自适应分块的多特征融合多目标跟踪  

Multi-target tracking method based on the adaptive fragment and multi-feature fusion

作　　者：别秀德 刘洪彬[1] 常发亮[1] 彭志勇[1] BIE Xiude LIU Hongbin CHANG Faliang PENG Zhiyong(School of Control Science and Engineering, Shandong Univ., Ji＇nan 250061, China)

机构地区：[1]山东大学控制科学与工程学院,山东济南250061

出　　处：《西安电子科技大学学报》2017年第2期151-157,共7页Journal of Xidian University

基　　金：国家自然科学基金资助项目(61673244;61273277);高等学校博士学科点专项科研基金资助项目(20130131110038)

摘　　要：针对多目标跟踪过程中存在目标遮挡、表观变化以及目标相似的情况,提出一种基于自适应分块的多特征融合粒子滤波多目标跟踪方法.该方法首先根据目标灰度投影进行自适应分块,融合颜色直方图及方向梯度直方图特征描述各子块,并引入加权Bhattacharyya系数计算粒子的子块匹配度;然后利用模糊C均值聚类获得每个目标的粒子群,得到目标最优状态估计;最后融入粒子空间信息更新子块权重.实验结果表明,该方法在多种复杂情况下,均能准确鲁棒地跟踪多个目标.In order to solve the problems of occlusion, appearance change and similar targets in multi- target tracking, this paper proposes a multi-target tracking method with the particle filter based on the adaptive fragment and multi-feature fusion. First, we divide each target into a few fragments adaptively according to its gray projection and describe each fragment with the color histogram and histogram of oriented gradient （HOG） feature, of which similarity can be obtained by adopting the weighted Bhattacharyya Coefficient. Then, we obtain particle sets of each target by FCM clustering. The optimal state estimation of each target is calculated through particles state in the subgroup. Finally, the weighting factor of each fragment is updated according to the reliability which is calculated by considering particles space information. Experimental results show that the proposed method can track the targets in many complex circumstances accurately and robustly.

关 键 词：多目标 自适应分块 多特征融合 HOG 粒子滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]