基于仿生探测的大视场小目标跟踪算法  

A Tracking Algorithm for Small Target in Large FoV Based on Biological Detecting

作　　者：寇巍巍 李东光[1] 张骢 KOU Weiwei;LI Dongguang;ZHANG Cong(The Key Laboratory of Electromechanical Dynamic Control, Beijing Institute of Technology, Beijing 100081, China)

机构地区：[1]北京理工大学机电动态控制重点实验室,北京100081

出　　处：《探测与控制学报》2017年第6期46-50,55,共6页Journal of Detection & Control

摘　　要：针对光学成像引信远距离探测目标时视场大、成像目标小,目标难以探测与跟踪的问题,提出了基于仿生探测的大视场小目标跟踪算法。该算法模仿鹰眼视觉原理,设计基于鹰眼视觉通路原理的探测器,通过提取颜色、亮度及方向特征构建目标显著图,实现对目标区域的粗定位,并根据探测器的探测结果,结合目标方向梯度直方图特征与贝叶斯分类器实现小目标的精确定位与跟踪。仿真实验表明,对装甲车与坦克模型分别进行跟踪,跟踪窗口重叠率分别为79.3%与79.0%,跟踪效果优于Mean-Shift算法,该算法可有效解决大视场下小目标的探测与跟踪问题。When imaging fuze carry out long distance detection,the target's image is small and the FoV(Field of View)is large which leads to the failure of detection.In order to solve this problem,a new tracking algorithm based on biological detecting for small target in large FoV was proposed,which imitated the mechanism of eagle's vision.The detector based on the visual pathway of eagle was established to locate the object,which extract the image's features of color,luminance and orientation to build the visual saliencymap;The tracker based on the HOG feature was established to track the small object in the large FoV,which combined the HOG with Nave Bayes.The results of experiments showed that the tracking window overlap rate of tank model was79.3%and panzer model's tracking window overlap rate was79.0%,which meant that the proposed algorithm was more effective and accurate than MeanShift,and it could detect and track the small target in large FoV efficiently.

关 键 词：成像引信 仿生探测 小目标跟踪 显著图 形态学滤波 

分 类 号：TJ430.1[兵器科学与技术—火炮、自动武器与弹药工程]