北京航空航天大学学报 ›› 2020, Vol. 46 ›› Issue (9): 1747-1755.doi: 10.13700/j.bh.1001-5965.2020.0075

• 论文 • 上一篇    下一篇

基于空间注意力机制的行人再识别方法

张子昊1,2, 周千里1,2,3, 王蓉1,2   

  1. 1. 中国人民公安大学 警务信息工程与网络安全学院, 北京 100038;
    2. 安全防范技术与风险评估公安部重点实验室, 北京 100038;
    3. 北京市公安局, 北京 100740
  • 收稿日期:2020-03-03 发布日期:2020-09-22
  • 通讯作者: 王蓉 E-mail:dbdxwangrong@163.com
  • 作者简介:张子昊 男,硕士研究生。主要研究方向:模式识别、人工智能;周千里 男,博士研究生。主要研究方向:模式识别、人工智能;王蓉 女,博士,教授,博士生导师。主要研究方向:模式识别、人工智能。
  • 基金资助:
    国家重点研发计划(A19808);中央高校基本科研业务费专项资金(2019JKF111)

Pedestrian re-identification method based on spatial attention mechanism

ZHANG Zihao1,2, ZHOU Qianli1,2,3, WANG Rong1,2   

  1. 1. School of Police Information Engineering and Network Security, People's Public Security University of China, Beijing 100038, China;
    2. Key Laboratory of Security Technology&Risk Assessment, Beijing 100038, China;
    3. Beijing Public Security Bureau, Beijing 100740, China
  • Received:2020-03-03 Published:2020-09-22
  • Supported by:
    National Key R & D Program of China (A19808); the Fundamental Research Funds for the Central Universities (2019JKF111)

摘要: 行人再识别是图像检索领域的一个重要部分,但是由于行人姿态各异、背景复杂等因素,导致提取到的行人特征鲁棒性和代表性不强,进而影响行人再识别的精度。在AlignedReID++算法基础上,提出了基于空间注意力机制的行人特征提取方法,应用在行人再识别中取得了很好的效果。首先,在特征提取部分,引入空间注意力机制来增强特征表达,同时抑制可能的噪声;其次,通过在卷积层中引入实例正则化层(IN)来辅助批正则化层(BN)对特征进行归一化处理,解决单一BN层对特征色调变化以及光照变化的不敏感性,提高特征提取对亮度、色调变化的鲁棒性;最后,在Market1501、DukeMTMC和CUHK03 3个行人再识别通用数据集上对所提改进模型进行测试评价。实验结果显示:改进后的模型在3个数据集上识别精度分别提升了2%、2.9%和5.1%,表明改进后的模型相较于改进前的模型,在精度以及鲁棒性上都有显著提高。

关键词: 深度学习, 空间注意力机制, 行人特征, 特征增强, 卷积神经网络

Abstract: Pedestrian re-identification has always been an important part of image retrieval. However, due to different pedestrian poses and complex backgrounds, the extracted pedestrian features are not robust and representative, which in turn affects the accuracy of pedestrian re-recognition. In this paper, based on AlignedReID++ algorithm, we proposes a pedestrian re-identification method based on spatial attention mechanism. First,in the feature extraction part, a spatial attention mechanism is introduced to enhance feature expression while suppressing possible noise. Second, the Instance-Normalization (IN) layer is introduced in the convolution layer to assist the Batch-Normalization (BN) layer to normalize the features and to solve the problem of single BN layer insensitivity to feature tonal and illumination changes, which enhances the robustness of feature extraction to tonal and illumination changes. Finally, to validate the proposed method, extensive experiment has been carried out on the Market1501, DukeMTMC, and CUHK03 pedestrian re-identification datasets. The experimental results show that the recognition accuracy of the improved model on the three datasets has been improved by 2%, 2.9%, and 5.1%, respectively, compared with model before modification, which indicates that the proposed method achieves higher accuracy and more robustness.

Key words: deep learning, spatial attention mechanism, pedestrian characteristics, feature enhancement, convolutional neural network

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发