基于后验概率度量的粒子滤波跟踪算法研究  

Particle filter algorithm based on posterior probability measurement

作　　者：李向军[1] 李良福[2] 

机构地区：[1]海军驻西安二十所军事代表室,陕西西安710068 [2]西安应用光学研究所,陕西西安710065

出　　处：《应用光学》2011年第4期646-651,共6页Journal of Applied Optics

基　　金：国家863计划(2006AA04Z222);中国博士后基金特别资助(200902593)

摘　　要：针对遮挡、光照变化、尺度变化等复杂环境中的视觉跟踪问题,提出一种基于后验概率度量的粒子滤波跟踪算法。由于后验概率指标与Bhattacharyya系数指标相比具有更强的峰值特性,采用后验概率指标作为相似性度量函数,通过粒子的更新、推广、观测、估计等步骤实现跟踪算法。通过对实际视频图像序列进行目标跟踪实验,实验结果表明:传统算法只有约50%的图像能够实现尺度自适应,而本文算法采用传统算法25%的粒子就能够收敛逼近目标的真实轨迹,达到更强的抗遮挡能力,90%以上的图像序列都能够实现良好的尺度自适应效果。To meet visual tracking requirements in complex environment where obscuration,illumination change and size variation may occur,this paper presents a particle filter algorithm based on posterior probability measurement.Compared with Bhattacharyya coefficient similarity measurement index,the posterior probability measurement index has stronger peak value characteristic.This paper uses the posterior probability index as similarity measurement function,and realizes the tracking algorithm by particle update,propagation,observation and estimation.The video image sequences were tested for object tracking.The experimental results show that only 50% of the image sequences can be scale-adaptive by the traditional algorithm,while this algorithm can converge to the real contrail of object by 25% particles of the traditional algorithm,the obscuration resistant capability is improved,and 90% of image sequences can get scale-adaptive effect.

关 键 词：后验概率 BHATTACHARYYA系数 粒子滤波 视觉跟踪 

分 类 号：TN29[电子电信—物理电子学] TP391[自动化与计算机技术—计算机应用技术]