基于联合直方图的自适应粒子滤波跟踪算法

An Adaptive Particle Filtering and Tracking Algorithm Based on Joint Histogram

针对传统的粒子滤波跟踪算法在跟踪过程中由于目标形态不断变化以及目标被部分遮挡导致跟踪效果不理想的问题,提出了一种基于颜色纹理联合直方图特征融合的自适应粒子滤波算法。该算法从视觉特征集中选取全局特征(颜色直方图)与局部特征(LBP纹理直方图),组成目标颜色纹理联合直方图;将当前目标的联合直方图与初始目标的联合直方图的巴氏距离,作为粒子更新权重的依据,从而算法可以自适应地选取一组最优粒子集,并且以较少的粒子数目来保持算法的多样性,解决了传统粒子滤波算法中存在的粒子退化以及重采样过程中的粒子匮乏问题。实验结果表明,与传统的粒子滤波算法相比,该算法可以更鲁棒地跟踪部分遮挡和形态剧烈变化的运动目标。

In order to solve the problem that traditional particle filter algorithm tracking the target is not ideal due to its appearance changing and partial occlusion,we propose an adaptive particle filtering algorithm based on joint histogram of color and texture.This algorithm selects the global feature（ color histogram） and local feature（ LBP texture） to combine color and LBP texture histogram.By calculating the Bhattacharyya distance of the current target combined histogram and the initial target combined histogram as the basis of particle update weight,it can not only select an optimal set of particles but also keep its diversity with a smaller number of particles then,which solves the problem of particle degradation and the shortage of particles in re-sampling process in the traditional particle filter algorithm.Experiment shows that compared with the traditional particle filter algorithm,the proposed algorithm can robustly track the moving target with changes of the appearance or partly occluded.