Investigating the Impact of Region-based Convolutional Neural Networks on Occluded Person Re-Identification in Social Scenes

This research explores the use of advanced image recognition algorithms to improve the identification of partially obscured individuals in surveillance footage. By investigating the impact of region-based convolutional neural networks on person re-identification, the study aims to enhance the effectiveness of smart surveillance systems, targeted tracking, and social anomaly detection. The research contributes to the development of a new dataset for occluded person segmentation and evaluates the effect of occlusion at different levels. The proposed model demonstrates promising performance, achieving a Rank-1 accuracy of 74% and Rank-5 accuracy of 90%, showcasing the model's effectiveness in detecting and re-identifying occluded persons. The findings offer insights for future improvements in person re-identification technology, with potential benefits for public safety and security.