Research On The Algorithm Of Posture Recognition In Diving Competition Video

With the rapid development of computer and multimedia technology, video image analysis has become one of important research fields. It has a wide range of applications in military, transportation and medical areas, etc. and promotes the developments of artificial intelligence, pattern recognition and other research areas to a higher and deeper level. As a challenging subject in the field of computer vision, visual analysis of human behaviors, including people detection, people tracking, motion recognition, behavior understanding and description, involves many subjects such as image processing, pattern recognition and so on. Nowadays, visual analysis of human behaviors is being applied into sports science to recognize and track athlete’s posture, thus helpful for their scientific training and technology diagnosis.The research aims at classification and recognition of human postures in dynamic scene by way of athlete segmentation and analysis of their postures in diving competition videos. The results of research can be applied in the automatic scoring, motion analysis, sports classification and many other areas. The basic theories and methods related to human motion recognition are introduced and the characteristics and difficulties of sports postures in diving competitions are analyzed. Moreover, the method of moving object segment with complex background and global motion has been presented as well as an effective method to classify and recognize of athletes’postures by multiple feature fusion. The research consists the two parts of object segmentation and posture recognition. In the stage of object segmentation, firstly, Lucas_Kanade optical flow algorithm is adopted to estimate the global motion area and object motion area. Then the color distribution characteristics in the YCbCr space is used to detect skin regions in the object motion area to determine the accurate position and shape of the object. After noise is eliminated by the projection method, the object is segmented accurately. In the stage of posture recognition, color, aspect ratio, area proportion and SIFT features are extracted after the segmented objects. Then support vector machine (SVM) is used to recognize every kind of diving posture. The experimental results show that our method is more effective in the object segmentation and posture recognition in diving competition videos compared with other methods.