Person Reidentification Model Based on Multiattention Modules and Multiscale Residuals

Yongyi Li,Di Fan,Shuang Dong,Changzhi Lv,Shiqi Wang,Xueling Lv,Rui Wang

doi:10.1155/2021/6673461

Yongyi Li, Di Fan + Show 5 more

Open Access

https://doi.org/10.1155/2021/6673461

Copy DOI

Abstract

At present, person reidentification based on attention mechanism has attracted many scholars’ interests. Although attention module can improve the representation ability and reidentification accuracy of Re-ID model to a certain extent, it depends on the coupling of attention module and original network. In this paper, a person reidentification model that combines multiple attentions and multiscale residuals is proposed. The model introduces combined attention fusion module and multiscale residual fusion module in the backbone network ResNet 50 to enhance the feature flow between residual blocks and better fuse multiscale features. Furthermore, a global branch and a local branch are designed and applied to enhance the channel aggregation and position perception ability of the network by utilizing the dual ensemble attention module, as along as the fine-grained feature expression is obtained by using multiproportion block and reorganization. Thus, the global and local features are enhanced. The experimental results on Market-1501 dataset and DukeMTMC-reID dataset show that the indexes of the presented model, especially Rank-1 accuracy, reach 96.20% and 89.59%, respectively, which can be considered as a progress in Re-ID.

Highlights

As an important video intelligent analysis technology, person reidentification (Re-ID) uses computer vision technology to realize the identification and matching of target pedestrians in a multicamera network with nonoverlapping fields of view. at is, given a pedestrian test image (Probe), the pedestrian image is retrieved under crossmonitoring equipment from all gallery images (Gallery) [1]
We designed a person reidentification model based on attention fusion and multiscale residuals. e model mainly solves the problems of poor coupling between attention mechanism and original network and scientific expression of local features. e model uses an improved ResNet50 as the backbone network and is designed with global and local branch structures. e paper added Combined Attention Fusion Module (CAFM) and Multiscale Residual Fusion Module (MSFM) to the original ResNet50 [18] to effectively concatenate the feature information between residual blocks and better integrate multiscale features. e global branch uses a Dual Ensemble Attention Module (DEAM) to enhance the network’s channel aggregation and location awareness capabilities. e local branch is divided into finegrained features by multiproportion block method to further refine the local features
Multiscale Residual Fusion Module is a dynamic selection mechanism that enables each neuron to select different receptive fields according to the size of the target feature [21]. e Multiscale Residual Fusion Module structure designed in this paper is shown in Figure 4(b), which is mainly divided into two parts: multiscale feature extraction and feature selective fusion

Summary

Research Article

Person Reidentification Model Based on Multiattention Modules and Multiscale Residuals. Person reidentification based on attention mechanism has attracted many scholars’ interests. Attention module can improve the representation ability and reidentification accuracy of Re-ID model to a certain extent, it depends on the coupling of attention module and original network. A person reidentification model that combines multiple attentions and multiscale residuals is proposed. E model introduces combined attention fusion module and multiscale residual fusion module in the backbone network ResNet 50 to enhance the feature flow between residual blocks and better fuse multiscale features. A global branch and a local branch are designed and applied to enhance the channel aggregation and position perception ability of the network by utilizing the dual ensemble attention module, as along as the fine-grained feature expression is obtained by using multiproportion block and reorganization. A global branch and a local branch are designed and applied to enhance the channel aggregation and position perception ability of the network by utilizing the dual ensemble attention module, as along as the fine-grained feature expression is obtained by using multiproportion block and reorganization. us, the global and local features are enhanced. e experimental results on Market-1501 dataset and DukeMTMC-reID dataset show that the indexes of the presented model, especially Rank-1 accuracy, reach 96.20% and 89.59%, respectively, which can be considered as a progress in Re-ID

Introduction

Global and local features extraction network

Matrix Softmax operation

Extract γ

Method

Conclusions