Research On Player Tracking Algorithm In Soccer Video

At present,there are many researches on moving target tracking technology,which have wide application value.Soccer has become one of the most popular sports in the world,with a high degree of concern,and player tracking have important practical significance for the general audience,coaches and referees,and is the theoretical basis of sports video content analysis.There is no tracking algorithm to deal with all the scenes perfectly,and the research on tracking algorithm for soccer video player should be integrated with its characteristics of specific domain to improve the performance of tracking.In order to describe the tracking target information better,global features and local features are combined.Color have significant distinction for the different team players,the HSV color space non-uniform quantization algorithm is used to remove the main color of the stadium,and the main color histogram based on the upper and lower of the player is extracted.The Haar-like local texture features are extracted to construct the weak classifier,using Integral Image to accelerate the computation.We improved the traditional tracking algorithm with online multiple instance learning,on the one hand,considering the contribution of different examples in the bag,the example is given weight.The example weights in positive bag are based on its center distance from the target position,and the example weights in negative bag are the same.On the other hand,due to the limitation of traditional algorithm for predicting the next frame position only within the fixed circular neighborhood,using the motion model of particle filter to generate the candidate set.in: order to measure the performance of different trackers in soccer video,the soccer video is divided into four scenes,and the data set is manually marked by the scene.The algorithm designed in this paper is compared with the other four algorithms which are also fixed tracking scale for performance evaluation and comparison.As the experiment shows,the proposed algorithm has strong robustness and can adapt to the player tracking in different scenes of soccer video and meet the real-time requirements.The next step of research is to consider how to use the player's context information to assist the player tracking to better handle the player tracking of the mixed scene.