Head-Hand Tracking Based Driving Behavior Analysis in Computer Vision System
Peiqun Lin, Jianmin Xu, Sibo Huang · 2011
In order to improve traffic safety and reduce traffic accidents, a real-time computer vision system is designed to identify unusual driving behaviours based on head-hand tracking. Firstly, an ellipse fitting algorithm is used to locate the steering wheel and the driver's head in every frame of image, then the skin color model is employed to extract the driver's hand regions, and the skeletons of the hand regions can be obtained by a chamfer distance transform algorithm. Moreover, Kalman filter is put forward to track the driver's head and hands. Finally, a naive Bayesian classifier, whose input parameters include position, movement direction and movement speed, is used to recognize several kinds of unusual driving behaviours. An experiment with 5257 frames of image from real environments is carried out, and its results demonstrate the effectiveness of the proposed methods.