中国科学院云南天文台机构知识库

针对标准粒子滤波在目标与背景颜色相似和存在遮挡时易出现跟踪丢失的问题,提出一种融合颜色和梯度直方图信息的免疫粒子滤波算法.颜色直方图能对目标进行全局描述,梯度方向直方图含有相应的结构信息,通过融合多特征信息,使得单纯依靠颜色特征不能很好适应环境变化的情况得到了有效改善.同时通过加入免疫粒子算法,提高了粒子的多样性,尤其是在目标被遮挡时,减少了“样贫”的影响.实验结果显示,采用综合直方图特征的改进粒子滤波算法在背景颜色干扰及目标被遮挡时仍能稳定地跟踪目标,具有较强的鲁棒性. 

Focusing on fails to track with standard particle filters when object and background have similar color and object is occluded, a new algorithm of immune particle filter object tracking based on histograms of color and oriented gradient is proposed.  Color histogram is the global description of targets in color image, histogram of oriented gradient contains some construction information. The gradient feature is added, the poor performance of only one feature is improved when the environment changes.  In addition, the proposed immune optimization algorithm improved the diversity of particles, especially the decline of the impact of sample impoverishment when occlusion occurs. Experimental results show that the proposed method is able to track the object stably when background color interference and object is occluded，which has a strong robustness.