基于事件相机的无人机目标跟踪算法  

UAV target tracking algorithm based on event camera

作　　者：朱强 王超毅 张吉庆 尹宝才 魏小鹏 杨鑫[1] ZHU Qiang;WANG Chaoyi;ZHANG Jiqing;YIN Baocai;WEI Xiaopeng;YANG Xin(School of Computer Science and Technology,Dalian University of Technology,Dalian 116000,China)

机构地区：[1]大连理工大学计算机科学与技术学院,辽宁大连116000

出　　处：《浙江大学学报（理学版）》2022年第1期10-18,共9页Journal of Zhejiang University（Science Edition）

基　　金：国家自然科学基金面上项目(61972067).

摘　　要：无人机目标跟踪可应用于消防、军事等重要领域,已成为计算机视觉领域热门研究课题之一。现有的无人机目标跟踪算法大多基于传统RGB相机结合深度学习算法,但此类算法一方面无法避免无人机机体抖动造成的运动模糊,另一方面因传统RGB相机在低光照或过曝光场景下成像质量较差,难以跟踪目标,为此提出采用无人机搭载DAVIS事件相机的方法进行目标跟踪。设计了基于事件与灰度图的双模态融合跟踪网络,用Vicon运动捕捉系统制作了无人机视角下的目标跟踪Event-APS 28数据集,实现了在复杂光照场景下对目标物的有效跟踪。The target tracking by unmanned aerial vehicle(UAV)has become a hot research topic in the field of computer vision.UAV target tracking can be applied to fire fighting,military and other important fields.At present,most UAV target tracking algorithms are based on traditional RGB cameras combined with deep learning algorithms.However,such methods are hard to deal with motion blur caused by UAV body jitter.Moreover,traditional RGB cameras have poor performance of imaging on low-illumination or over-exposure scenes.In order to solve the above problems,this paper presents an approach which takes the UAV with DAVIS event camera for target tracking,and designs a dual mode fusion tracking network based on events and grayscale,To better train the network,this paper employs the Vicon motion capture system to make the target tracking data sets:Event-APS 28 under the UAV perspective,ensuring that the target can be tracked effectively according to the image information even in complex lighting scenes.

关 键 词：目标跟踪 事件相机 双模态融合跟踪网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]