Recognition of visual-related non-driving activities using a dual-camera monitoring system

For a Level 3 automated vehicle, according to the SAE International Automation Levels definition (J3016), identification of non-driving activities (NDAs) that driver is engaging with great importance in design an intelligent take-over interface. Much existing literature focuses on strategy associated Human-Machine Interaction design. This paper proposes dual-camera based framework identify and track NDAs require visual attention. achieved by mapping driver's gaze using nonlinear system approach, object scene, recognised deep learning algorithm. A novel gaze-based region interest (ROI) selection module introduced contributes about 30% improvement average success rate 60% reduction processing time compared results without this module. has been successfully demonstrated five types NDA required attention 86.18%. The outcome research could be applicable other tracking within certain window potentially used evaluate level for both human-driving vehicles.