Functional near-infrared spectroscopy (fNIRS) and Eye tracking for Cognitive Load classification in a Driving Simulator Using Deep Learning

Mehshan Ahmed Khan,Houshyar Asadi,Mohammad Reza Chalak Qazani,Chee Peng Lim,Saied Nahavandi

DOI: https://doi.org/10.48550/arxiv.2408.06349

IF: 6.4588

2024-07-24

Human-Computer Interaction

Abstract:Motion simulators allow researchers to safely investigate the interaction of drivers with a vehicle. However, many studies that use driving simulator data to predict cognitive load only employ two levels of workload, leaving a gap in research on employing deep learning methodologies to analyze cognitive load, especially in challenging low-light conditions. Often, studies overlook or solely focus on scenarios in bright daylight. To address this gap and understand the correlation between performance and cognitive load, this study employs functional near-infrared spectroscopy (fNIRS) and eye-tracking data, including fixation duration and gaze direction, during simulated driving tasks in low visibility conditions, inducing various mental workloads. The first stage involves the statistical estimation of useful features from fNIRS and eye-tracking data. ANOVA will be applied to the signals to identify significant channels from fNIRS signals. Optimal features from fNIRS, eye-tracking and vehicle dynamics are then combined in one chunk as input to the CNN and LSTM model to predict workload variations. The proposed CNN-LSTM model achieved 99% accuracy with neurological data and 89% with vehicle dynamics to predict cognitive load, indicating potential for real-time assessment of driver mental state and guide designers for the development of safe adaptive systems.