Sports Video, Moving Target Detection And Tracking

Nowadays, sport videos are becoming more and more popular. Therefore, it is useful and has commercial value to analyse sport videos. In this paper, soccer video is taken as an example. Detection, recognition and tracking of moving targets people pay attention to include soccer players and football in the soccer videos are studied.This paper is divided into four parts: moving targets detection, recognition, football tracking and players tracking. And the main work and research results are as followings:(1) This paper adopts the improved K-means clustering algorithm to segment the court area. Then, HSV with RGB model is used to eliminate court area. And ellipse court lines are extracted by the curve fitting method, straight court lines by the classic Hough transform. At last moving targets are detected after eliminating court lines.(2) On target recognition, this paper selects the 7 normalized and invariants Hu moments as characteristics of targets. Targets can be determined whether or not belonging to the same category as the sample by comparing the characteristics of targets and sample.(3) For soccer tracking, this paper adopts Kalman filter to predict the estimated trajectories in order to achieve the tracking of football. And this paper uses the least square method to deal with the football occlusion.(4) For players tracking, CamShift algorithm, based on color model, can take full advantage of players color information, to attain to be real-time. To deal with the player occlusion by other players, CamShift algorithm with Kalman filter is suggested . This paper determines an occlusion factor based on the search window size, and weights the results of Kalman and CamShift according to the occlusion factor.Algorithms proposed in this paper are realized by VC and OpenCV. The experiments show that the detection and tracking algorithms by this paper, are robust and efficient.