Abnormal motion detection in an occlusion environment

We present a motion classification approach to detect movements of interest (abnormal motion) based on optical flow. By tracking all feature points of a moving human in successive frames, we calculate the coordinate space and create feature space. This is done directly from the intensity information without explicitly computing the underlying motions. It requires no foreground segmentation, no prior learning of activities, no motion recognition and no object detection. First, we determine the abnormal scene and speed by using the velocity histogram. Then by using k-means clustering over velocity orientation and magnitude, we determine the abnormal direction. The performance of the proposed method is experimentally shown.SICE Annual Conference 2013 - International conference on Instrumentation, Control, Information Technology and System Integration September 14-17, 2013, Nagoya University, Nagoya, Japa