Replay-Based Soccer Video Highlights Detection

Due to the existence of massive soccer videos on the Internet, fast-paced life makes people concern not how to transfer and play all the matches, but focused on interesting or highlight clips. Therefore, how to analyze video content to meet the different equipment, different types of audiences is of great significance.Based on systematic analysis of the soccer highlights extracting and semantic event detection methods, effectively solve the problems that the existing affect arousal model can not reflect the intensity of matches and highlights location and event detection are not ideal, by realizing replay and goal-mouth detection. Improved shot boundary discriminate method, and then proposed a soccer shot boundary detection algorithm based on statistical confirmed. The replay scene detection based on Logo, which can obtain by two ways. Videos whose Logo transition fit motion characteristic, automatically acquire Logo pixels by comparing the similarity of motion vectors. A convenient tool is designed for any other videos whose Logo transition is complex, and Logo pixels could be acquired by only a small amount of interaction. Construct a feature vector containing the width of the goalposts, based on SVM recognize goal image. Structure replay feature to improve affect arousal modeling, and a new highlights locating method with short, medium and long three different lengths is proposed. Based on the goal, shooting and foul rules to achieve highlights event detection, and affect arousal modeling and event detection are used in www.showball.net.The experiments show that the proposed method of replay detection is with high recall and precision. Affect arousal curve added with replay features more accurately reflect the intense situation of soccer video, and semantic event detection also has higher recall and precision, especially for the goal event. However, further processing is needed for some special replay model, and the rules of shooting and foul events need to further enhance to improve recognition accuracy.