Efficient Person Re-Identification via Progressive Filter Pruning and Body Part-Aware Feature Learning

Abstract Person re-identification (Re-ID) is a critical task in modern surveillance, enablingthe tracking of individuals across multiple camera views despite variationsin appearance, pose, and background. As public safety demands grow, Re-ID systems must operate efficiently and accurately in real-time, especially onresource-constrained devices. However, traditional deep learning models are oftentoo computationally intensive for practical deployment. This research presents an optimized Re-ID framework that integrates ProgressiveSoft Filter Pruning (PSFP) with local feature learning. PSFP reduces model com-plexity while preserving accuracy, and local feature learning enhances robustnessagainst occlusions and appearance variations. Extensive evaluations on bench-mark datasets demonstrate that the proposed model achieves state-of-the-artaccuracy with significantly reduced inference time, FLOPs, and memory usage.For instance, on the Market-1501 dataset, our method achieves 84.61% Rank-1accuracy, with a 37% reduction in FLOPs and a 16.7% decrease in memory usagecompared to the baseline. These results confirm the feasibility of lightweight, scalable Re-ID solutionsfor public safety and autonomous systems, while also supporting the ethi-cal and sustainability goals of energy-efficient AI for smart city environments.The complete implementation is available at https://github.com/women-ssniffp/Person-ReID-PSFP.git.