Calculated Typical Behavior Recognition Algorithm Based On Optical Flow

Human motion vision analysis is a new research direction. It includes many science domains, such as: Pattern Classification, Image Processing, Computer Vision and Artificial intelligence, etc. Its goal is to detect, track, identify the people, then understand and describe their actions through video images. These research results can be widely applied to the smart surveillance, virtual reality, and perceptual interface, and have very good social and economic benefits.In general, the human motion vision analysis includes the follow four steps: motion detection, moving object classification, people tracking, motion understanding and description. The thesis emphasizes on high-level vision module (human action understanding and description). The main works are as follows:1 , Compare various features in video series, and then analyze the main current optical-flow algorithms and its experimental results, finally analyze the experimental results of different algorithms. When the translation between consecutive frames is large, the hierarchical optical-flow algorithm can get optical-flow field more precisely.2, Propose an action recognition algorithm based on optical-flow feature and Sequence Alignment. This algorithm uses the direction histogram from optical-flow field to generate the common spatiotemporal template and the index sequences warehouse, and achieves action recognition via sequence alignment. Because of reducing the number of template by using common template warehouse, the algorithm can handle 320x240 image sequence and the average processing speed will be up to 10 frames per second.3^ An action classification software system is developed. The results through extensive experiment show that the system can make on-line typical action recognition, and is robust to the object's size change, certain incline degree and revolve.