Research On Basketball Video Annotation And Realization Of Prototype System

With the rapid growth of the video data's amount, how to efficiently retrieve the useful information become very urgent. As the base of video indexing and searching, video annotation has great significance for its application prospect and research value. This thesis which is supported by Natural Science Fund of Shandong (Y2008G09) focuses on basketball match video annotation, and the main object is constructing the video index and prepare for video retrieve.This thesis is mainly researching on the basketball match video annotation. In the process of annotation, shot is the process unit, which boundary detection has high priority. This paper detects the shot boundary based on the difference between the histogram of key region. In the contracture of the sports video, scene's level is higher than shot's and it can be obtained by clustering shot into different classes. RFCM cluster can be used to classify the shots and annotate the scene. In the semantic detection, moving object detection and tracing is the basis. In this paper adaptive Gaussian Mixture Model is used to background model, and CamShift and Kalman filter are used to trace the players and ball. In the annotation of the semantic, DBN is adopted to match the low-level features with high-level semantic. For the basketball expert, the basketball tactics are their interest, this paper uses HMM to annotate the tactic model. What's more, a prototype system, which includes manually annotation, auto annotation and virtual tactical panel, is designed to annotate the video and show the tactics.The implement of the algorithms are all based on Visual c++6.0and Visual c# 2008, and also OpenCV and Aforge.net class base are used too. Experiments result show that the method introduced in this paper annotate well.