ILFMT-LS: Specific Target Detection in Mobile Crowds Based on Deep Learning

Detecting and locating specific individuals using non- overlapping multi-camera setups is a key focus in visual systems. However, targets are often obscured, making globally sensitive reidentification methods impractical. To address this, a novel method, Integrated Local Feature Matching with Transformers at Light Speed (ILFMT-LS), was proposed. It integrates a Local Feature Transformers (LoFTR) module and Transformer for rapid matching of local image features. A Convolutional Block Attention Module (CBAM) enhances salient region features and suppresses masked regions, improving visible area importance. This method achieved 92.27% precision and 93.25% recall on DukeMTMC-reID and CUHK03 datasets, indicating its superiority over other methods.