Moving Object Extraction Method Based on Background in Sports Video

摘要

In order to conform to the psychological needs of the audience, director usually reproduces the wonderful lens in the form of slow motion, so the slow motion of sports video contains a wealth of semantic information, and advanced semantic content analysis of slow motion is extremely import. Due to the relative motion between the camera and monitoring scene, background also moving while targets moving between two frames. To solve this problem, this paper continued using the method of background compensation for moving object extraction. In order to transform motion background into a static background, a new frame is filled between two frames, and then frame difference method is used to extract moving targets. In the process of background compensation, this paper remove most of the feature points according to the characteristic of long-distance between harris angular which are line in the tennis court, thus the amount of computation is reduced when feature points matching. When the feature points selected for parameter estimation, relevant domain knowledge is used, and the speed of the algorithm is improving.