改进的粒子群优化粒子滤波的人脸跟踪  

Face Tracking Based on Improved Particle Swarm Optimization on Particle Filter

作　　者：严石[1] 杨定礼[2] 张宇林[2] 杨玉东[2] 季仁东[2,3] 皇甫立群[2] 

机构地区：[1]淮阴工学院成人教育处,江苏淮安223001 [2]淮阴工学院电子信息工程学院,江苏淮安223003 [3]南京航空航天大学理学院,南京210016

出　　处：《淮阴工学院学报》2016年第1期12-18,共7页Journal of Huaiyin Institute of Technology

基　　金：淮安市科技支撑计划工业项目(HAG2013064);淮阴工学院科研基金项目(HGB1202)

摘　　要：针对粒子群优化粒子滤波的人脸跟踪方法出现的"粒子退化"、"粒子贫乏"、"局部最优"、"粒子早熟"等问题,提出改进的粒子群优化粒子滤波(IMPSO-PF)的人脸跟踪方法。该方法首先提出了非均匀的空间直方图的颜色跟踪线索,提高了跟踪的鲁棒性,其次提出了多样性函数diversity以及粒子集中度函数difference,并根据diversity与difference的关系更新速度与位置,使粒子不断向高似然区域运动。实验结果表明,该方法既保证了粒子多样性,同时也防止了粒子较早的成熟。In order to solve the ＂ particle degeneracy＂, ＂ particle poor＂, ＂ local optimum＂, ＂ particle precocious＂ and other questions which appeared in face tracking based on particle swarm optimization particle filtering,a method of face tracking based on improved particle swarm optimization particle filter（ IMPSO- PF） was proposed. In this method,firstly,tracking clue based on non- uniform color space was proposed,it improved the robustness of the tracking. Secondly,the diversity function（ diversity） and concentration degree function（ difference） were proposed. Speed and position were updated according to the relationship between diversity and difference. All particles moved to the high likelihood area. Experiments showed that the method not only ensured the particles diversity,but also prevented particles from being precocious.

关 键 词：粒子群 粒子滤波 人脸跟踪 非均匀的空间直方图 多样性函数 集中度的函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]