基于多特征融合的粒子滤波目标跟踪算法  

Multi-feature Fusion Based Particle Filter Algorithm for Object Tracking

作　　者：刘士荣[1] 朱伟涛[1] 杨帆[2] 仲朝亮[2] 

机构地区：[1]杭州电子科技大学自动化研究所,浙江杭州310018 [2]华东理工大学自动化研究所,上海200237

出　　处：《信息与控制》2012年第6期752-759,共8页Information and Control

基　　金：国家自然科学基金资助项目(61175093)

摘　　要：为提高背景混淆下的视频目标跟踪的效果,同时结合视频目标跟踪中状态模型和观测模型的非线性、非高斯特点,提出了一种基于多特征融合的粒子滤波目标跟踪算法.为了获得准确的目标颜色模型,采用了一种自适应选取目标颜色直方图的方法.针对光照变化、背景混淆对目标跟踪的影响,提出了在粒子滤波框架下颜色直方图和边缘方向直方图相结合的方法来提高目标跟踪的鲁棒性.实验结果表明,该算法不仅能够对单一或多目标进行有效跟踪,而且在受到类似目标颜色干扰情况下,也能得到较好的跟踪效果.A multi-feature fusion based particle filter algorithm for object tracking is proposed to improve the tracking performance with background confusion and to deal with the nonlinear or non-Gaussian characteristics of state model and observation model in vision object tracking field. An adaptive method of choosing object color histogram is presented to get an accurate color model of the object. To cope with the influence of illumination variation and background confusion, a hybrid algorithm of color histogram and edge orientation histogram under particle filter framework is put forward to improve robustness of object tracking. The results of the experiment show that the proposed algorithm can track both single objects and multi-objects, even with similar color disturbance.

关 键 词：目标跟踪 粒子滤波 自适应颜色直方图 边缘方向直方图 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]