Driver gaze tracking and eyes off the road detection

Abstract

Driver distraction is the main cause of vehicle collisions rather than mechanical faults. This paper proposes a vision-based continuous driver gaze tracking and eyes off the road detection system. The main blocks of the system are Facial points and head pose tracking, Gaze prediction, and Eyes off the road detection. The IR cut filter camera is fitted on the dashboard in line with the steering wheel which continuously records the driver’s facial region. This feed is processed to track facial features and estimate head pose. Subsequent deep learning algorithm predicts the gaze region using features like yaw, pitch, roll, number of pixels in the facial area, and eye distance from the top edge. The IR cut filter camera can produce good quality images for both day and night conditions. The system detects the eyes off the road in real-time with 20 fps and 96% accuracy. The system can also perform in diver-independent calibration mode with 85% accuracy.