Vision-based vehicle behavior monitoring method using a novel clustering algorithm

Abstract

Vehicle behavior analysis is an important task in the area of intelligent transportation systems. The purpose of this study is to monitor vehicle real-time risk and behavior using soft computing techniques. It is inspired by the fact that feature points of the same vehicle have the same motion status in the 3D space. This paper proposed an effective vehicle motion segmentation method using rigid motion constraints. Specifically, we first recovered the 3D information of the feature points using the height enumeration and camera affine matrix. Then, the rigid motion constraints of two feature point trajectories in two directions were computed to analyze the characteristic. Thereafter, an affinity matrix was built by using Gaussian kernel, where the larger value in the affinity matrix means that the feature point trajectories are more likely belonging to the same vehicle. Finally, a segment and merge procedure is adopted to cluster the feature point trajectories. Experimental results on the traffic videos and the Hopkins 155 datasets demonstrate that the proposed method achieves a good performance. This paper also gives a new idea to the field of intelligent transportation system and application.