Anomaly Behavior Detection Based On Global Motion Orientation

With the continuous development of science and technology, intelligent video surveillance system is aroused more and more people's attention, and abnormal behavior detection system is an important aspect of it. It plays important role in public safety and intelligent monitoring in hospital wards. This paper defines three normal behaviors-a single person walking, multiplayer walking and shaking hands, and two anomaly behaviors-fainted and fighting. According to different motion direction rules for different events, we extracted direction of movement as the feature. Then, abnormal events are detected with support vector machine (SVM).First, the silhouette is extracted via the temporal difference. To reduce the noise impact, the silhouette is processed via dilation, erosion, and smoothing. It used motion image sequence for motion history image, then we took the gradient of the motion history image, we would get motion direction of human object in every frames.In order to get more detailed information, we introduce block-based motion directions to model those events, and use LIBSVM to learn and detect the abnormality actions.Experimental result shows that this method proposed in this paper is of low computation complexity and can be used in real-time surveillance system.