基于可变形掩膜对齐卷积模型的行人再识别  

Person Re-identification Based on Deformable Mask Alignment Convolution Model

作　　者：刘畅[1] 邱卫根[1] 张立臣[1] LIU Chang;QIU Weigen;ZHANG Lichen(School of Computers,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学计算机学院,广州510006

出　　处：《计算机工程与应用》2021年第5期146-152,共7页Computer Engineering and Applications

基　　金：国家自然科学基金(61572142)。

摘　　要：行人再识别是计算机视觉领域的一个重要研究方向,在视频监控等非常广阔的领域有极其重要的应用前景。行人再识别研究中遇到的一个重要挑战就是行人图像对齐问题。利用全卷积模型和全局平均池化操作,提出了一种新的可变形掩膜对齐的深度卷积神经网络模型,它不仅可以解决行人图像对齐问题,而且实现了行人图像的多信息融合。该方法在Market-1501和DukeMTMC-reID两大数据集上进行了验证,整体准确率得到了很大提高。Person re-identification is an important research direction in the field of computer vision,and has extremely important application prospects in very wide fields such as video surveillance.An important challenge in person re-identification research is the problem of person image alignment.In this paper,a new deformable mask aligned deep convolutional neural network model is proposed using the fully convolutional network model and global average pooling operation.It not only solves the problem of person image alignment,but also implements multi-information fusion of person images.The method in this paper is verified on the two large data sets of Market-1501 and DukeMTMC-reID,and the overall accuracy rate has been greatly improved.

关 键 词：行人再识别 对齐 全卷积模型 信息融合 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]