Detection of abnormal moving vehicles for intelligent driver assistance system

Driving in nighttime environment is a difficult work due to low visibility, drowsy and impaired drivers. It is crucial to make a good algorithm to detect automatically abnormal moving vehicles surrounding the host vehicle to indicate early collision warning. This paper proposes an efficient approach to detect on-road abnormal moving vehicles in nighttime. Oncoming, change speed, change lane, roadside parking, and overtaking vehicles are detected. The proposal method is useful for vehicle behavior analysis system of IDAS (Intelligent Driver Assistance System). Firstly, all motion vectors of moving objects are estimated from frames. Most of safety moving vehicles are eliminated by using a new proposal threshold range and ROI setting. The remaining motion vectors are grouped by using K-means clustering algorithm to obtain segment abnormal vehicle candidates. The segmented candidates are classified using learning algorithm Support Vector Machines (SVMs) and various features to eliminate non-vehicle candidates. The experimental results show that the proposal method is high ability to detect the abnormal moving vehicles in front of the host vehicle in nighttime driving. It is useful to be the first processing step for behavior analysis systems in Intelligent Driver Assistance Systems.