Recognition Of Forehand And Backhand Stroke Action In Tennis Video

Tennis game attracts widespread attention due to a large audience and great prospect, and the research on tennis video is very popular in the field of video analysis. Users are concerned about the stroke event, so this dissertation focuses on the recognition of forehand and backhand stroke action in tennis video.First of all, summarize the previous approach of moving object detection. Considering the existing problems in background modeling method, this dissertation studied a two-layer background model method based on codebook and running average, then moving object is detected using the novel approach. Experimental results show that, the background modeling approach has low computational complexity, fast background update characteristic and strong anti-noise ability, what's more, it can also adapt to the light change. It is available for moving object detection in other complex environment.Secondly, process the foreground image using morphology method and extract the edge with different edge detector. The dissertation analyzes the advantages and disadvantages of Sobel operator, Laplacian operator and Canny operator by contrast experiment, finally we choose Canny operator.Thirdly, a 2D model is built based on player's edge, it's necessary to fix positions on every part of the 2D model.Finally, the forehand or backhand stroke action is recognized. Detect the sound of stroke based on audio keywords to locate swing sequence in video sequences. In every frame we use the position relation of 2D model's parts to judge the swing direction, according to it we get comprehensive identification results. In the end, the corresponding relationship between swing direction and stroke action is constructed through the domain knowledge.