Advancing Cognitive Load Detection in Simulated Driving Scenarios Through Deep Learning and fNIRS Data

Area of Science:

• Neuroscience
• Human-Computer Interaction
• Automotive Engineering

Background:

• The transition to automated driving systems presents new safety challenges, including increased crash risks from human factors like cognitive overload.
• Driving simulators offer a controlled environment to study these complex human factors.
• Advanced Driving Assistance Systems (ADASs) are key to this transition.

Purpose of the Study:

• To dynamically assess cognitive load in a realistic driving simulator using Functional Near-Infrared Spectroscopy (fNIRS).
• To evaluate the effectiveness of deep learning models in analyzing fNIRS data for cognitive load monitoring.
• To investigate the impact of temporal segmentation strategies on classification performance.

Main Methods:

• Utilized a realistic driving simulator with a challenging night-time-rain scenario.
• Employed Functional Near-Infrared Spectroscopy (fNIRS) to measure brain activity.
• Participants engaged in an auditory n-back task to simulate multitasking demands.
• Applied a sliding window approach to time-series fNIRS data and analyzed it using the EEGNet deep learning model.
• Compared overlapping and non-overlapping temporal segmentation strategies.

Main Results:

• Classification performance was significantly influenced by learning rate and windowing method.
• A learning rate of 0.001 achieved the highest accuracy: 100% with overlapping windows and 97% with non-overlapping windows.
• The study demonstrated the feasibility of real-time cognitive load monitoring.

Conclusions:

• Combining fNIRS and deep learning shows significant potential for real-time cognitive load monitoring in simulated driving.
• Optimizing temporal modeling, including windowing methods and learning rates, is crucial for accurate physiological signal analysis.
• This approach can inform the design of safer automated driving systems by addressing human factors.