Fatigue warning system for driver nodding off using depth image from Kinect

One of the major road traffic accidents is from an exhausted driver such as drowsiness and a lack of attention over driving. This paper presents a driver fatigue warning system using Kinect depth image, which monitors a driver attention and alerts driver while nodding off. Our algorithm transforms Kinect depth images into gradient images to detect driver face and applies Discriminative Random Regression Forests to get angles of head rotation. The sensitivity result is 93.75% by measuring with a dataset that consists of 160 images of four head positions of 40 people. The system can handle many situations both in daytime and night time. For testing one passenger in real situations, the sensitivity results are 94.28% with 2,676 frames at daytime and 95.13% with 2,036 frames at night time.