基于多特征融合的权值优化粒子滤波目标跟踪  

Target tracking based on multiple features fusion and weight optimization particle filter

作　　者：熊杰 吴晓红[1] 卿粼波[1] 何小海[1] 尤晓建[2] 

机构地区：[1]四川大学电子信息学院,四川成都610065 [2]西南科技大学,四川绵阳621000

出　　处：《电视技术》2017年第9期155-160,共6页Video Engineering

基　　金：成都市科技惠民项目(2015-HM01-00293-SF);特殊环境机器人技术四川重点实验室开放基金(14zxtk03);国家自然科学基金委员会和中国工程物理研究院联合基金(11176018)

摘　　要：针对传统粒子滤波目标跟踪算法在光照变化、遮挡和姿态变化较大时对目标跟踪的鲁棒性不高的缺陷,提出了一种多特征融合的权值优化粒子滤波目标跟踪算法。该算法在建立目标模型时提取了运动目标区域内的HSV空间的核加权颜色特征、经过优化处理后的LBP纹理特征和边缘特征,再将提取的特征进行加权融合;同时采用一种权值优化组合方法在粒子进行重采样前,对粒子的权值进行优化处理,避免粒子贫化问题的出现,保持粒子的多样性;最后结合模板更新策略对目标模板实时更新。实验结果表明:本文算法对光照变化、遮挡和姿态变化下的目标跟踪具有较好的鲁棒性。Given that the robustness of traditional particle filter algorithm for target tracking is not very well, especially in illumina- tion change, occlusion and pose variation, therefore, a target tracking algorithm based on multiple features fusion and weight opti- mization particle filter is proposed. The kernel weighted color feature based on HSV, optimized LBP texture feature and edge fea- ture are extracted in the moving target area when building object model in the algorithm. Meanwhile, the weighted value of parti- cles are optimized before particle resembling, magnificently avoiding particles impoverishment and increasing the diversity of parti- cles. Finally, the introduction of template updating strategy is combined to update target template. Experiments show that： the im- proved algorithm has good robustness when tracking in the situation of illumination, occlusion and pose variation.

关 键 词：多特征融合 权值优化 目标跟踪 粒子滤波 模板更新 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]