Pedestrian attribute recognition: A survey

PATTERN RECOGNITION

卷:121

Article

Wang, Xiao^1,3  Zheng, Shaofei¹  Yang, Rui¹  Zheng, Aihua²  Chen, Zhe⁴  Tang, Jin¹  Luo, Bin¹ 

[1] Anhui Univ, Sch Comp Sci & Technol, Hefei, Peoples R China

[2] Anhui Univ, Sch Artificial Intelligence, Hefei, Peoples R China

[3] Peng Cheng Lab, Shenzhen, Peoples R China

[4] Univ Sydney, Sch Comp Sci, Fac Engn, Sydney, NSW, Australia

关键词: Pedestrian attribute recognition;    Multi-label learning;    Multi-task learning;    Deep learning;    CNN-RNN;   

DOI  :  10.1016/j.patcog.2021.108220

来源: Elsevier

【 摘 要 】

Pedestrian Attribute Recognition (PAR) is an important task in computer vision community and plays an important role in practical video surveillance. The goal of this paper is to review existing works using traditional methods or based on deep learning networks. Firstly, we introduce the background of pedestrian attribute recognition, including the fundamental concepts and formulation of pedestrian attributes and corresponding challenges. Secondly, we analyze popular solutions for this task from eight perspectives. Thirdly, we discuss the specific attribute recognition, then, give a comparison between deep learning and traditional algorithm based PAR methods. After that, we show the connections between PAR and other computer vision tasks. Fourthly, we introduce the benchmark datasets, evaluation metrics in this community, and give a brief performance comparison. Finally, we summarize this paper and give several possible research directions for PAR. The project page of this paper can be found at: https://sites.google.com/view/ahu-pedestrianattributes/ . (c) 2021 Elsevier Ltd. All rights reserved.

【 授权许可】

Free