Research On Clustering And Retrieval On Goal Event Of Soccer Video

With the rapid development of Internet technology and fast-growing amount of videoinformation, more efficient video search is required. It is an important branch of videosearch field to analyze and search the wonderful event in sports video. Goal event is thetypical kind of wonderful events in soccer video. However, the relationship among thedata used to describe the video information is complex and unique. It definitely increasesthe difficulty to realize the analysis and search on goal event of soccer video. Therefore itis necessary to do further research on analysis and search on goal event of soccer video.Although there are many kinds of goal event retrieval, such as the method based onthe goal position or field line distribution, the method based on slow motion retrieval, themethod integrated by audio characteristics and so on, which can play a quite goodperformance on small amount of video data set. These methods cost too much to handlelarge amount of video data set. Therefore we consider doing goal event detection on smallamount of video data set to gain samples firstly. Then cluster the samples to analyze thepersonality and common of them which belonged to different clusters. At last do retrievalon large amount of video data set using classification.The prototype system consists of goal event detection module, clustering module andretrieval module, which are in charge for the goal event detection on small amount ofvideo data set, sample partition and classifier’s train and decision separately. We judgewhether the three kinds of shot match the rules to realize the first module, use K-meansclustering algorithm whose similarity measurement is improved to realize the secondmodule, and construct level binary tree which determine the order of classifiers to realizethe third module.