基于局部分块和背景加权的视觉跟踪算法

针对视觉运动目标的鲁棒跟踪问题,提出了一种基于局部分块和背景加权的跟踪算法。首先对目标的前景和背景区域进行分块采样,然后利用基于积分直方图的局部快速穷搜索算法计算每一个分块在当前帧中的后验概率图,最后对后验概率图确定的对应分块的搜索结果赋予不同的权值,进而计算出目标在当前帧中的位置。实验结果表明:基于局部分块加权的跟踪算法比单纯的背景加权跟踪和分块跟踪具有更高的跟踪精度和成功率,且算法复杂度较低。

A Visual Tracking Algorithm Based on Local Patches and Weighted Background

In order to improve the robustness of visual moving object tracking, a tracking algorithm is proposed based on local patches and weighted background. Firstly, the local patches are sampled respectively from the foreground area and background area of the object. Secondly, the probability map of each patch to the frame image is calculated through the integral histogram based local exhaustive search. Finally, different weights are assigned to the searching results of patches by maximizing each probability map, and the final localization of the object in current frame is obtained by the weighted sum of the searching results of local patches. The experimental results indicate that the proposed algorithm exceeds the background weighted tracking and local patch based tracking in both tracking precision and tracking successful rate. Besides, the proposed algorithm is not too complex.