Ball Detection And Tracking In Soccer Video

As the most popular sports video, soccer video has drawn attention from more andmore researchers. How to detect and track the targets effectively in soccer video is veryimportant to the follow-up analysis and it has become a very intractability problem inthe information and multimedia fields. In this paper, the research status of ball andplayer detection and tracking is introduced at first. After that, the theory of shot segmentand moving target detection and tracking is detailedly studied. An effective solution ofball detection and tracking in soccer video is presented in this paper.Firstly, this paper proposes an algorithm of targets detection in soccer video. In thismethod, the soccer video is decoded into consecutive frames and the histograms of H, Sand I of the training images are calculated. After the domain color is calculated out, thismethod takes an image segment on the image sequences based on cylindrical distance.Because of the holes and burrs in the images after segmentation, a morphologicalprocessing is taken on these images. After that, the field area of each image is detectedand the objects in the field area are marked as the candidate balls and players. Toremove the influence of the field lines to the detection algorithm, the Hough transformis used to detect the field lines and to wipe them away. The results of the experimentsshow that the method is effective.Secondly, a ball tracking framework is presented in this paper. According to theresult of object detection algorithm above, several ball selection rules based on shapeanalysis are defined. With these rules, the candidate balls are extracted from the fieldarea in each image. These ball candidates are tracked with Kalman Filter. In this paper,the solutions of occluding and trajectory split are provided. After the process of KF,there will form several trajectories. The real ball trajectory is picked out according tothe rules of trajectory selection.Thirdly, a target tracking algorithm based on particle filter is proposed. Thisalgorithm uses the color and HOG features of balls to solve the nonlinearity problem intracking. The algorithm extracts the color and HOG features out and fuses them togetheras a feature template. In the particle filter framework, the features of each particle arecalculated and the distance between the feature template and the feature of each particleis computed and normalized as the weight of the particle. According to calculating theweighted sum of all particles, the position of the ball will be got. The experimental results indicate that the algorithm is effective and has the ability to resist to the targetoccluding.