High-level Semantics Analysis Of Soccer Video

In recent years, Automatic extraction of highlights and summarization is animportant task in Context-Based Video Retrieval field, especially in the soccor sportdomain.Soccer is a very popular game,which has a large number of audiences allover the world.This paper proposed a framework for soccer video high-level semanticanalysis, including shoot event, foul event，penalty event. The mainly works aredescribed as follows:(1) Shot segmentation: A video are composed of a series of scenes, shotsegmentation is the first step, the algorithm sets the value of the four threshold. It isdivided into visible flied and invisible flied by comparing threshold.In the visibleflied double threshold segmentation algorithm is used. In the invisible flied singlethreshold segmentation algorithm is used. This paper mainly combines singlethreshold with double threshold algorithms to segment video by cutting shotdetection.(2) Shot classification: select key frames in video segments, with key frame as adescription of the shots, using field grass rate to detect field region in eachframe.Shots are classified into three types: long-view shot, middle-view shot, close-upshot and out of field shot.(3) Slow-motion shots detection, slow-motion often appers after highlightevents.This paper uses interframe histogram difference method to locate and detectslow motion. Through detecting the slow motion, we can effectively detect whetherthere is a highlight event.(4) Forbidden zone detection, as shoot often happens near forbidden zone, andforbidden zone lines in a long-view shot can be detected to find forbidden zone, whichcan help shoot event detection. This paper uses the Hough transform on the forbiddenzone to detect white line detection.Though the forbidden zone detection can beeffectively assisted shooting events detection.(5) Detection of caption, if a goal occurs,the score will change,so there will be score caption in the video.In this paper, the K-means clustering algorithm based onwavelet transform analysis and extraction on the captions.therefore,detection ofcaption are used to help detect shoot event and determine whether there is a goalevents.Through the analysis above, this paper proposed a framework for soccer videohigh-level semantic analysis.It uses an algorithm triggered by slow-motion replays todetect highlights events, assisted by forbidden zone detecting and caption regionextraction.This paper mainly idea is through the low-level feature extraction,middle-level fearture analysis and high-level semantic events detection.