基于动态显著性特征的粒子滤波多目标跟踪算法  

The Images Tracking Algorithm Using Particle Filter Based on Dynamic Salient Features of Targets

作　　者：张焱[1] 张志龙[1] 沈振康[1] 鹿小莺[2] 

机构地区：[1]国防科技大学电子科学与工程学院ATR国家重点实验室,湖南长沙410073 [2]北京跟踪与通信技术研究所,北京100094

出　　处：《电子学报》2008年第12期2306-2311,2305,共7页Acta Electronica Sinica

基　　金：总装备部武器装备预研基金(No.9140A21040306KG0195;No.9140C8002010704)

摘　　要：针对复杂背景条件下图像序列中运动多目标跟踪问题,提出一种基于动态显著性特征的粒子滤波多目标跟踪算法,该算法借鉴心理学中关于视觉注意的研究成果,综合目标的灰度、细节和运动特性形成稳健的动态显著性特征,用来作为粒子滤波的状态向量.由于该算法中的显著性特征来源于目标的多种底层特性,因此算法具有很强的稳健性,同时,粒子滤波可实现非线性非高斯状态空间模型的最优估计.故而,该算法能够同时处理多个目标跟踪过程中的航迹管理问题,以及目标出现、消失、合并、分裂、被障碍物遮挡等问题.实验结果表明,该算法能够很好地实现复杂图像序列中的多目标跟踪.Focusing on the problem of tracking moving targets in image sequences with complex background,a particle-filter tracking algorithm based on dynamic salient feature of targets is proposed. Using for reference the research results of visual attention of psychology,the algorithm fuses the gray-level and the detail and motion of the targets to form a robust dynamic salient feature as the state vector of the particle filter. The salient feature of the targets in the algorithm sources from the bottom characteristics of the targets, so the algorithm is robust. What＇s more, the particle filter could solve non-linear and non-Gaussian state estimation. In a word, the algorithm could deal with the trace management of the multi-targets tracking, and cope with appearing and disappearing and combining and dividing of the targets, and partial occlusions then recovering the tracks after temporary loss. The experimental results show that the new algorithm could accomplish the multi-targets tracking in complex image sequences.

关 键 词：显著性 目标特征 粒子滤波 多目标跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]