摘要： 基于模型的跟踪方法难以处理足球视频中球员形态发生较大变化的情况。为此，提出一种改进的多特征自适应融合的球员跟踪算法。利用自适应高斯混合模型检测球场和球员区域，使用球员HUE颜色特征的Bhattacharyya距离度量法代替传统的模板匹配方法，辨别球队归属，自适应地融合目标模型的颜色、形状和时空特征信息，实现对球员的跟踪，采用三点估算预测方法解决球员完全遮挡现象。实验结果表明，该算法能较好地解决球员之间的遮挡问题，在球员形态变化较大时能实现稳定的跟踪。

关键词: 自适应权重, 特征融合, 时空特征, 三点估算, 完全遮挡, 目标跟踪

Abstract: An improved player tracking algorithm based on multi-feature adaptive fusion is proposed to solve existing problems that the model-based tracking method is difficult to deal with greater change of players’ form in football video. This paper uses the adaptive Gaussian mixture model to detect football playfield and players. The Bhattacharyya distance of players’ HUE color features is used to distinguish ownership of the team instead of traditional template matching methods. The method fuses the color, shape and temporal-spatial feature information of target model adaptively for tracking the players and uses three-point prediction method to solve the complete occlusion between players. Experimental results show that the algorithm deals well with the occlusion between players, and can track robustly when the players’ shape changes greatly.

Key words: adaptive weight, feature fusion, temporal-spatial feature, three-points estimation, complete occlusion, object tracking

中图分类号: 

• TP301.6