基于特征融合的粒子群优化粒子滤波跟踪方法  

Particle Filter Tracking Method base on Feature Fusion and Particle Swarm Optimized

作　　者：袁开宇[1] 储珺[1] 冷四军 朱陶[1] 曾接贤[1] 

机构地区：[1]南昌航空大学计算机视觉研究所,南昌330063

出　　处：《南昌航空大学学报（自然科学版）》2015年第1期33-41,48,共10页Journal of Nanchang Hangkong University(Natural Sciences)

基　　金：国家自然科学基金(61263046);江西省自然科学基金(20122BAB201037)

摘　　要：针对传统粒子滤波跟踪算法重采样时存在粒子退化、目标与背景颜色相似和尺度变化导致的目标定位不准确问题,本研究提出了一种基于特征融合的粒子群优化粒子滤波跟踪算法,算法利用粒子群优化进行粒子权值更新,用当前状态估计值与各粒子状态的差值大小作为评价标准,促使粒子采样向真实状态区域移动,减缓粒子退化,提高了粒子滤波跟踪算法的跟踪性能。针对跟踪目标尺度变化导致的定位不准确情况,引入了归一化转动惯量(Normalized moment of inertia,NMI)特征,并将它与颜色特征采用乘性融合策略进行融合来描述目标特征,提高目标复杂场景下的定位精度。通过在多个标准测试视频上进行试验,实验结果表明,本研究提出的方法对动态背景场景中尺度变化目标和背景颜色相似目标的跟踪具有较好的准确性和鲁棒性。For the presence of particles of degraded traditional particle filter tracking algorithm when resampling, Not a good solution to partial occlusion morph targets and target tracking problems, This paper introduces the multi-feature fusion based particle swarm optimized particle filter tracking method. Particle swarm optimization algorithm uses particle weights be updated with the current estimate of the difference between the size of the state and the state of each particle as evaluation criteria, prompting the true state of the particle sampling area to move, reduce particle degradation, improve the tracking performance of the particle filter tracking algorithm. For target deformation and occlusion, This article introduced the Normalized moment of inertia （NMI） feature, Will it with color features muhiplicative fusion strategy fusion is used to describe the target characteristics. Through the experiments on several standard test video, experimental results show that the proposed method for dynamic background scene morph targets and partial occlusion target tracking with better accuracy and robustness.

关 键 词：目标跟踪 粒子滤波 NMI特征 粒子群优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]