基于Kalman滤波和区域匹配的视频多目标跟踪

Multi-target Tracking for Videos Based on Kalman Filter and Region Matching

提出了1个基于Kalman滤波和局部区域匹配的足球球员检测算法。首先利用改进的主颜色分割提取场地,使得预设参数个数降为1个,增加了算法的适用性;其次对球员进行运动跟踪,根据Kalman滤波器预测的位置设置局部搜索区域,采用局部构图的方法获得球员最佳匹配,并在球员遮挡情况下对不同队球员和裁判间的分离采用颜色直方图匹配和质心判定,对同队队员遮挡辅助相对位置及运动轨迹跟踪。实验证明,该算法对于多目标在场景中的遮挡、分离、消失及出现均具有很好的处理效果。

This paper proposes a soccer player detection and tracking method based on Kalman filtering and local matching. Firstly, the improved dominant color segmentation is used to extract the soccer field, which makes the predefined parameter drop to one and also enhance the adaption. Secondly, the local searching region is predicted by Kalman filter during the player tracking, and the composition strategy is used to get the optimal matching objects. Histograms and centroids are used to judge the occlusion between different teams and the referee, and the relative position and motion trajectory are furthermore used to detect the occlusion in the same team. Experiments show that this proposed algorithm has good performance for multiple targets tracking in the scene of occlusion, separation, disappearance and reappearance.