Pedestrian Re-Identification Method Based on Combined Multiple Loss of Attention Mechanism

In recent years, pedestrian re- recognition has become an increasingly hot topic in the field of computer vision, and has been widely used in many fields such as intelligent video surveillance and security systems. Although significant progress has been made in the research of person re-identification of video sequences, the existing methods do not use comprehensive temporal information contained in the image data, and there are also problems such as the influence of interference information in scenarios such as occlusion or complex background. Our paper first proposes a bilateral network, which can fully extract the spatial information in the image sequence, and secondly, design the TAL module, introduce the spatial attention mechanism and the loss function, and insert it into the bilateral network at will, so as to construct a module that can perform joint feature extraction on the spatiotemporal information of the video sequence. In the training stage, the network was optimized by using the method of cross-entropy loss, triplet loss and MSE loss, and tested on the public MARS dataset, Duke dataset and LS-VID dataset, compared with the current mainstream methods, experiments on three datasets showed that the Rank-1 and mAP of the algorithm were improved.