberg

et al., 2010). The ﬁeld of view (FOV) is affected by

https://orcid.org/0009-0004-4267-4289

https://orcid.org/0000-0003-4124-8697

https://orcid.org/0000-0002-9351-2747

https://orcid.org/0000-0003-0098-9456

https://orcid.org/0000-0002-4762-4342

head-mounted devices (HMDs), which can inﬂuence

gaze behavior in complex VR scenarios in various

ways, such as object manipulation tasks or distance

estimation for example (Mizuchi and Inamura, 2018).

Moreover distance judgment is commonly underes-

timated in virtual environments when users wear an

HMD, partly due to the FOV restrictions but also due

to the weight of the device (Willemsen et al., 2009).

The use of HMDs also affects the vergence movement

of the eye, which consists in keeping the gaze posi-

tioned on an object depending on its distance and lo-

cation, so researchers aim to improve its computation,

tracking methods and calibration in 3D environments

(Duchowski et al., 2022). These studies demonstrate

an important level of interest for gaze behavior in vir-

tual environments.

This paper presents a methodological approach to

generate standardized and comparable eye-tracking

and behavioral data in real and virtual environments.

This workﬂow is focused on interaction and behav-

ior for visual exploration of an object. Our approach

addresses the challenges of comparing data collected

using different devices and environments, enabling a

comprehensive analysis of eye-tracking and behav-

Dumonteil, M., Macé, M. J., Gouranton, V., Nicolas, T. and Gaugne, R.

A Method for Standardizing Eye-Tracking and Behavioral Data in Real and Virtual Environments.

DOI: 10.5220/0013124100003912

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 20th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2025) - Volume 1: GRAPP, HUCAPP

and IVAPP, pages 211-222

ISBN: 978-989-758-728-3; ISSN: 2184-4321

211

Figure 1: Eye-tracking and motion tracking during the same task in reality (left), an immersive room (center) and a virtual

environment (HMD) (right).

ioral data in different modalities, presented in Fig.1: a

real environment, a virtual environment using an im-

mersive room setup and a virtual environment using

a VR HMD. However, comparing 3D gaze data in

different modalities presents unique challenges. Eye

and body movement tracking devices in real and vir-

tual environments produce heterogeneous streams of

information and performing comparable analysis re-

quires standardized methods of data collection and

analysis. Despite these challenges, several studies

have highlighted the validity of virtual environments

as counterparts to real-world scenarios in various

ﬁelds of research, with some researchers revealing

similarities in exploration patterns, especially for eye

ﬁxations between physical and virtual contexts (Gul-

han et al., 2021).

Our methodology is illustrated through an archae-

ological use case study, completely presented in (Du-

monteil et al., 2024), which focuses on characterizing

differences in an observation task to analyze a corpus

of ancient artifacts reproduced across the modalities.

Experimental results and description are detailed in

the mentioned work.

This methodology outlines a comprehensive

workﬂow for data preparation and analysis, starting

from experiment design, through data acquisition and

processing, and ending with the analysis and interpre-

tation, based on eye tracking results and movements

analysis. For each of these stages, several solutions

and recommendations are provided, along with de-

tailed descriptions of each step.

2 RELATED WORKS

This section ﬁrst lists a number of frameworks and

tools that facilitate the analysis of user behavior. It

then presents a group of studies that have processed

eye-tracking data in a real and mobile context.

Some solutions handle eye-tracking data in a

virtual context, such as PLUME, an open-source

toolbox for recording, replaying and analyzing XR

behavioral data, including physiological signals,

such as eye tracking or EEG for example (Javerliat

et al., 2024). Similarly other framework have been

developed but mainly focused on behavioral data

processing in virtual environment such as XREcho

(Villenave et al., 2022) or UXF (Brookes et al.,

2020), which collect and register behavioral events.

Another solution manages the recording and visual-

ization of gaze data during an exploration of a car

interior in VR (Li, 2021). In addition, commercial

software is used to capture or visualize behavioral

data from VR applications, such as Tobii Ocumen,

a toolkit designed for Tobii devices integrated into

HMDs (https://developer.tobii.com/xr/solutions/).

Cognitive3D proposes the same functionalities

(https://cognitive3d.com/). However the aforemen-

tioned studies only consider the processing of eye

tracking in a virtual environment. Furthermore all

of these behavior analysis solutions mainly handle

the HMD modality; they do not consider the con-

straints of the standardization and integration of other

modalities.

To compare human behavior between real and vir-

tual environments, we need a framework that can per-

form such a behavioral analysis. Therefore, it is im-

perative to improve the processing for a recording in

a real mobile context (Takahashi et al., 2021; Pfeif-

fer et al., 2016; Wang et al., 2017). To estimate the

gaze on a 3D model with respect to a real environ-

ment, some researchers (Paletta et al., 2013) develop

a method to generate a 3D map of the environment,

using a SLAM (Simultaneous Localization And Map-

ping) method then image descriptor matching without

ﬁducial markers to estimate and then project the user’s

attention on this virtual map. Similarly, some re-

searchers present pipelines to process 3D eye tracking

from 2D data but without any kind of markers, using

a structure-from-motion method (Li et al., 2020; Jo-

geshwar and Pelz, 2021). Even though these solutions

are efﬁcient, it is less suitable for mapping gaze with

GRAPP 2025 - 20th International Conference on Computer Graphics Theory and Applications

212

the precision required on a speciﬁc object than on an

entire environment. Finally, another study proposes a

workﬂow and an implementation to deﬁne an experi-

mental scenario for an eye tracking study in a virtual

environment, including preparation, data collection,

analysis, and visualization (Ugwitz et al., 2022). De-

spite the adapted methods that are presented and the

well explained implementation, this solution, like the

previous one, does not illustrate modality comparison

in eye-tracking processing and it is primarily focused

on one type of environment.

3 WORKFLOW

3.1 Overview

The objective of the presented methodology is the

generation of standardized and comparable eye-

tracking and behavioral data across different modal-

ities, for an object observation task. As presented in

Figure 2, it is structured around four components:

• Experimental settings: a deﬁnition of the proto-

col, the constraints and the resulting rules that are

necessary to establish a comparison between mul-

tiple modalities;

• Acquisition: a description of the required data for

processing and analysis and how to generate them,

using different devices for different modalities;

• Processing: a description of the required compu-

tational processes to transform the raw acquired

data into the standardized format of 3D gaze data;

• Interpretation: an overview of the analytical tech-

niques that can be applied to these data sets, facil-

itating the examination of cross-modal behavioral

and gaze pattern study.

3.2 Experimental Settings

The experimental settings section gathers three dif-

ferent activities. The ﬁrst concerns the deﬁnition of

the execution protocol for the task to be compared

between the different modalities. The next two are

protocol-dependent and concern the design of the data

acquisition apparatus and the 3D environment.

3.2.1 Protocol

The aim of the protocol is to enable a similar task to

be performed in interaction with an object, between

real and virtual environments. In order to remain as

general as possible, we consider two types of virtual

environments, the ﬁrst based on a VR HMD, and the

second based on a CAVE-like immersive space. Thus,

our purpose considers three different modalities:

• Real modality: the task is performed on the real

object;

• HMD modality: the task is performed in VR, on a

digital 3D copy of the object, wearing an HMD;

• Immersive modality: the task is performed in VR,

on a digital 3D copy of the object, in a CAVE-like

immersive room.

Both HMD and Immersive modalities are consid-

ered because of potential differences in user behavior.

In an immersive room, users can see their own body,

allowing for more natural interactions, which could be

particularly interesting for users unfamiliar with VR.

However, since HMDs are widely used for VR, it is

essential to include this modality in our study due to

its accessibility.

In order to obtain comparable behaviors, it is im-

portant to make the tasks that the participants have to

perform as similar as possible between the different

modalities. For the sake of generality, we will focus

on an object observation task. For this purpose, the

scenario of the task must be based on three elements:

(i) the scene, (ii) the positioning and the ability of the

participant to move in the scene, and (iii) the ability of

the participant to interact with the object considered

in the scene. Thus, we can state that the scene con-

sists of the object, the participant’s movement space,

and the interaction media between the participant and

the object.

The positioning of participants in a space and their

ability to move within that space is a central element

in a behavioral study. It is possible to constrain partic-

ipants’ movements by, for example, restricting them

to a seated position in the scene. However, it is more

interesting to take advantage of the natural movement

capabilities of virtual reality by allowing the user to

move around in the scene.

Similarly, it is important to clearly deﬁne the par-

ticipants’ ability to interact with the object. Ideally,

participants should be allowed to pick up the object

with their hands and handle it freely during the obser-

vation task. However, such a level of freedom in the

user’s interaction with the object can create complex

constraints in the management of data production and

analysis, particularly related to the object’s position

tracking capabilities and the risk of the object being

occluded by the participants’ hands. In order to give

participants a sense of control over their observation

task, we recommend implementing a moderate level

of interactions to provide a feeling of agency while

avoiding unnecessary complications.

Furthermore, it is recommended that a framework

A Method for Standardizing Eye-Tracking and Behavioral Data in Real and Virtual Environments

213

Figure 2: Workﬂow Components schema, structured in four blocks: Experiment Settings (purple), Acquisition (green), Pro-

cessing (blue) and Interpretation (red).

Figure 3: An assembly combining a system of Vulfoni

stereoscopic glasses with Tobii 2 eye tracking glasses, at-

tached together with a 3D-printed clip which is also de-

signed as the tracking constellation system.

for the task be established that allows for active obser-

vation by the participants. This can be accomplished

through the use of an object observation questionnaire

that can be completed by participants during the task

and includes questions related to the objects’ colors,

shapes, and textures.

3.2.2 Apparatus for Acquisition

The acquisition equipment is mainly concerned with

the acquisition of eye-tracking data and the position-

ing of moving elements in the environment, i.e. the

participant and the object.

Given the participant’s choice of mobility in the

scene, eye tracking is performed by equipment worn

by the participant. For the Real modality, there are

several models of eye-tracking glasses that meet this

constraint. Similarly, various models of VR HMDs

incorporate eye-tracking systems. The immersive

modality is often more complex, as there are no off-

the-shelves stereoscopic goggles with eye-tracking.

We present a solution example in the 3.2.4 section that

details an implementation for a speciﬁc use case.

Participant’s head tracking and object tracking are

directly taken into account in the two virtual modali-

ties (default API features for HMD and external cam-

era tracking for the immersive room). Different so-

lutions can be adopted for the Real modality such

as external IR-based mocap systems, inertial track-

ing systems or video-based systems. We chose to use

Figure 4: Augmented Reality markers (ArUco ﬁducial

markers) placed around the object of interest in order to

track the participants’ position relative to the object.

the built-in capabilities of eye-tracking goggles rely-

ing on the on-board camera and IMU. In this case,

the video captured by the on-board camera is used

as input to a processing algorithm to track the dis-

placement of elements using AR markers placed in

the scene as in Figure 4. Note that in this particular

case, as the markers are placed on a turntable moving

with the observed object, this method measures a rel-

ative position between the head and the object, which

may not be enough, depending of the use case.

3.2.3 3D Models

This part is related to the design of the virtual scene

that represents the real scene. It is important to de-

vote particular attention to the similarity between the

object representations in each modality, real or vir-

tual. All concerned objects of interest must be pro-

vided with an actual and a virtual version. Thus two

possible scenarios may be identiﬁed: the production

of a virtual version of a real object, or the generation

of a real object from a virtual model. The ﬁrst case

can be handled by a scanning method that produces a

3D mesh with high accuracy, such as photogramme-

try. In the second case, the real copy is made from an

original 3D virtual model using 3D printing for exam-

ple. In both scenarios, there are several points to bear

in mind: the quality of the produced (real or digital)

copy, its appearance (texture, color, ...) compare to

the original object and its accurate size.

GRAPP 2025 - 20th International Conference on Computer Graphics Theory and Applications

214

3.2.4 Implementation

We have implemented an archaeological use case ad-

hering to our workﬂow to allow comparison of the

same observation task on a corpus of archaeological

artifacts performed in different modalities.

The archaeological artifacts under observation

were placed on a turntable that the user could ma-

nipulate. They were allowed to freely move around

the table and to rotate the turntable during the obser-

vation task but could not directly touch and manip-

ulate the object. The observation questionnaire, an-

swered by the participants after each observation, was

based upon an actual archaeological artifact analysis

grid (Cauliez et al., 2002).

For the Real modality, the participant wore Tobii

Pro Glasses 2, a mobile eye-tracking system. ArUco

markers were disposed on the turntable around the ar-

tifact, and used to compute the relative position of

the user head with respect to the object. For the

HMD modality, they wore an HTC Vive Pro Eye

VR headset and ﬁnally for the Immersive modality,

they were placed in an immersive space constituted

of a 4-sided viewing screen measuring 10 m x 3 m

x 3 m (width, depth and height), tracked using pas-

sive markers detected by an Optitrack system. Stere-

oscopy was guaranteed by Vulfoni glasses, to which

Tobii eye-tracking goggles were attached using a 3D-

printed clip (Figure 3).

The observation task was based on a corpus of

three distinct potteries selected for their stylistic di-

versity. Facsimiles of the pottery were used during

the experiment as real objects, and virtual 3D models

were then generated using photogrammetry. The vir-

tual scene was a faithful representation of the real en-

vironment used in the experiment. The virtual scene

was designed and implemented using the Unity game

engine (version 2021.3.8f1).

3.3 Acquisition

The acquisition part comprises three different activ-

ities. The ﬁrst is the calibration of the different de-

vices involved in data acquisition between the dif-

ferent modalities. We separate the acquisition tasks

for the eye-tracking data and the different mobile ele-

ments of the scene, as the nature of the data collected

is signiﬁcantly different.

3.3.1 Calibration

The calibration process is an essential step to en-

sure data quality during eye tracking recording. This

mandatory step is completed before each task begins,

in all modalities and for each iteration. Manufactur-

ers usually provides a standard calibration procedure

that is simple to integrate in an experimental protocol.

Nevertheless even if the standard procedure is rela-

tively simple to setup, its convenience comes at the

cost of its quality. Some research work aim at enhanc-

ing the calibration process for eye-tracking glasses

(Onkhar et al., 2023; Liu et al., 2020) while keep-

ing it’s simplicity. However, in most case, record-

ing high-quality data requires a more elaborate and

time-consuming procedure. The decision depends on

the desired quality and, most importantly, the feasi-

bility of integrating a more complex procedure into

the protocol. With regard to elements tracking, both

lighthouse (for HMD modality) or IR tracking (for

Real or Immersive modality) have to be calibrated be-

forehand. The same is true of the inside-out HMD

tracking system, although it may require to repeat the

tracking calibration process if light conditions change

in the room for example.

3.3.2 Eye-Tracking Data

The initial objective of the eye-tracking data obtained

during the acquisition phase for the observation of an

object is to facilitate the calculation of user’s gaze,

based on the direction and the point of origin of the

gaze. Eye-tracking glasses, such as the Tobii Pro

Glasses 2, record the coordinates of the gaze on the

image of the scene (indicated by x

and y

in Fig-

ure 5), recorded by a camera placed at the center of

the glasses between the eyes. The output video cor-

responds to the user’s point of view, onto which the

gaze position is projected. In the case of HMD, the in-

tegrated eye-tracking system records a gaze ray, with

a point of origin and a direction.

In addition, eye-trackers record information about

eye blink and pupil diameter. The information used

to identify blinks varies with the device. For exam-

ple, on a Vive Pro Eye HMD, the data is taken di-

rectly from the Tobii API outputs, whereas on Tobii

Pro Glasses 2, the information is not directly avail-

able, but can be inferred from pupil data and gaze

position. A blink is identiﬁed as a loss of data for

both eyes (with null coordinates for gaze position).

Missing data can also occur if the user looks out of

the tracker’s ﬁeld of view, or if the device moves too

much on the head. Nevertheless, the manufacturer

also recommends using eye openness as a marker

for blinks onsets and offsets. The information of

eye blink and pupil diameter can be used to measure

speciﬁc user’s behavioral aspects such as cognitive

workload (Pomplun and Sunkara, 2019; Zhang et al.,

2015).

A Method for Standardizing Eye-Tracking and Behavioral Data in Real and Virtual Environments

215

Figure 5: Representation of the elements involved in calcu-

lating the intersection point (red) of the line of sight (green)

on the observed object mesh from 2D eye tracking (yellow)

and user’s head position data.

3.3.3 Elements Tracking Data

The tracking of environmental elements, such as the

user’s head and the object being observed (real or vir-

tual), necessitates the independent acquisition of po-

sition and rotation data for each element. This type

of data is typically generated directly by standard op-

tometric or inertial tracking systems. It can also be

complemented by videos via pose estimation for the

Real modality, or directly available in the engine soft-

ware for the irtual modalities.

3.3.4 Implementation

We used standard calibration procedures recom-

mended by the providers for the different tracking

system. The ET calibration was performed before

each observation task. In both the real and immer-

sive modalities, we collected the 2D coordinates of

gaze on the image of the scene provided by the ET

glasses. In the case of the HMD, we collected the

point of origin and gaze direction recorded by the in-

tegrated ET system. The positions and rotations of the

user’s head and the object were recorded for the HMD

and Immersive modalities. For the Real modality, we

implemented the method presented in the section 3.2,

collecting the videos recorded by the on-board cam-

era of the ET glasses, to perform pose estimations.

3.4 Processing

The main goal of the processing stage is to generate

3D gaze data about the observed object (in red in the

ﬁgure 5 in the form of coordinates (x,y,z)). The ﬁrst

step is to extract and prepare the necessary data from

the various data collected. This data is then used to

accurately calculate the coordinates of the user’s head

and the object observed in the virtual environment.

The ﬁnal step is then to calculate the gaze ray using

the ET data.

3.4.1 Data Extraction

Data extraction involves the collection and format-

ting of eye-tracking data and elements position and

rotation data from multiple acquisition sources. This

preliminary phase is based on simple data extraction

tools and depends on the acquisition devices used. It

can be performed on video data, CSV or JSON ﬁles.

3.4.2 3D Environment Data

The 3D environment data does not require speciﬁc

processing for the HMD and Immersive modalities.

They are directly collected in the log ﬁles of the ap-

plication. For the Real modality, the solution to re-

trieve the element tracking data depends on the ac-

quisition solution implemented. In the case of AR

marker-based tracking, a pose estimation is required.

It is presented in section 3.4.4 as we used this method

in our implementation of the workﬂow.

3.4.3 3D Gaze Data

The generation of 3D gaze data corresponds to com-

puting the position of the gaze on the object, i.e. the

point of intersection between the line of sight, in the

form of a vector (vx, vy, vz) (in green on Figure 5) and

the 3D model of the object. Three pieces of data are

therefore required: the direction of gaze, the position

and orientation of the user’s head, and the position

and orientation of the object. The calculation of the

last two is mentioned in the previous section.

Since the sampling frequency of the different data

is not the same between the different acquisition de-

vices, it is necessary to perform a synchronization cal-

culation of these data by interpolation. As standard,

this calculation takes as a reference the data ﬁle with

the lowest sampling rate and applies a linear interpo-

lation calculation to the other data ﬁles for each times-

tamp of the reference data.

Once all elements are correctly placed in relation

to each other, the line of sight is computed as a ray

cast in the direction of the gaze. This process depends

on the modality in which data was collected. Regard-

ing the HMD modality, as mentioned in section 3.3.2,

the line of sight is already provided as a vector us-

able for gaze ray-casting by the Tobii XR software

package associated to the HTC Vive Pro Eye that we

used in our implementation. For Real and Immer-

sive modalities, i.e. modalities that use eye-tracking

GRAPP 2025 - 20th International Conference on Computer Graphics Theory and Applications

216

glasses, there are two main methods to compute the

line of sight. The ﬁrst one is to use the gaze direction

provided by the ET system for each eye. The second

one is to use the projection of the sight on the model

of the ET system embedded camera. In this case, the

sight is projected on the virtual screen of the camera,

giving (x

, y

) coordinates that can be used to calcu-

late the gaze ray and subsequently the projection of

the sight on the 3D object.

3.4.4 Implementation

For the HMD modality, eye-tracking data (origin and

direction of gaze) was collected and aggregated from

the device API into a JSON ﬁle, while elements

movement data (position and rotation vectors) were

directly collected from Unity.

For the other two modalities that use ET glasses,

we collected and processed the raw data produced

by the device with a custom script that retains only

the necessary data, such as gaze coordinates in the

device’s video image. This video from the glasses

scene camera is exploited for both elements tracking

and eye tracking, with the 2D coordinates of the gaze

in the video frame. Therefore, it is crucial to syn-

chronize this high-frequency eye-tracking data with

the scene video frequency, interpolating these coor-

dinates to get coherent data from the eye-tracking

glasses.

In order to compute the 3D gaze data, we have

implemented a process to replay the records and thus

cast a ray to get the intersection point on the ob-

served object. For the tracking of elements in the

Real modality, we implemented the method presented

in (Takahashi et al., 2018), using an image process-

ing and pose estimation method with ArUco markers

(Garrido-Jurado et al., 2014) captured on the video

captured by the glasses. For the two modalities that

use ET glasses, a virtual representation of the ET sys-

tem embedded camera is used inside the processed

scene for computing the gaze ray from eye-tracking

data. This solution provides a more uniform handling

of the three modalities because, for the HMD modal-

ity, the solution is to use the straightforward gaze di-

rection data provide by the API.

3.5 Interpretation

The interpretation step of the methodology provides

analysis and restitution (graphical representations and

metrics) to compare the user’s behavior during the ob-

servation task between the different modalities.

3.5.1 Analysis

We are considering different tools to analyze the

user’s gaze and behavior in turn.

Gaze Event Detection There exist many different

eye movements studied in the literature, from the

most obvious and common ones to more subtle and

complex ones to detect. In order to keep the focus

on a comparative workﬂow between real and virtual

contexts, we will only consider the following three:

• Fixations: a movement when eyes are locked on

an object in order to stabilize the object on the

fovea for clear vision.

• Saccades: a fast eye movement between two ﬁxa-

tions.

• Smooth Pursuit (SP): a more complex movement

that tracks a moving target to keep it within the

fovea.

Many strategies have been developed to distin-

guish gaze events from each other, depending on the

target event and its deﬁnition. Examples of popu-

lar algorithms are presented in Table 1. The most

commonly used algorithms are based on two crite-

ria: velocity and dispersion. In velocity-based al-

gorithms, such as ”identiﬁcation by velocity thresh-

old” (IVT)(Salvucci and Goldberg, 2000) the veloc-

ity of the eye is exploited in order to identify ﬁx-

ations as low values, while saccades as higher val-

ues. Dispersion-based algorithms, such as ”dispersion

threshold identiﬁcation” (IDT)(Salvucci and Gold-

berg, 2000), consider the distance between observed

points that correspond to the same temporal and spa-

tial information, which is less pronounced in ﬁxa-

tions than in saccades. Based on these core prin-

ciples, some algorithms improve these methods by

adding other thresholds, such as ”velocity and disper-

sion threshold identiﬁcation” (IVDT)(Komogortsev

and Karpov, 2013), or by implementing a second pro-

cessing step, such as ”velocity and movement pat-

tern identiﬁcation” (IVMP)(Komogortsev and Kar-

pov, 2013). Finally, other algorithms rely on machine

learning (Zemblys et al., 2018; Startsev et al., 2019)

to detect gaze events, avoiding the primary limitations

of the aforementioned methods, such as the necessity

to set optimal parameters. These methods require spe-

ciﬁc implementation and execution procedure, in ad-

dition to model training, which is not feasible in every

experimental protocol.

We have implemented an algorithm based on the

I-VDT event detection methods (Llanes-Jurado et al.,

2020; Komogortsev and Karpov, 2013; Duchowski

et al., 2022). This choice is based on its widespread

A Method for Standardizing Eye-Tracking and Behavioral Data in Real and Virtual Environments

217

Table 1: Gaze event detection algorithm examples.

Method Principle Events detected Drawbacks

IDT Dispersion threshold Fixation and saccades Threshold settings & Robustness

IVT Velocity threshold Fixation and saccades Threshold settings & Robustness

IVDT Dispersion and velocity threshold Fixation and saccades (and SP) Threshold settings

IVMP Velocity threshold and movement magnitude Fixation, saccades and smooth pursuit Threshold settings

NH Adaptive velocity threshold Saccades, PSO then ﬁxation Restricted detection

DBSCAN Unsupervised learning clustering Fixation, saccades and SP Parameters settings

Machine Learning Machine Learning algorithms (...) Fixation and saccades (and PSO) Model training

use in the community, its ease of implementation and

its reduced computational complexity in terms of pro-

cessing time and space, allowing real-time execution

if necessary. According to the recommendations from

(Llanes-Jurado et al., 2020; Salvucci and Goldberg,

2000), we used the values 1.3° for dispersion thresh-

old and 0.1 s for time threshold.

For SP events, detection becomes more challeng-

ing because it also depends on user and object mo-

tions. The setup proposed in our workﬂow allows to

detect the motions of these elements independently

from the eye-tracking data. Therefore, the distinction

between ﬁxation and SP depends on whether the ele-

ments are moving or not.

User Activity Analysis. User behavior can be ana-

lyzed by examining their movements in the environ-

ment and their interactions with the observed object.

To do this, it is possible to use position elements and

their derivatives, such as trajectory, speed or velocity,

and identify, levels of effort for example. It is also

interesting to analyze the spatial relationship between

the user and the object, through their distance or rel-

ative position, for example. Finally, quantifying the

interaction, in terms of interaction time or the amount

of movement applied on the object, in the case where

the user can control the position of the observed ob-

ject, constitutes an indicator for comparison between

the different modalities.

The spatial relationship and interaction quantiﬁca-

tion are straightforward information to compute from

the information collected with the tracking data. The

spatial relation part is particularly relevant in the con-

text of the observation task we consider. Behavioral

analysis from motion requires models that are compa-

rable to those used for gaze event analysis, but are less

relevant to the scope of our work and are therefore not

presented in detail here.

3.5.2 Restitution

We consider two families of restitution of the ana-

lyzed data, (i) visual representations, and (ii) metrics

and statistical information. Visual representations are

presented in four categories, (i) semantic visualiza-

tion for ET (examining gaze behavior as a function

of a semantic decomposition of the object), (ii) ge-

ometric visualization for ET (representing results as

a function of the 3D coordinates of gaze on the ob-

served object), (iii) temporal visualization for ET, and

(iv) visualization of user activity. Regarding metrics,

we consider two categories, (i) ET metrics and (ii)

user activity metrics.

Semantic Visualization of Eye-Tracking. For a

more relevant study of gaze behavior in relation to the

observed object, it is common practice to use an ab-

straction of that object, to represent areas of interest

(AOIs) (Blascheck et al., 2017). This abstract descrip-

tion can be deﬁned with respect to the saliency aspect

of the object (Wang et al., 2016; Kim et al., 2010), or

can correspond to regions on the object that are corre-

lated with a visual stimulus (Blascheck et al., 2017).

The partitioning of the object can also represent a sig-

niﬁcant semantic decomposition of the object.

Figure 6 shows a decomposition of the simple

mesh (a) into different AOIs, which are identiﬁed

by different colors (b). This decomposition can be

straightforwardly represented by an abstract version

of the object (c), which allows easy visualization of

gaze behaviors such as ”most viewed parts”. This

type of decomposition makes it possible to analyze

gaze behavior at a higher level of abstraction than the

observed object, and to derive relevant metrics that are

easier to interpret. Furthermore, this type of decom-

position makes it possible to reason about collections

of similar objects, which is precisely the case in the

underlying archaeological study.

Geometric Visualization of Eye-Tracking. The

most used visualization in eye tracking study is the

heat map (Figure 7), or attention maps, which is a vi-

sual representation of ﬁxations distribution on the ob-

served object (Herman et al., 2023; Stellmach et al.,

2010; Sundstedt and Garro, 2022). Each area is rep-

resented by a zone with a color gradient which could

indicate the frequency of gaze on a point. The ren-

dering of this visualization strongly depends on its

parameters, which correspond to the value threshold

GRAPP 2025 - 20th International Conference on Computer Graphics Theory and Applications

218

Figure 6: Examples of a semantic visualization: a teapot

3D model (a), segmented in different AOI due to the mesh

proﬁle (b) and an abstract version of the model to clearly

visualize results on it (c).

for the heat zone to appear, its radius, and the color

scale. In the case of a 3D environment, the cone of

vision is used to produce a distribution of gaze over

the surface of the observed object, taking particular

account of potential occlusion due to the object’s pro-

ﬁle and elements. Accordingly, heatmap processing

uses a Gaussian blur ﬁlter (Stellmach et al., 2010;

Pfeiffer and Memili, 2016) or, more effectively, an

adapted gaze projection (Javerliat et al., 2024) that

treats the gaze ray as a cone, opening the frustum in

depth through the use of a Gaussian distribution. Fi-

nally, some works also propose to generate 3D atten-

tion volumes, using volume-based rendering to repre-

sents the distribution of visual attention in the envi-

ronment (Pfeiffer, 2012).

The other important visualization method is the

gaze plot (Figure 7), which is a graph with ﬁxation

points as nodes and successive saccades as edges.

An interesting representation for ﬁxation points uses

cones to represent user’s position associated to the

concerned movement (Stellmach et al., 2010). The

cone’s apex represents the ﬁxation center position, its

radius is relative to the event duration, its height rep-

resents the distance between the gaze origin (user’s

head) and gaze intersection point, and its orientation

is based on the viewing direction.

Figure 7: Example of a visualization directly represented

on the concerned object, respecting its mesh (left: heatmap,

right: gaze plot).

Temporal Visualization of Eye-Tracking. An-

other aspect addressed by some ET visualization

methods involves considering and representing time.

Most of these methods are deﬁned the context of

2D data, e.g. gaze stripes (Kurzhals et al., 2016)

and space-time cubes (Li et al., 2010; Kurzhals and

Weiskopf, 2013) Extending these visualization meth-

ods to 3D gaze data requires a fourth dimension

(Blascheck et al., 2017). In this case, the most com-

mon solution is to use an animated visualization en-

vironment. For example, the GazEnViz solution al-

lows to explore a recording and view the results over

a speciﬁc period of time. The main risk of this type

of visualization is the cluttering up of the informa-

tion displayed. The visualization method proposed

by (Pathmanathan et al., 2023), which is a 3D ex-

tension of space-time cubes method, implements ﬁl-

tering systems per recorded user to simplify the dis-

played data. This approach could be extended in our

case to a ﬁltering system per modality considered, but

again at the risk of complicating the visualization en-

vironment.

Visualization of Movement. The visualization of

the user’s movements can be based on the represen-

tation of a 3D trajectory as in (Javerliat et al., 2024),

with the possibility to limit it to a certain time pe-

riod or to display several trajectories simultaneously.

It may also be interesting to visualize users through

avatars, with different type of representations, such

as hands to focus on interactions, or feet to focus

on walking in the environment (Reipschl

ager et al.,

2022)). In this case, the visualization is an animation,

as in (Pathmanathan et al., 2023), with the same draw-

backs of data overload as in the case of the temporal

visualization of the ET.

Eye-tracking Metrics. In eye tracking studies, ﬁx-

ations are often used and examined using a variety of

metrics. Fixation duration, number of ﬁxations in a

time period, ﬁxation frequency, deﬁned by the num-

ber of ﬁxations on the observed object divided by the

duration of the time period studied, are all signiﬁ-

A Method for Standardizing Eye-Tracking and Behavioral Data in Real and Virtual Environments

219

cant values. In addition, a description of gaze pat-

terns, such as the ﬁrst ﬁxation, can provide informa-

tion about the user’s attentional focus. Similarly, sac-

cades and SP can be examined by their duration, am-

plitude, or velocity.

When combined with AOIs, gaze events can be

used to calculate a variety of metrics, such as the

amount of time spent on each area of the object, or the

number of visits (which can be deﬁned by the passage

of the gaze point from one AOI to another).

User Activity Metrics. As for gaze events, user ac-

tivity can be transformed into a variety of metrics. We

can consider values related to displacements, such as

distance traveled, velocities and accelerations or sta-

tionary phases, values related to relative positions be-

tween the user and the object, such as average dis-

tance, relative height, or values related to interactions,

such as the number of interactions, the amount of ob-

ject displacement caused by these interactions.

3.5.3 Implementation

For the visualizations, heatmaps and gazeplots are

generated, as presented in the Figure 7, and then for

the semantic representations, ﬁve AOIs are consid-

ered on the pots: the inner part, the top, body and foot

parts, and then the raised parts, which can be handles

or buttons. This decomposition corresponds to a stan-

dard structural decomposition used in archaeology.

To compare the user activity in relation to the ob-

ject, we considered two metrics: (i) the distance be-

tween the user’s head and the center of the object and

(ii) a vertical angle. The latter is deﬁned as the an-

gle between two vectors originating from the center

of the artifact: a vector towards to the user’s head and

a vertical vector pointing upwards. As the vertical

angle value approaches zero, the head is positioned

above the artifact. This allows to measure how much

the user is above the artifact while looking at it, which

could be useful if the inside of the object is relevant to

the observation task. The metrics used to analyze eye

tracking and user’s behavior are presented in the Table

2. Additionally, we also inspect acceptability with a

TAM questionnaire and object analysis grid answers.

4 DISCUSSION

The workﬂow presented here is methodological in the

sense that it is not software-based, but can be seen

as a structured guideline for designing, implementing

and conducting a user study across three modalities to

enable comparison.

Table 2: Gaze-related and behavioral metrics summary.

Subject Metric Requirement

Fixation Duration -

Dispersion -

Frequency Ofﬂine analysis

First ﬁxation Duration -

Dispersion -

Area of interest Associated ﬁxation Fixations description

Order in scan path Ofﬂine analysis

Activity Head distance Elements position

Vertical angle Elements position

The workﬂow has been designed to be generic so

that it can be applied to different eye-tracking and

behavioral studies conducted in virtual and real en-

vironments. In this way, it is possible to compare

the performance of the same task in different modal-

ities, making it possible to identify differences dur-

ing a simulation on user behavior, such as cognitive

workload, or to identify biases in behavior between

modalities. In addition, this methodology can also be

used to validate the suitability of VR tools against the

same use case in a real context. Another original as-

pect of the workﬂow is to enable a behavioral compar-

ison between the VR HMD and the Immersive room

modality, which is very rarely addressed in the exist-

ing literature.

We have illustrated the implementation of this

workﬂow with an archaeological use case of an ar-

tifact observation task, detailed in (Dumonteil et al.,

2024). In this case, the aim of this on-going study is

to detect possible biases in the performance of archae-

ological tasks in virtual reality. The same approach

can also be applied to other application domains.

Nevertheless, this methodology still has remain-

ing challenges depending on the implementation

choices. First of all, the use of disparate ET sys-

tems for the different modalities, in spite of data post-

processing, may lead to differences in eye-tracking re-

sults only due to the modalities distinction, in terms

of data quality or gaze estimation accuracy. This is

speciﬁcally signiﬁcant in immersive rooms because

we believe that a better and more comfortable device

for eye-tracking in this environment could improve

the comfort and the user’s behavior, according to the

potential discomfort of the set up with two superposed

glasses. Subsequently , the sampling rate of the sys-

tems used for each modality can vary considerably

(from 50 Hz to 200+ Hz, depending on the devices’

parameters). So it is necessary to adapt all data logs

to a uniform rate in order to ensure the consistency of

the data across all modalities.

As mentioned above, the restitution part of the

workﬂow could be enriched with a multitude of ad-

ditional measurements, either for metrics production,

or for gaze events detection. In this case, the impor-

GRAPP 2025 - 20th International Conference on Computer Graphics Theory and Applications

220

tant point is to use the same algorithm in the different

modalities. For this reason, the data acquisition and

processing part of the workﬂow has also been treated

with particular care.

5 CONCLUSION

This paper presents a workﬂow design for the genera-

tion of standardized and comparable eye-tracking and

behavioral data, including directions and suggestions

for analyzing and presenting the results using metrics

and visualizations adapted to a 3D context. While

most of the used tools and methods to analyze eye-

tracking data are based on a 2D context, generalizing

their use to a 3D context enables a more complete un-

derstanding of user behavior, despite some challenges

to correctly process and display such data. The goal

of this methodology is to compare use cases in real

and virtual environments, addressing the challenges

of comparing data collected using different devices

and in different environments. The approach is spe-

cially focused on visual exploration tasks on a single

object, in real and virtual modalities. The approach

was illustrated with the implementation of an archae-

ological use case.

Further works will involve the evaluation of our

method in different contexts, and on the other hand,

measure the effectiveness of the proposed implemen-

tation over other methods of eye-tracking. Our work-

ﬂow could also be applied to other application do-

mains and extended to tasks beyond object observa-

tion.

ACKNOWLEDGEMENTS

This work was partially funded by Equipex+ Con-

tinuum ANR-21-ESRE-0030 and CNRS MITI AAP

Digital Twin AJAX.

REFERENCES

Blascheck, T., Kurzhals, K., Raschke, M., Burch, M.,

Weiskopf, D., and Ertl, T. (2017). Visualization of

Eye Tracking Data: A Taxonomy and Survey. Com-

puter Graphics Forum, 36(8):260–284.

Brookes, J., Warburton, M., Alghadier, M., Mon-Williams,

M., and Mushtaq, F. (2020). Studying human behav-

ior with virtual reality: The Unity Experiment Frame-

work. Behavior Research Methods, 52(2):455–463.

Cauliez, J., Delaunay, G., and Duplan, V. (2002). Nomen-

clature et m

ethode de description pour l’

etude des

eramiques de la ﬁn du N

eolithique en Provence.

ehistoires M

editerran

eennes, (10-11).

Duchowski, A. T., Krejtz, K., Volonte, M., Hughes, C. J.,

Brescia-Zapata, M., and Orero, P. (2022). 3d gaze in

virtual reality: vergence, calibration, event detection.

Procedia Computer Science, 207:1641–1648.

Dumonteil, M., Gouranton, V., Mac

e, M. J., Nicolas, T.,

and Gaugne, R. (2024). Do we study an archaeolog-

ical artifact differently in vr and in reality? In ICAT-

EGVE 2024-joint 34th International Conference on

Artiﬁcial Reality and Telexistence & the 29th Euro-

graphics Symposium on Virtual Environments, pages

1–9.

Garrido-Jurado, S., Mu

noz-Salinas, R., Madrid-Cuevas,

F. J., and Mar

ın-Jim

enez, M. J. (2014). Auto-

matic generation and detection of highly reliable ﬁdu-

cial markers under occlusion. Pattern Recognition,

47(6):2280–2292.

Gulhan, D., Durant, S., and Zanker, J. M. (2021). Simi-

larity of gaze patterns across physical and virtual ver-

sions of an installation artwork. Scientiﬁc Reports,

11(1):18913.

Herman, L., Popelka, S., and Hejlova, V. (2023).

Eye-tracking Analysis of Interactive 3D Geovisu-

alization. Journal of Eye Movement Research,

10(3):10.16910/jemr.10.3.2.

Javerliat, C., Villenave, S., Raimbaud, P., and Lavou

e, G.

(2024). PLUME: Record, Replay, Analyze and Share

User Behavior in 6DoF XR Experiences. IEEE TVCG,

30(5):2087–2097.

Jogeshwar, A. K. and Pelz, J. B. (2021). GazeEnViz4D: 4-

D Gaze-in-Environment Visualization Pipeline. Pro-

cedia Computer Science, 192:2952–2961.

Kim, Y., Varshney, A., Jacobs, D. W., and Guimbreti

ere, F.

(2010). Mesh saliency and human eye ﬁxations. ACM

Transactions on Applied Perception, 7(2):12:1–12:13.

Kollenberg, T., Neumann, A., Schneider, D., Tews, T.-K.,

Hermann, T., Ritter, H., Dierker, A., and Koesling, H.

(2010). Visual search in the (un)real world: how head-

mounted displays affect eye movements, head move-

ments and target detection. In Proc. of ACM Symp. on

ETRA, pages 121–124.

Komogortsev, O. V. and Karpov, A. (2013). Automated

classiﬁcation and scoring of smooth pursuit eye move-

ments in the presence of ﬁxations and saccades. Be-

havior Research Methods, 45(1):203–215.

Kurzhals, K., Hlawatsch, M., Heimerl, F., Burch, M., Ertl,

T., and Weiskopf, D. (2016). Gaze Stripes: Image-

Based Visualization of Eye Tracking Data. IEEE

TVCG, 22(1):1005–1014.

Kurzhals, K. and Weiskopf, D. (2013). Space-Time Visual

Analytics of Eye-Tracking Data for Dynamic Stim-

uli. IEEE Transactions on Visualization and Com-

puter Graphics, 19(12):2129–2138. Number: 12.

Li, T. (2021). 3d representation of eyetracking data: An im-

plementation in automotive perceived quality analysis

(dissertation). Digitala Vetenskapliga Arkivet.

Li, T.-H., Suzuki, H., and Ohtake, Y. (2020). Visualization

of user’s attention on objects in 3D environment using

A Method for Standardizing Eye-Tracking and Behavioral Data in Real and Virtual Environments

221

only eye tracking glasses. Journal of Computational

Design and Engineering, 7(2):228–237. Number: 2.

Li, X., C¸

oltekin, A., and Kraak, M.-J. (2010). Visual Explo-

ration of Eye Movement Data Using the Space-Time-

Cube. In Fabrikant, S. I., Reichenbacher, T., van Krev-

eld, M., and Schlieder, C., editors, Geographic Infor-

mation Science, pages 295–309. Springer.

Liu, M., Li, Y., and Liu, H. (2020). 3D Gaze Estimation

for Head-Mounted Eye Tracking System With Auto-

Calibration Method. IEEE Access, 8:104207–104215.

Llanes-Jurado, J., Mar

ın-Morales, J., Guixeres, J., and

Alca

niz, M. (2020). Development and Calibration of

an Eye-Tracking Fixation Identiﬁcation Algorithm for

Immersive Virtual Reality. Sensors, 20(17):4956.

Mizuchi, Y. and Inamura, T. (2018). Evaluation of Human

Behavior Difference with Restricted Field of View in

Real and VR Environments. In IEEE Int. Symp. on

Robot and Human Interactive Communication (RO-

MAN), pages 196–201.

Onkhar, V., Dodou, D., and de Winter, J. C. F. (2023). Eval-

uating the Tobii Pro Glasses 2 and 3 in static and dy-

namic conditions. Behavior Research Methods.

Paletta, L., Santner, K., Fritz, G., Mayer, H., and Schram-

mel, J. (2013). 3D attention: measurement of visual

saliency using eye tracking glasses. In ACM CHI EA

’13 Extended Abstracts on Human Factors in Comput-

ing Systems, pages 199–204.

Pathmanathan, N.,

Oney, S., Becher, M., Sedlmair, M.,

Weiskopf, D., and Kurzhals, K. (2023). Been There,

Seen That: Visualization of Movement and 3D Eye

Tracking Data from Real-World Environments. Com-

puter Graphics Forum, 42(3):385–396.

Pfeiffer, T. (2012). Measuring and visualizing attention in

space with 3D attention volumes. In Proc. of the ACM

Symp. on ETRA, pages 29–36.

Pfeiffer, T. and Memili, C. (2016). Model-based real-time

visualization of realistic three-dimensional heat maps

for mobile eye tracking and eye tracking in virtual re-

ality. In Proc. of the ACM Symp. on ETRA, pages 95–

102.

Pfeiffer, T., Renner, P., and Pfeiffer-Leßmann, N. (2016).

EyeSee3D 2.0: model-based real-time analysis of

mobile eye-tracking in static and dynamic three-

dimensional scenes. In Proc. of the ACM Symp. on

ETRA, pages 189–196.

Pfeil, K., Taranta, E. M., Kulshreshth, A., Wisniewski, P.,

and LaViola, J. J. (2018). A comparison of eye-head

coordination between virtual and physical realities. In

Proc. ACM Symp. on Applied Perception, pages 1–7.

Pomplun, M. and Sunkara, S. (2019). Pupil dilation as an

indicator of cognitive workload in human-computer

interaction. In Human-Centered Computing, pages

542–546. CRC Press.

Reipschl

ager, P., Brudy, F., Dachselt, R., Matejka, J., Fitz-

maurice, G., and Anderson, F. (2022). AvatAR: An

Immersive Analysis Environment for Human Motion

Data Combining Interactive 3D Avatars and Trajecto-

ries. In Proc. of ACM CHI, pages 1–15.

Salvucci, D. and Goldberg, J. (2000). Identifying ﬁxations

and saccades in eye-tracking protocols. In Proc. of

ACM Symp. ETRA, pages 71–78.

Startsev, M., Agtzidis, I., and Dorr, M. (2019). 1D CNN

with BLSTM for automated classiﬁcation of ﬁxations,

saccades, and smooth pursuits. Behavior Research

Methods, 51(2):556–572.

Stellmach, S., Nacke, L., and Dachselt, R. (2010). Ad-

vanced gaze visualizations for three-dimensional vir-

tual environments. In Proc. of ACM Symp. ETRA,

pages 109–112.

Sundstedt, V. and Garro, V. (2022). A Systematic Review of

Visualization Techniques and Analysis Tools for Eye-

Tracking in 3D Environments. Frontiers in Neuroer-

gonomics, 3.

Takahashi, N., Inamura, T., Mizuchi, Y., and Choi, Y.

(2021). Evaluation of the difference of human behav-

ior between vr and real environments in searching and

manipulating objects in a domestic environment. In

Proc. IEEE Int. Conf. on Robot & Human Interactive

Communication (RO-MAN), pages 454–460.

Takahashi, R., Suzuki, H., Chew, J. Y., Ohtake, Y., Na-

gai, Y., and Ohtomi, K. (2018). A system for three-

dimensional gaze ﬁxation analysis using eye tracking

glasses. Journal of Computational Design and Engi-

neering, 5(4):449–457.

Ugwitz, P., Kvarda, O., Ju

ıkov

a, Z.,

sinka,

C., and

Tamm, S. (2022). Eye-Tracking in Interactive Virtual

Environments: Implementation and Evaluation. Ap-

plied Sciences, 12(3):1027.

Villenave, S., Cabezas, J., Baert, P., Dupont, F., and Lavou

G. (2022). XREcho: a unity plug-in to record and

visualize user behavior during XR sessions. In Proc.

of ACM Multimedia Systems Conf., pages 341–346.

Wang, X., Lindlbauer, D., Lessig, C., and Alexa, M. (2017).

Accuracy of Monocular Gaze Tracking on 3D Geom-

etry. In Eye Tracking and Visualization, pages 169–

184. Springer: Mathematics and Visualization.

Wang, X., Lindlbauer, D., Lessig, C., Maertens, M., and

Alexa, M. (2016). Measuring the Visual Salience of

3D Printed Objects. IEEE Computer Graphics and

Applications, 36(4):46–55.

Willemsen, P., Colton, M. B., Creem-Regehr, S. H., and

Thompson, W. B. (2009). The effects of head-

mounted display mechanical properties and ﬁeld of

view on distance judgments in virtual environments.

ACM Transactions on Applied Perception, 6(2):8:1–

8:14.

Zemblys, R., Niehorster, D. C., Komogortsev, O., and

Holmqvist, K. (2018). Using machine learning to de-

tect events in eye-tracking data. Behavior Research

Methods, 50(1):160–181.

Zhang, L., Wade, J., Swanson, A., Weitlauf, A., Warren, Z.,

and Sarkar, N. (2015). Cognitive state measurement

from eye gaze analysis in an intelligent virtual real-

ity driving system for autism intervention. In 2015

international conference on affective computing and

intelligent interaction (ACII), pages 532–538. IEEE.

GRAPP 2025 - 20th International Conference on Computer Graphics Theory and Applications

222