Research On Key Techniques Of Video Analysis System For Diving Sport

Recently, the motion and video computing analysis has become a hot topic in the field of computer vision.According to characteristics of diving sports video, the thesis first concentrates on how to extract the moving objects from video sequences in dynamic scene. Furthermore, the proposed moving object extraction method is applied to analysis of the diving Sports video. Based on these work, a Diving sports Video Analysis System, Diving VAS for short, is developed. The main contributions are listed as follows:We develop an automatic algorithm which is used to extract moving objects in video sequences of a dynamic scene. The proposed algorithm first adopts the globe camera motion models to obtain the corresponding coordinates of pixels in neighbor frames and then the difference between them are calculated to get the segmented foreground and the background. The pixels that belong to the background in neighbor frames are then modeled by Gaussian models. At last, the particle filtering method is applied to predict the possible foreground region in the coming frame and calculate probability for all possible coming pixels to segment the coming frames. Experimental results from our algorithm show that the accumulating error caused by inaccurate estimating parameter of globe camera motion models can be effectively reduced.We also develop a Diving Sports Video Analysis System (Diving VAS) based upon some key techniques such as video segmentation, 3D human motion tracing, motion parameters calculation, inverse computing model parameters and so on. Diving VAS can provide general supports for application of video motion analysis and especially provide scientific assistances for sports training.