复杂背景下多信息融合的粒子滤波跟踪算法  

A particle filter based tracking algorithm with cue fusion under complex background

作　　者：胡昭华[1] 宋耀良[1] 梁德群[2] 樊鑫[2] 

机构地区：[1]南京理工大学电子工程与光电技术学院,江苏南京210094 [2]大连海事大学信息工程学院,辽宁大连116026

出　　处：《光电子．激光》2008年第5期680-685,共6页Journal of Optoelectronics·Laser

基　　金：国家科技攻关计划子项目“基于图像处理技术的城市交通调度管理”资助项目(2004BA111B01)

摘　　要：将颜色和运动这两种信息融合进粒子滤波跟踪器,并提出分层抽样的方法,克服了利用单一信源所带来的跟踪不稳定问题;与典型的基于边缘特征或仅基于颜色信息的粒子滤波器相比,计算简单,并能够有效解决由于目标形状或颜色模糊而产生的跟踪困难问题。实验结果表明,该粒子滤波融合算法在复杂背景下能够稳健可靠的跟踪目标。We present a particale filter（PF） approach to target tracking that fuses color and motion,in which a layered sampiing strategy is embedded. This method overcomes the instability brought by using a single measurement source. Compared with the traditional PF methods that merely use either edge features or color cues,this method has lower computational expense and effectively resolves the tracking difficulties due to imprecise silhouette and/or color models. The experimental results on real world videos show that the proposed method can robustly track moving objects of interest in complex background.

关 键 词：目标跟踪 粒子滤波 信息融合 状态估计 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]