A Person Re-Identification Method Based on Feature Fusion

Person re-identification (Re-ID) aims to accurately retrieve specific pedestrians from large-scale image databases and is widely applied in public safety and intelligent surveillance. Traditional methods often fail to fully capture the relationships between local features. To address this, we propose a novel Re-ID method based on feature fusion. The approach uses ResNet50 as the backbone for feature extraction and incorporates both global and local feature extraction branches. In the global branch, hybrid pooling (combining global average and maximum pooling) enhances feature expressiveness. In the local branch, the feature maps are vertically segmented, and adjacent local features are fused for a more precise representation. Experiments on the Market1501 dataset show the proposed model achieves 87.7% mAP and 94.7% Rank-1 accuracy, outperforming existing methods, demonstrating its effectiveness in Re-ID tasks.