基于身份导向自监督表示学习的智能寻人系统
Identity-seeking self-supervised representation learning for a smart person-tracking system
夏冉 1雷晓艳 1郭梦晴 1王文韬1
作者信息
- 1. 周口师范学院网络工程学院,河南周口 466000
- 折叠
摘要
行人重识别(person re-identification,Person ReID)技术对安全监控和个人跟踪等领域至关重要,但标注数据的稀缺性和高成本限制了其广泛应用.针对这一问题,设计一种基于身份导向自监督表示(Identity-seeking Self-supervised Representation,ISR)学习方法的寻人系统,能够从大规模无标注视频数据中学习并提取人员的特征表示.系统架构分为三个模块:首先,利用YOLOV8模型对视频流中的人物进行检测,并自动裁剪出galley图片;其次,通过ISR学习方法对galley图片进行特征提取,并构建特征数据库;最后,在特征数据库中,检索与查询图片相似的galley图片,并关联到对应的视频帧.实验结果证明,系统搜索准确高效,具有广泛的应用价值和实用潜力.
Abstract
Person re-identification(Person ReID)technology is crucial for applications such as security monitoring and individual tracking.However,the scarcity and high cost of annotated data limit its widespread use.To address this issue,a person search system based on Identity-seeking Self-supervised Representation(ISR)learning method is designed to learn and extract feature representations of individuals from large-scale unlabeled video data.The system architecture consists of three modules:First,the YOLOV8 model is used to detect individuals in video streams and automatically crop gallery ima-ges;Second,the ISR learning method is employed to extract features from gallery images and construct a feature database;Finally,the system retrieves gallery images similar to the query image from the feature database and associates them with corresponding video frames.Experimental results demonstrate that the system is accurate and efficient in search,with broad application value and practical potential.
关键词
行人重识别/身份导向自监督表示学习/无标注数据学习Key words
person re-identification/identity-seeking self-supervised representation learning/unlabeled data learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据