基于粒子滤波与多特征融合的视频目标跟踪

Visual Object Tracking Based on Particle Filter and Multiple Cues

提出了一种基于粒子滤波和多特征融合的视频目标跟踪方法。以粒子滤波为跟踪框架,根据颜色跟踪中存在的问题提出将颜色与目标运动信息融合,利用融合后的信息确定粒子的权值。利用重采样策略缓解退化现象对粒子滤波的影响。针对2段不同的视频进行了不同算法的仿真与性能的比较,实验结果表明,本文方法在计算量增加不多的情况下大大改善了跟踪的性能与鲁棒性,尤其当目标与背景颜色相近时仍然能够准确地对目标进行跟踪。

A visual object tracking method based on particle filter and multiple cues is presented. According to the problems in color-based tracking,our proposed method integrates color information with motion information,and the weights of particles are determined by this integration. The resample strategy is also used to cope with the impact of degeneration. We simulate different algorithms and compare their performance on two pieces of video. The experiments show that our method makes the tracking more robust with a moderate increase in computation. Especially when the object is similar with the background in color,it still can be accurately locked.