基于粒子滤波的空-地目标跟踪算法  

Particle filter based algorithm for tracking ground target in airborne environment

作　　者：宋策[1,2] 张葆[1] 尹传历[1] 王超[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,航空光学成像与测量重点实验室,吉林长春130033 [2]中国科学院大学,北京100039

出　　处：《光电子．激光》2013年第10期2017-2023,共7页Journal of Optoelectronics·Laser

基　　金：国家高技术研究发展计划(863计划)(2008AA121803)资助项目

摘　　要：针对空-地目标跟踪中目标大幅度变速运动而引起的跟踪失败问题,基于Kristan等人提出的双步(TS)动态模型框架,对空-地目标跟踪中目标运动特点进行分析与建模,改进TS模型中的保守模型以适应加速运动,提出适于描述大幅度变速运动的加速度双步(TSA)动态模型作为粒子滤波(PF)跟踪算法的动态模型,实现对粒子状态的精确预测,进而达到使用较少粒子即可对目标鲁棒跟踪的目的。对空-地目标跟踪的测试视频进行测试,结果表明,本文算法可对大幅度变速运动目标稳定跟踪,正确跟踪率为92%,对目标尺寸约为25pixel×30pixel时的处理帧率为29frame/s。本文算法具有较好的鲁棒性与实时性。For the tracking failing caused by target moving in large scale variable velocity when tracking ground target with the camera in the airborne environment, a two-stage acceleration （TSA） dynamic mode particle filter tracking algorithm is proposed. Based on the framework of the two-stage （TS） dy- namic model proposed by Kristan et al. , the motion features of ground target which is tracked with the camera in the airborne environment are analyzed and modeled. The conservative model of the two-stage （TS） dynamic model is improved to accommodate the accelerated motion. The two-stage acceleration dy- namic model which can describe target moving in large scale variable velocity is proposed as the dynamic model of particle filter tracking algorithm,the states of particles can be predicted precisely,and then ro- bust tracking can be realized using less particles. The proposed algorithm is tested with video sequences for tracking ground target with the camera in airborne environment, and the results show that the pro- posed algorithm can track ground target moving with large scale variable velocity stability, the accurate rate is 92%, and the average computing frame rate is 29 frame/s when the object scale is about 25 pixel X 130 pixel. The algorithm proposed in this paper has better robustness and real-time quality.

关 键 词：粒子滤波(PF) 动态模型 目标跟踪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]