A Method For Extracting Player Trajectory In Broadcast Soccer Video

The main content of this research is player trajectory extraction in the broadcast soccer videos.Soccer matches attract great attention and there is a growth of demand for automatically analyzing soccer matches.However,due to the diversity of conditions such as the venue of the soccer matches and the color of the clothing,there is currently no universal tracker that fits perfectly into all scenarios.Deep learning technology has been widely used in computer vision in recent years because of its excellent feature extraction ability.This paper attempts to use the deep learning technology to automatically extract the feature of the player to achieve object detection and tracking.In terms of object detection,this paper improves the existing network model to make it more suitable for small object detection tasks in soccer match scenarios.In terms of object tracking,by constructing the similarity matrix,the multi object tracking problem is transformed into a data association problem that can be solved by the Hungarian algorithm.When building a similarity matrix,we calculate the motion information of the object using Kalman filter,extract target detection features using convolutional neural networks and extract the apparent features using Siamese network,and then comprehensively consider the similarity of these three features.For the addition,disappearance and occlusion of the targets in a soccer match,we propose corresponding solutions respectively.Experiments show that the proposed algorithm can do the multi object tracking in the soccer match videos well and adapt to different scenes.This paper is a successful application of deep learning technology in video processing of soccer matches.