Identification of Changing Personnel with Double-Layer Network Fusion and Bi-Level Monitoring Mechanism

Abstract

Person re-identification (Re-ID) is an essential part of visual surveillance that aims to identify and locate persons from multiple network cameras without conflicting viewpoints. Although significant advances have been made in recent years with the use of deep learning, there are still many challenges in vision such as occlusion, exposure, background clutter, misalignment, scale, perspective, low resolution and illumination, and cross-camera methods. Dressing redefinition is a hot topic in education right now. Most existing methods assume that people's clothes do not change in a short time, but they do not apply when people change clothes. Accordingly, this article introduces a double-layer garment changer re-identification network that integrates the secondary care process through clustering and fine-grained knowledge in space and training the garment classification branch to increase the sensitivity of the network to garment characteristics. In this method, auxiliary equipment such as human bone is not used and the complexity of the model is greatly reduced compared to other methods. This article runs experiments on the famous redefined PRCC data and large-scale long-term dataset (LaST). Experimental results show that the method in this article is superior to existing methods.