Improving person re-identification by multi-scale features and texture enhancement
Aiming at the problem that the existing pedestrian re-identification algorithm has a low accuracy rate when extract-ing pedestrian features due to illumination differences,variable poses and camera angles,and low image resolution,a method com-bining multi-scale network models and image textures is proposed.In this method,first,a sliding window-based self-attention mechanism module is constructed to obtain richer receptive fields and global features;then a progressive multi-scale network mod-ule is proposed to improve the representation of multi-scale networks at a finer-grained level to mine features in depth;finally,a tex-ture feature enhancement module is constructed to integrate global and local features at different spatial levels,thereby reducing the impact of factors such as occlusion,illumination,and low-resolution images on pedestrian re-identification.And on this basis,a multi-loss function joint strategy is adopted to fuse the global and local features of pedestrians,avoiding the decrease in accuracy caused by the unbalanced segmentation area.The experimental results show that the Rank-1 and mAP of the method on the two mainstream person re-identification datasets Market-1501 and DukeMTMC-reID have reached 95.5%,87.6%,89.3%and 79.4%,respectively.
万方数据
维普
