Global-local Feature Enhancement for Occluded Person Re-identification

Artificial Intelligence is rapidly evolving and Person re-identification techniques stand out. Although some occluded person re-identification methods have high recognition capabilities, they tend to have less accurate feature extraction under severe occlusion, which affects the model’s recognition capability to some extent. To address this problem, we propose a global-local feature enhancement network (GLFE). Firstly, we propose a local feature extraction module (LFEM), which can obtain the key point information of a person through the openpose pose estimation model. Secondly, a global attention feature obtaining module (GAFOM) is proposed, which can effectively extract local features of the person and dynamically classify the importance weights of the pedestrian through the attention mechanism, so as to reduce the interference of some noise features to the model. Finally, we feature splice LFEM with GAFOM to obtain the final pedestrian feature vector. This process enhances the concentration on the region of differentiating pedestrians. Extensive experimental results have demonstrated that the proposed GLFE achieves superior identification performance compared to several state-of-the-art methods.