双重特征粒子滤波目标跟踪算法

Particle Filter Object Tracking Algorithm Based on Dual-feature

针对传统粒子滤波目标跟踪算法在光照变化及遮挡情况下跟踪准确性不高的缺陷,提出一种双重特征粒子滤波目标跟踪算法.首先,提出一种融合边缘信息具有抗光性的Sobel中值二值模式特征描述算子,建立视频每帧的纹理直方图;对视频每帧分别提取HSV颜色空间的色度核函数直方图、饱和度核函数直方图,再通过线性加权得到颜色直方图;在此基础上,分别对每帧的纹理直方图与和颜色直方图进行归一化处理,并引入到粒子滤波算法中计算目标特征粒子权值,再通过加权准则确定当前帧目标的最终位置.对比实验结果表明,本文算法能够有效提高在光照变化和遮挡情况下目标跟踪的准确性.

To improve the tracking robustness of the traditional particle filter target tracking algorithm under the condition of illumina- tion change and occlusion,particle filter target tracking algorithm based on dual-feature is proposed. Firstly, the paper presents a novel Sobel Med-value Binary Pattern （ SMBP） feature description operator with light resistance and edge information, and establish texture histogram for video each frame. Color histogram is obtained by linear weighting of the hue kernel function histogram and saturation kernel function histogram in HSV color model. On this basis, for each frame of color histogram and texture histogram are normalized respectively, and introduced into the particle filter algorithm to calculate the particle weight of target characteristics, the final location of the current frame target is determined through the weighted criteria again. The experimental results show that the proposed algorithm can improve the accuracy of target tracking effectively under the condition of illumination change and occlusion.