摘要
将行人检测算法和行人重识别算法相结合,提出一种多目标跨摄像头跟踪算法,该算法由行人检测、行人重识别和行人数据关联三大模块组成.首先利用基于YOLOv3的行人检测改进算法检测视频中出现的行人,并保存视频号、帧号和行人的全身位置信息;其次,利用基于生成对抗网络和重排序的行人重识别改进算法,为已检测的行人图片赋予一个标签;最后整合前两步得到的行人信息,生成包含视频中所有行人信息的JSON(JavaScript Object Notation)文件.该算法可以快速、高效地完成多目标跨摄像头跟踪任务,有一定实用价值,在南京市举办的全球人工智能应用大赛中获得了单项奖.
Combining the person detection algorithm and the person re⁃identification algorithm,a multi⁃target cross camera tracking algorithm is proposed.The algorithm consists of three modules:person detection,person re⁃identification and person data association.Firstly,the improved person detection algorithm based on YOLOv3 is used to detect the person appearing in the video and to save the video number,frame number and person body position information.Secondly,the improved person re⁃identification algorithm based on generative adversarial network and reranking is used to assign a label to the detected person images.Finally,the person information obtained in the first two steps is integrated to generate a JavaScript Object Notation(JSON)file containing all information of the person in the video.The algorithm can complete the multi⁃target cross camera tracking task quickly and efficiently,and has certain practical value,which won a single award in the Global(Nanjing)Artificial Intelligence Application competition.
作者
戴臣超
王洪元
曹亮
殷雨昌
张继
Dai Chenchao;Wang Hongyuan;Cao Liang;Yin Yuchang;Zhang Ji(School of Computer Science and Artificial Intelligence Aliyun School of Big Data,Changzhou University,Changzhou,213164,China)
出处
《南京大学学报(自然科学版)》
CAS
CSCD
北大核心
2021年第2期227-236,共10页
Journal of Nanjing University(Natural Science)
基金
国家自然科学基金(61976028,61572085,61806026,61502058)
江苏省自然科学基金(BK20180956)。
关键词
多目标跨摄像头跟踪
行人重识别
行人检测
重排序
生成对抗网络
multi⁃target cross camera tracking
person re⁃identification
person detection
reranking
generative adversarial network