一种基于M-APF的实时UAV目标跟踪算法  

Real Time UAV Target Tracking Method Based on M-APF

作　　者：席庆彪[1,2] 董阳霞[1] 刘慧霞[2] 

机构地区：[1]西北工业大学自动化学院,西安710072 [2]西北工业大学第365研究所,西安710065

出　　处：《系统仿真学报》2014年第1期107-111,118,共6页Journal of System Simulation

基　　金：国家自然科学基金(61074155)

摘　　要：针对中小型UAV(Unmanned Aerial Vehicle)平台运动目标跟踪系统由于受机载摄像条件及数据传输时延等因素,造成跟踪结果实时性不高、目标被遮挡时易丢失的问题,提出一种基于M-APF(Meanshift-Auxiliary Particle Filter)的UAV运动目标跟踪算法。该方法采用APF作为跟踪算法的主体框架,同时引入Meanshift计算少量辅助采样粒子的偏移,并将其移动到观测值的局部最优位置,解决了APF算法计算量大的问题,提高算法实时性。仿真结果显示:算法满足UAV平台运动目标跟踪要求,实时性及鲁棒性优于Meanshift和APF等算法。Focusing on the problems of the target information estimated delay and the target lost for the occlusion reasons in the UAV moving target tracking system, a novel target tracking method based on M-APF （Meanshift-Auxiliary Particle Filter） was proposed. It adopted the APF as the main framework of the tracking algorithm, then the Meanshift was applied to calculate the offset of a few auxiliary particle and moved them to the local optimum position of the observed values. As a result, it solved the large calculation of the APF algorithm, and improved the real-time of the method. Simulation results show that the calculation cost and robustness of the M-APF is superior to Meanshift and APF algorithm, and satisfies the UAV target tracking requirements.

关 键 词：无人机 目标跟踪 均值漂移 辅助粒子滤波 遮挡 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]