Research on Pedestrian Re-identification Methods Based on Local Feature Enhancement Using Self-Feedback Human Analysis

Aiming at the issue of background clutter in person Re-identification (ReID), this paper proposes a context-aware and feature-enhanced ReID model by analyzing the spatial relationships between persons and their surrounding environments. At the sample level, an adaptive random background erasure strategy is introduced, which combines semantic segmentation and distance transformation to randomly remove background regions and reduce interference. At the feature level, a context-aware multi-branch dilated convolutional structure is designed to better capture spatial correlations between the person and the background, enhancing the model's ability to extract comprehensive global and fine-grained local features. Furthermore, beyond the most salient regions, the model learns additional discriminative features to improve feature diversity and recognition accuracy. Extensive experiments on standard benchmarks demonstrate the effectiveness of each component. Our method achieves state-of-the-art performance, with a Rank-1 accuracy of 92.6% on mainstream person ReID datasets.