CogMamba: Multi-Task Driver Cognitive Load and Physiological Non-Contact Estimation with Multimodal Facial Features.

Area of Science:

• Human-Computer Interaction
• Automotive Safety
• Biomedical Engineering

Background:

• Driver cognitive load is critical for advanced driving assistant systems (ADAS) and autonomous driving safety.
• Current methods for detecting driver cognitive load are often invasive or limited, hindering practical application.
• Non-driving-related tasks (NDRTs) can increase cognitive load, requiring rapid driver re-engagement.

Purpose of the Study:

• To develop a non-contact, multi-task model for estimating driver cognitive load and physiological state.
• To address the limitations of existing invasive or eye-tracking-based cognitive load detection methods.
• To enhance the safety and practicality of ADAS and autonomous driving systems.

Main Methods:

• Proposed a novel non-contact cognitive load and physiological state estimation model named CogMamba.
• Utilized multimodal features extracted from RGB facial videos.
• Introduced the Mamba architecture to capture temporal dependencies for joint estimation of cognitive load, heart rate (HR), and respiratory rate (RR).

Main Results:

• CogMamba demonstrated superior performance on two public datasets.
• The model showed excellent robustness in cross-dataset generalization tests.
• Successfully achieved joint estimation of cognitive load, HR, and RR using non-contact video data.

Conclusions:

• CogMamba offers an efficient and practical solution for non-contact driver state monitoring.
• The findings provide valuable insights for real-world applications in autonomous driving.
• This non-contact approach can improve driver safety and system usability in future vehicles.