针对分布场目标跟踪算法中使用分布场的目标模型估计鲁棒性较弱的问题,提出一种将分布场与其他特征有效融合的方法,来提高分布场特征表示的有效性。在对每个像素点进行分布场估计时,原始算法仅通过该点的灰度直方图来估计其在灰度空间上的分布,并没有考虑该点的位置与结构信息。为了实现在分布场中对目标结构信息的有效表示,通过对目标中包含结构信息的特殊点进行特殊编码以实现结构信息的融合。实验表明,对于一些复杂环境下的挑战视频序列,融合了结构信息的分布场比原始分布场在目标跟踪的成功率上具有显著优势,且优于当前流行的4种目标跟踪方法。
In order to improve the robustness of the distribution fields ( DF) as an object model in object tracking , we propose a mutli‐feature fusion framework for the distribution fields . In the original DF‐based method , the density histogram was used to estimate the DF of a pixel , but the structural information was ignored . For effective representation of the structural information in the DFs , a special type of coding for the featured points which contain structural information is merged into the DFs . Experiments show that the new method outperforms the original method and four other state‐of‐the‐art tracking algorithms for some challenging video clips .