Traffic Abnormality Detection through Directional Motion Behavior Map

Abstract

Automatic traffic abnormality detection through visualsurveillance is one of the critical requirements for IntelligentTransportation Systems (ITS). In this paper, wepresent a novel algorithm to detect abnormal traffic eventsin crowded scenes. Our algorithm can be deployed with fewsetup steps to automatically monitor traffic status. Differentfrom other approaches, we don’t need to define region ofinterests (ROI) or tripwires nor to configure object detectionand tracking parameters. A novel object behavior descriptor- directional motion behavior descriptors are proposed.The directional motion behavior descriptors collectforeground objects’ direction and speed information from avideo sequence with normal traffic events, and then thesedescriptors are accumulated to generate a directional motionbehavior map which models the normal traffic status.During detection steps, we first extract the directional motionbehavior map from the newly observed video and thenmeasure the differences between the normal behavior mapand the new map. If new direction motion behaviors arevery different from the descriptors in the normal behaviormap, then the corresponding regions in the observed videocontain traffic abnormalities. Our proposed algorithm hasbeen tested using both synthesized and real surveillancevideos. Experimental results demonstrated that our algorithmis effective and efficient for practical real-time trafficsurveillance applications.