• 首页
  • 期刊简介
  • 编委会
  • 投稿指南
  • 收录情况
  • 杂志订阅
  • 联系我们
引用本文:沈江霖,魏 丹,罗一平.基于姿态交换图像生成的行人重识别[J].软件工程,2023,(7):29-33.【点击复制】
【打印本页】   【下载PDF全文】   【查看/发表评论】  【下载PDF阅读器】  
←前一篇|后一篇→ 过刊浏览
分享到: 微信 更多
基于姿态交换图像生成的行人重识别
沈江霖, 魏 丹, 罗一平
(上海工程技术大学机械与汽车工程学院, 上海 201620)
goatlmljrf@163.com; weidan@sues.edu.cn; lyp777@sina.com
摘 要: 不同摄像设备之间存在角度、分辨率等差异,同时行人兼具刚性和柔性的特性且外观易受穿着、姿态、遮挡物和视角等因素的影响。基于此,文章从生成对抗网络与姿态特征等方面对行人重识别问题展开深入研究,提出了一种姿态可交换行人重识别框架(PSG-Net)。该框架将样本中的每个行人编码为姿态代码,视觉代码,通过切换姿态代码,生成高质量的姿态合成图像。在Market-1501、DukeMTMC-reID和CUHK03数据集上的实验结果表明,该方法实现了识别性能改进,并在Market-1501数据集上的排序第一(rank-1),结果能达到95.1%,优于大多数先进的方法。
关键词: 行人重识别;生成对抗网络;图像生成
中图分类号: TP181    文献标识码: A
基金项目: 国家自然科学基金项目(62101314).
Pedestrian Re-identification Based on Pose-switched Image Generation
SHEN Jianglin, WEI Dan, LUO Yiping
(School of Mechanical and Automobile Engineering, Shanghai University of Engineering Science, Shanghai 201620, China)
goatlmljrf@163.com; weidan@sues.edu.cn; lyp777@sina.com
Abstract: There are differences in angle and resolution between different camera devices, and pedestrians have both rigid and flexible characteristics, and their appearance is easily affected by wearing, posture, obstruction, and perspective. Based on this, this paper conducts in-depth research on pedestrian recognition from the aspects of Generative adversarial network and attitude characteristics, and proposes a gesture exchangeable pedestrian recognition framework (PSG-Net). This framework encodes each pedestrian in the sample as pose code and visual code, generate high-quality pose synthesis images by switching pose codes. The experimental results on Market-1501, DukeMTMC reID, and CUHK03 datasets show that the proposed method has achieved recognition performance improvement, with ranking first (rank-1) results reaching 95.1% on Market-1501 dataset, which is superior to most advanced methods.
Keywords: pedestrian re-identification; generative adversarial network; image generation


版权所有:软件工程杂志社
地址:辽宁省沈阳市浑南区新秀街2号 邮政编码:110179
电话:0411-84767887 传真:0411-84835089 Email:semagazine@neusoft.edu.cn
备案号:辽ICP备17007376号-1
技术支持:北京勤云科技发展有限公司

用微信扫一扫

用微信扫一扫