监控场景中的行人属性识别研究综述

Pedestrian Attribute Recognition in Surveillance Scenes:A Survey

监控场景中的行人属性识别任务旨在为监控场景中视频摄像头捕捉的行人图片预测其属性类别,由于监控场景环境的复杂以及行人属性的细粒度标签,监控场景中的行人属性识别任务极具挑战,受到业界和学界的广泛关注.文中对监控场景中的行人属性识别研究进展进行梳理,首先给出了其概念范畴与任务定义,并与其他相似的属性识别任务进行对比.其次,文中对目前主流的行人属性识别数据库进行了简单介绍,并从图片和标注两个角度分析了不同数据库之间的异同.再次,文中对深度学习时代以来所提出的各种行人属性识别方法进行了归纳和总结,综述了目前行人属性识别领域的研究现状.最后,文中对监控场景中的行人属性识别存在的问题进行了思考和讨论,并对未来的发展趋势进行了展望.

The task of pedestrian attribute recognition in surveillance scenes aims to predict the attribute categories of pedestrian images captured by video cameras in surveillance scenes.Due to the complexity of the surveillance scene environment and the fine-grained labeling of pedestrian attributes,the task of pedestrian attribute recognition in surveillance scenes is extremely challenging and has received extensive attention from the industry and academia.This paper reviews the research progress of pedestrian attribute recognition in surveillance scenes.Firstly,we give its conceptual scope and task definition and compare it with similar attribute recognition tasks.Secondly,this paper briefly introduces the mainstream pedestrian attribute recognition datasets and analyzes the similarities and differences between different datasets from picture and annotation perspectives.Again,this paper summarizes and concludes the various pedestrian attribute recognition methods proposed since the era of deep learning and reviews the current research status in pedestrian attribute recognition.Finally,this paper considers and discusses the problems of pedestrian attribute recognition in surveillance scenes and provides an outlook on the future development trend.