LAUSR

Person re-identification is a challenging research issue in computer vision and has a broad application prospect in intelligent security. In recent years, with the emergence of large-scale person datasets and the rapid development of deep learning, many outstanding results have been achieved in person re-identification researches, which mainly involves two critical technologies: feature extraction and distance metric. Among them, feature extraction has been well summarized in the current literature of person re-identification, but there is no systematic analysis of the distance metric method in the current review literature. However, effective and reliable distance metric is crucial to improve the accuracy of person re-identification. Therefore, it is necessary to systematically review and summarize the metric learning methods in person re-identification, so as to provide some references for the researchers of metric learning. In this paper, we make a comprehensive analysis of metric learning methods in the past five years, which can be summarized into three aspects: distance metric method, metric learning algorithm, and re-ranking for the metric results. Then, we compare the performance of some representative metric learning methods and discuss them in-depth. Finally, we make a prospect for the future research direction of metric learning in person re-identification.