Research On Semantics Of Sports Video Based On Rule

With the rapid growth of the amount of videos, how to use it efficiently is a big problem. Video's semantic analysis becomes a focus of the video processing. This thesis, which is supported by Natural Science Fund of Shandong (Y2008G09) focuses on video semantic analyze, especially basketball match video. The aim is to process, analyze and understand the video content to construct the architecture and index, with which the user can locate the interest video easier.In the analysis of the video, the extraction of semantic events is the most important thing. Because of the semantic gap which is between low-level features processed by computer and high-level semantics understood by human, the analysis is difficult. The features extraction is the base of video analysis, color feature is extracted and converted into the color space that is consistent with human perception in our process. As the processing unit, shot boundaries should be detected before the analysis. In this paper, the key domain is defined and its histogram's difference is used to detect the boundaries. The core of the sport video analysis is extraction of the semantic events and their relationship between the raw video data. Our research emphasis on the video analysis combined with the domain knowledge. Firstly, the events that are replayed are defined as wonderful events, then the common events are detected and the video's structure is formed. Based on the fundamentals of compiling, this paper uses context-free grammar to establish the framework of video analyze. Each shot will be marked on its semantic content and the shots in the whole video will form a sequence, at last the catalogue of the video will be formed.On the part of application, shot boundary detection is firstly realized on basis of the key domain's histogram. And then the main semantic are analyzed and detected. Furthermore, a sport video analyze raw framework is developed with C#.