Video-Based Person Re-Identification Using Spatial-Temporal Memory Coupling Network

Video-based Person Re-identification (Re-ID) aims to automatically retrieve video sequences of the same individual across non-overlapping cameras. While methods based on deep convolutional neural networks have achieved significant success in video-based person re-identification, they often suffer from interference caused by similar background interference and person occlusion, leading to subpar performance in Re-ID. Therefore, we propose a Spatial-temporal memory-coupled network framework, referred to as STMCN, which alleviates the influence of spatial and temporal interference by incorporating spatial and temporal features into a memory module, effectively addressing issues such as similar background interference and person occlusion. Additionally, we introduce a temporal saliency erasure module for head prediction to extract complementary features from consecutive video frames, enabling us to capture holistic characteristics of the target identity and achieve improved Re-ID performance. Experimental testing on the Mars dataset demonstrates that our framework outperforms most existing methods, achieving a Rank-1 accuracy of 92.3% and a mean average precision (mAP) of 87.5%.