In this paper, an intelligent-video-analysis-based vehicle abnormal behavior detection method was presented to handle the real-time problem in vehicle abnormal behavior detection. When vehicle abnormal behavior occurs, vehicle position, velocity and moving direction change rapidly. To extract the changes of the three parameters mentioned above, the background subtraction approach was adapted to detect vehicles. Furthermore the meanshift algorithm was utilized to track the detected vehicles. Vehicle behavior decision can be concluded by weight fusion of the three parameters. To verify the proposed method, experiments on real videos were operated. Experimental results demonstrate that the proposed method can detect vehicle abnormal behavior effectively in real traffic scene.