Visual Target Detection And Tracking In Soccer Video

In recent years,deep-learning technology is widely used in computer vision tasks and has achieved huge success in the fields of visual detection and tracking.On this basis,how to translate the technology from theory to practice has become a problem to be solved.Visual target detection and tracking in soccer video is a challenging task which has great practical and commercial value.The traditional soccer game object trajectory extraction usually adopts the way that the athlete carries the recording chip with him,but this kind of method cost is very high,and it is difficult to be popularized in the amateur stadium.Apart from this,some research only uses camera in soccer video processing,but as the goal in the soccer video share similar appearance and have frequent occlusions,these methods can only segment the goal and football from image,rather than on the track of tracking or only for a short time.Based on related research of computer vision and deep-learning,this paper designs a method of multiple-camera target tracking which is suitable for soccer video,by using multiple cameras in course of image acquisition,it has the ability to do a long accurate tracking for multiple targets in the field.The main research contents of this paper are as follows:The paper designs a single-camera and multi-target framework which is capable of tracking multiple targets accurately based on deep-learning detector and correlation-filter tracker,the framework uses data correlation algorithm to fuse the result of detector and tracker.Through the analysis of the target in the stadium,the paper select a variety effective feature descriptions to optimize the multiple modules in the system,and designs the division algorithm based on CN color features which could complete the team-divison task in soccer video.In soccer video,the scope of the single camera can't cover the entire stadium and player movement will cause its frequent disappearance and reproduce in a single camera,we use multiple-camera system to solve this problem.Firstly,the method uses multiple cameras to collect image data independently and use the single-camera and multiple-target tracking method to extract the player movement in the image,and then it designs a multiple-camera data fusion method to fuse data in these camera systems.After that,it preserve the result.The paper illustrate the hardware construction and software design of the whole system,and do experiments in the actual scene.The experimental results show that the multiple-camera and multiple-target tracking algorithm designed could do a long-term and accurate tracking for the targets in the soccer video based on the accurate single-camera and multiple targets algorithm and multiple-camera system,it also has great robustness and real-time performance.