Analyzing Gaze Data During Rest Time/Driving Simulator Operation Using Machine Learning

Fujikake, Kazuhiro; Itadu, Yoshiyuki; Takada, Hiroki · 2022 · Crossref

DOI: 10.1007/978-3-031-05028-2_28

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of objectively evaluating visually induced motion sickness (VIMS) in driving simulator (DS) experiments, particularly among elderly drivers. While DSs are valuable for research, they often induce VIMS, a condition caused by sensory mismatch between visual and vestibular systems. Existing evaluation methods, such as subjective questionnaires or physiological measurements requiring electrodes, are either non-objective or burdensome to participants. The authors propose a less intrusive method using non-contact eye-tracking data to detect VIMS, hypothesizing that machine learning models can identify gaze characteristics, such as rotational eye movements, associated with sickness. The experiment involved eight elderly participants with no visual or balance impairments. Participants underwent six DS trials: one one-minute practice run and five five-minute test runs involving stops and turns. Gaze data were collected using a Tobii Pro X2–30 eye-tracker at rest (before and after trials) and during driving. Participants completed the Simulator Sickness Questionnaire (SSQ) to subjectively report symptoms. Based on SSQ scores, participants were divided into two groups: four who experienced VIMS and four who did not. The researchers applied a one-dimensional convolutional neural network to classify gaze data into VIMS-positive or VIMS-negative categories. The model was trained on 60-second resting periods and 30-second segments of driving trials, using a leave-one-out validation method. Results indicated that the machine learning model achieved higher accuracy when analyzing gaze data after DS trials compared to before. For resting gaze data post-trial, the model’s mean accuracy, precision, recall, and F-score all exceeded 70%. Statistical analysis showed that model accuracy was significantly higher after driving than before (p < 0.1). For driving gaze data, the model performed better on the fifth trial than the first, with mean accuracy exceeding 65% and recall reaching 87.60% in the fifth run. Notably, for the VIMS group, the accuracy of the model using resting gaze data (before and after driving) was significantly higher than when using driving gaze data (p < 0.05). Visual analysis of gaze plots showed that participants with VIMS exhibited more diffuse gaze patterns after driving compared to those without VIMS. The study concludes that machine learning applied to non-contact gaze data is an effective, objective index for evaluating VIMS, particularly when analyzing resting periods after simulator use. The findings support the hypothesis that gaze characteristics change in response to VIMS. However, the lower accuracy during active driving suggests that task-related gaze behaviors complicate detection. The authors recommend increasing dataset sizes and further validating the system in complex environments, such as public roads, to improve real-time detection capabilities. This approach offers a promising, low-burden alternative to traditional VIMS assessment methods.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-25; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• gaze based attention detection
• drowsiness detection algorithms
• eye movements scanning
• dms validation

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: tool software, validation psychometrics
• Theoretical Contribution: computational model