融合颜色与纹理特征的粒子滤波目标跟踪

Object tracking based on combined feature of color and texture with the particle filter

针对采用单一颜色特征的粒子滤波目标跟踪算法在背景相似、光照变化复杂的场景下会导致跟踪失败的问题,提出一种基于LBP纹理和颜色特征融合的粒子滤波跟踪目标算法。综合加权颜色直方图和LBP纹理直方图进行目标特征描述,建立目标观测模型;同时粒子滤波进行状态预测,利用Bhattacharyya系数进行相似度测量,作为目标区域参考模型更新准则,实现权值更新;最后对权值归一化处理,得到目标位置状态的最终估计。实验结果表明该算法不仅提高了跟踪方法的鲁棒性,而且在目标遮挡、光照变化等干扰下,具有较好的准确性。

In some complex scene such as similar color distribution backgrounds and frequently light changing, color-based particle filter algorithm may easily Cause the unstable tracking or even lost the target. To solve the problem, a combined feature of color and LBP texture based object tracking with the particle filter algorithm is proposed. Firstly, color and LBP texture histograms are used to describe the feature to build objective observation model. Secondly, the particle filter system goes the state prediction. We apply the Bhattacharyya coefficients to compute the similarity between the reference location distribution and the candidate location distribution. We get the Bhattacharyya distance to update the weights and judge whether the model should be update or not. Finally, we normalize the weights and get the final estimate of the state of the object location. Experimental resuhs show that the proposed method is not only robust but also accurate in object occlusion and light changing environment.