基于局部敏感直方图的分布场跟踪算法研究

Distribution fields tracking method based on locality sensitive histograms

文章提出了一种基于局部敏感直方图的分布场目标表示方法,克服了原始分布场跟踪方法对光照和参数敏感的缺点。利用局部敏感直方图与分布场模型表示目标空间结构上的相似性,提出利用局部敏感直方图作为一种新的分布场构建方式来表示目标。与分布场目标表示方法相比,局部敏感直方图对像素所在层取更大的权值,因此能够更好地保留目标空间结构。在基准视频序列上的实验结果表明,与其他具有代表性的算法相比,基于局部敏感直方图的分布场跟踪算法用固定的一组参数取得了最佳的跟踪结果。

A new distribution fields tracking method based on locality sensitive histograms is presented. It overcomes the limitation of the original distribution fields tracking method for light change and sen- sitiveness to parameters. By the similarity between locality sensitive histograms and distribution fields in representing the spatial structure, the new method uses locality sensitive histograms as a new kind of distribution fields to represent a target. Compared with the target representation method based on distribution fields, locality sensitive histograms assign a greater weight for the layer of pixels, so it can keep better target spatial structure. The testing results on challenging benchmark video sequences show that the proposed method with a fixed set of parameters obtains the best tracking performance a- mong the common representative methods.