A Pedestrian Re-Identification Method Integrating ResNet and Residual Self-Attention Mechanism

Person Re-identification is an important task in the field of computer vision, which aims to identify the same person across cameras and scenes. However, due to the influence of perspective changes, illumination differences, occlusion and other factors, person re-identification still faces great challenges. In order to solve these problems, this paper proposes a deep learning method combining ResNet and residual self-attention mechanism to improve the robustness and feature expression ability of the model for complex scenes. In this method, the ResNet-18 network is used as the backbone, and a self-attention module is embedded in each residual block to enhance the network's ability to perceive global context information by learning the relationship between different positions in the feature map. At the same time, the output of the self-attention module is fused by residual connection, which can not only effectively avoid gradient disappearance, but also retain the original feature information and promote the training of the deep network. Experimental results on the public dataset Market-1501 demonstrate that the proposed method achieves mAP and accuracy scores of 95.21% and 93.01%, respectively, outperforming existing methods. These results confirm the effectiveness of the proposed approach in the person reidentification task.