摘要
行人重识别是智能视频分析领域的研究热点,得到了学术界的广泛重视。行人重识别旨在非重叠视角域多摄像头网络下进行的行人匹配,即确认不同位置的摄像头在不同的时刻拍摄到的行人目标是否为同一人。本文根据研究对象的不同,将目前的研究分为基于图像的行人重识别和基于视频的行人重识别两类,对这两类分别从特征描述、度量学习和数据库集3个方面将现有文献分类进行了详细地总结和分析。此外,随着近年来深度学习算法的广泛应用,也带来了行人重识别在特征描述和度量学习方面算法的变革,总结了深度学习在行人重识别中的应用,并对未来发展趋势进行了展望。
The intelligent video analysis method based on pedestrian re-identification has become a research focus in the field of computer vision, and it has received extensive attention from the academic community. Pedestrian re-identifica- tion aims to verify pedestrian identity in image sequences captured by cameras that are orientated in different directions at different times. This current study is classified into two categories: image-based and video-based algorithms. For these two categories, using feature description, metric learning, and various benchmark datasets, detailed analysis is performed, and a summary is presented. In addition, the wide application of deep-learning algorithms in recent years has changed pedestrian re-identification in terms of feature description and metric learning. The paper summarizes the application of deep learning in pedestrian re-identification and looks at future development trends.
出处
《智能系统学报》
CSCD
北大核心
2017年第6期770-780,共11页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金项目(61471201)
关键词
行人重识别
特征表达
度量学习
深度学习
卷积神经网络
数据集
视频监控
pedestrian re-identification
feature representation
metric learning
deep learning
convolutional neural networks
datasets
video surveillance