Research On Football Match Event Detection Method Based On Video

Football game video last in a long time,but most of which are not attractive to audience.How to extract attractive video segments from the whole video has been becoming a technical problem in the field of video analysis.Recently,many problems still exited in events detecting from football game video,for example,high cost of video processing time,low accuracy of event detecting,difficulty of obtaining standard training sample,etc.The paper put forward to a set of solution.To soccer videos,the shots are segmented and the key frames are extracted.The soccer videos are divided into semantic shot sequences,such as long-view shot,middle-view shot,close shot,close-up shot,replay shot and out of field shot.The semantic annotation makes the most cost of time in the whole algorithm.This dissertation combines the classical algorithm with artificial rules.Through analyzing the semantic structure of goal events,combining with soccer domain knowledge and using special shots,achieves the preliminary position of the target event sequence.The improved HMM is used to classify shot sequences which have been preliminary positioned.Since the algorithm of Baum-Welch in HMM is easy to fall into the local optimal solution,this dissertation adopts the improved scheme combined with genetic algorithm.The feasibility of algorithm has been proved,it shows that:1.The time cost can be saved in the case of ensuring accuracy,when combined classic semantic annotation algorithm with simple artificial rules.2.The detecting of special shots contribute to the preliminary positioning.3.The higher accuracy of improved Baum-Welch algorithm can be realized in the case of shortage of training samples.