基于模糊测度的多特征融合鲁棒粒子滤波跟踪

Multi-feature fusion robust particle filter tracking based on fuzzy measure

针对基于单一颜色特征的粒子滤波跟踪算法易受光照变化、部分遮挡及相似干扰物的影响,而利用多特征融合的粒子滤波方法存在各特征权值、跟踪模板及窗口大小自适应选取问题,提出了一种基于模糊测度的多特征融合鲁棒粒子滤波跟踪算法。采用颜色及边缘方向直方图来描述目标量测模型,通过分别计算这两类特征在候选目标与参考目标之间的Bhattacharyya距离来确定其各自特征的模糊测度,通过查取模糊规则表来自适应地确定两类特征的权重;将连续帧的多特征联合模板更新机制用于对初始目标模板的更新;针对目标发生尺度变化造成跟踪窗口难以自适应的问题,通过引入粒子离散度实现了跟踪窗尺寸的自适应调整。实验结果表明:所提出的跟踪算法位置平均误差小于8个像素,相比于传统方法可以有效克服光照、部分遮挡以及相似目标干扰等影响,具有较高的跟踪精度及较强的鲁棒性。

In order to overcome the problem that particle filter tracking based on the single color feature is susceptible to illumination changes, partial occlusion and the interference of the similar, and the feature weight, tracking template and tracking window size are difficult to adaptive when the particle filter tracking method based on multi-feature fusion is used, a multi-feature fusion particle filter tracking based on the fuzzy measure is presented. A color histogram and a edge orient histogram are used to describe the target measure model, and Bhattaeharyya distance of these two features between the candidate and reference targets is used to determine their separate fuzzy measures. Then, the weights of these two features are adaptively determined by referring to the fuzzy rule table. Besides, a combined template update mechanism of multi-feature based on successive frames is adopted to update the initial target template. Finally, particle dispersion is introduced to solve the problem that the tracking window cannot adapt to changes of the tracking target scale. Experimental results in- dicate that the average error of the proposed tracking algorithm is less than 8 pixel errors. Compared with the traditional tracking algorithm, the proposed algorithm can effectively solve the problem of illumination changes, partial occlusion and the interference of the similar, and it can satisfy the system requirements of higher preci- sion and strong robustness.