A Survey on Metric Learning for Object Re-identification in Intelligent Surveillance

Abstract

In recent years, object re-identification has been a significant topic in computer vision, which is expected in the application of artificial intelligent surveillance. To tackle the problem, metric learning has become an optimal method, and the approaches are concerned with learning a distance function tuned to a particular identification task, and have been shown to be useful in conjunction with nearest-neighbor methods and other techniques. This paper provides a survey of existing metric learning approaches for object re-identification, which focuses on the methods based on the application of metric learning. The detailed description of methods is presented for comparison, and several related main datasets are briefly introduced. In this paper, all papers are published ranging from 2013 to 2016, which provides an overview of progress in this area.