基于多策略融合的快速核相关滤波目标跟踪算法  

Fast Kernel Correlation Filter Target Tracking Algorithm Based on Multi-Strategy Fusion

作　　者：张晗 康国华[1] 张琪 邱钰桓 张雷[2] Zhang Han;Kang Guohua;Zhang Qi;Qiu Yuhuan;Zhang Lei(Academy of Astronautics,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China;China Xi’an Satellite Control Center,Xi'an 710043,China)

机构地区：[1]南京航空航天大学航天学院,南京210016 [2]西安卫星测控中心,西安710043

出　　处：《航天控制》2020年第3期31-38,共8页Aerospace Control

基　　金：空间智能控制技术重点实验室开放基金资助项目(KGJZDSYS-2018-07)。

摘　　要：针对核相关滤波算法(Kernel Correlation Filter,KCF)对快速运动目标跟踪精度较低、实时性较差的问题,提出多策略融合的快速核相关滤波(Multistrategy KCF,MSKCF)算法。该算法基于KCF框架,融合多个策略,将Faster Regin-CNN网络结构、特征极差、滤波尺度因子引入目标图像识别窗口标定和尺寸自适应更新,解决了识别窗口与目标大小不适应的问题,实现了自动跟踪。本文采用北斗导航卫星模型进行了验证,结果表明MSKCF可以自主获得初始跟踪窗口,目标跟踪精度与速度均有所提升。A multi-strategy KCF(MSKCF)algorithm with multi-strategy fusion is proposed for the kernel correlation filter(KCF),which has low tracking precision and low real-time performance.The algorithm is based on the KCF framework and integrated with multiple strategies,the Faster RCNN network structure,feature range and filter scale factor are introduced into the target image recognition window calibration and size adaptive update,which solves the problem that the recognition window and the target size are not suitable and realize automatic tracking.In this paper,The Beidou navigation satellite model is used to verify the results.The results show that MSKCF can obtain the initial tracking window autonomously and the target tracking accuracy and speed are improved.

关 键 词：图像识别 核相关滤波 Faster Rcnn 多策略融合 目标跟踪 

分 类 号：V448.25[航空宇航科学与技术—飞行器设计]