摘要
针对移动机器人视觉目标检测与跟踪任务,首先,阐述了其研究背景及意义,分析了当前目标检测与跟踪方法研究中存在的难点;其次,结合特征表达的类型和检测步骤对传统目标检测算法、基于候选区域的目标检测算法、基于回归的目标检测算法与基于增强学习的目标检测算法等4种目标检测算法的优缺点进行了分析比较;接着,分析比较了传统跟踪算法、基于稀疏表示的跟踪算法、基于相关滤波的跟踪算法与基于深度学习的跟踪算法等4个阶段视觉跟踪算法的性能;最后,总结了现阶段目标检测与跟踪方法存在的局限性,并指出未来进一步探索的方向。
Aiming at the visual target detecting and tracking performed by a mobile robot,the background and significance of the research are expounded,and the difficulties in the research are analyzed.Then,according to the types of feature representation and the detecting steps,a comparative analysis is made on the advantages and disadvantages of the following four types of target detection algorithms,i.e.,traditional target detection algorithms,candidate region based target detection algorithms,regression-based target detection algorithms and reinforcement learning based target detection algorithms.Further,a comparative analysis is conducted on the performance of the following visual tracking algorithms in four stages,i.e.,traditional tracking algorithms,sparse representation based tracking algorithms,correlation filter based tracking algorithms and deep learning based tracking algorithms.Finally,the limitations of current methods are summarized,and the directions for future improvement are pointed out.
作者
刘文汇
巢渊
唐寒冰
徐鹏
LIU Wenhui;CHAO Yuan;TANG Hanbing;XU Peng(Jiangsu University of Technology,Changzhou 213000,China)
出处
《电光与控制》
CSCD
北大核心
2022年第4期59-67,88,共10页
Electronics Optics & Control
基金
国家自然科学基金(51905235)
江苏省自然科学基金(BK20191037)
常州市科技计划(CJ20190069)
江苏省研究生实践创新计划(SJCX20_1045)
江苏理工学院研究生实践创新计划(XSJCX20_32)。
