Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (7): 126-133.DOI: 10.3778/j.issn.1002-8331.2110-0376

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Research on Person Re-Identification Based on Background Adaptive Learning

HE Ruhan, XIONG Jiefan, XIONG Mingfu   

  1. 1.Engineering Research Center of Hubei Province for Clothing Information, Wuhan 430200, China
    2.Hubei Provincial Engineering Research Center for Intelligent Textile and Fashion, Wuhan 430200, China
    3.School of Computer Science and Artificial Intelligence, Wuhan Textile University, Wuhan 430200, China
  • Online:2023-04-01 Published:2023-04-01

基于背景自适应学习的行人重识别算法研究

何儒汉,熊捷繁,熊明福   

  1. 1.湖北省服装信息化工程技术研究中心,武汉 430200
    2.纺织服装智能化湖北省工程研究中心,武汉 430200
    3.武汉纺织大学 计算机与人工智能学院,武汉 430200

Abstract: The research of person re-identification based on human semantic information has become one of the most prevalent research highlights in recent years. However, most of the existing research based on this still stay in how to extract more accurate semantic information, ignoring the relationship between the semantic information of human body itself and the semantic information of human body and environment. In order to solve the problem, this research proposes a human semantic spatial relationship model based on background adaptive learning. The model can be mainly divided into three parts:semantic separation, rough feature extraction and spatial relation learning. Semantic separation is mainly used to distinguish human semantic information from environmental semantic information, rough feature extraction is mainly used to extract shallow features of different semantic information, and spatial relation learning is mainly used to enhance the above shallow features at the spatial relation level. Extensive experiments show that the method has achieved good results in two groups of public data sets(DukeMTMC-reID, CUHK-03).

Key words: person re-identification, semantic segmentation, spatial relationship

摘要: 现有的基于语义分割的行人重识别研究大多还是停留在人体语义信息的提取本身,忽视了人体自身语义信息之间以及人体与环境语义信息之间的相互关系,为了解决这一问题,此项研究提出了基于背景自适应学习的人体语义空间关系模型。该模型主要分为语义分离,特征粗提取以及空间关系学习三部分,语义分离主要用于区分人体语义信息和环境语义信息,特征粗提取则是用于提取不同语义信息的浅层特征,空间关系学习主要是对上述的浅层特征进行空间关系维度的特征关联。通过广泛的实验证明,该方法在两组公开数据集中(DukeMTMC-reID、CUHK-03)均取得了较好的效果。

关键词: 行人重识别, 语义分割, 空间关系