基于分块颜色直方图和粒子滤波的物体跟踪

在使用粒子滤波的跟踪方法中，颜色直方图经常被用来作为目标特征。但是普通的颜色直方图易受与跟踪物颜色相似的背景和其他物体的干扰，并且在跟踪目标被部分遮挡后性能也将下降。为解决这些问题，受hog特征启发，提出一种分块重叠的颜色直方图，并且根据分块直方图特点，重新设计了粒子滤波系统的权重计算方法和模型更新方法。实验证明该系统优于传统的颜色直方图特征。

Object tracking based on sub-block color histogram and particle filter

In traditional object tracking with particle filter, color histogram is often used as feature. But color histogram suffers from the interference of background and other objects in the scene which are of the similar color to the tracked objects. Also simple color histogram will fail when the objects are occluded. In order to solve these problems, inspired by hog feature, an overlapping sub-block color histogram is designed as the feature for particle filter tracking system to improve the tracking performance. Also based on the new feature, weight calculation and model update method in particle filter have been redesigned. Experiment has proved its effectiveness.