Research On Character Target Detection And Motion Recognition Methods In Video

Characters of target detection and gesture recognition is an important application of computer vision field of understanding, with the development of intelligent community,people pay high attention on vigilance and attention of safe living, the subject attracted more researchers gradually. Because of the complexity of the background, the complexity of the action and the complexity of the process, this is also a very challenging task. The main purpose of this paper is to achieve target detection under the premise of simple background and to achieve six kinds of action recognition,they are walking, hand waving,jogging, boxing, hand clapping and running.For target detection, this paper studies three kinds of commonly used methods,they are background subtraction method, inter frame difference method and optical flow method, we selected background subtraction method as the final target detection method combined with shoot video background and the advantages and disadvantages of the three methods. For the shortage of mixed Gaussian model method in background subtraction method,we proposed a video images based on feature point sparse optical flow field in the background modeling method. Experimental results show that this method can effectively remove the problem of shadow in the background modeling, which can significantly improve the accuracy of detection.After the foreground segmentation, the feature extraction of the target area is carried out and the classification and recognition of the action are carried out. This paper focuses on the research method based on space-time interest points feature, using the method of feature extraction feature description, then use the bag of words model for behavior modeling, finally using KNN algorithm for model training and final action recognition test. The experiment on the KTH action database of experimental tests, the results show that, the methods proposed in this paper can extracting the ideal feature vector and can get good recognition effect, preliminary validation of the effectiveness of the system and provide theoretical foundation and experimental basis for practical application.