Design and implementation of a dangerous driving behavior analysis system

Abstract

Dangerous driving behavior is the major factor to cause most of driving accidents. We present a novel method for dangerous driving behavior analysis. When the dangerous driving is detected, the system will issue an alert message to remind driver to pay more attention. The proposed system can detect three kinds of dangerous behavior: dozing, eating, and phoning. Furthermore, since we propose a novel algorithm to adaptively detect skin color, the proposed system can run in various light conditions with good quality. The proposed algorithm is verified and implemented on an in-vehicle embedded system -TREK-668. The performance on PC can reach 150 fps in average on input video with QVGA (320×240) resolution. On embedded systems, the proposed system can achieve 24fps in average on input video with QVGA (320×240) resolution.I.