Person Re-identification Algorithm Based on Siamese LSTM
Daxiang Li, Rui Meng, Ying Liu · 2021
In this paper, we propose a new siamese long short-term memory (SLSTM) network model to solve the problem that the recognition accuracy is affected by the occlusion of pedestrian images. The backbone network is composed of two modules, CNN and LSTM, in order to use the local detail information of pedestrian images, firstly, the image is divided into blocks and the features of each image sub-block are extracted, then use the LSTM module to learn the dependency relationship between the features of the pedestrian image block, obtain the feature representation of pedestrian images through memory coding; at the same time, A new loss function is designed by combining the new triplet verification loss and Softmax recognition loss to reduce the intra-class difference and increase the inter-class difference. Comparative experiments were carried out on three pedestrian benchmark datasets of PRID-2011, Market-1501 and CUHK03. The experimental results show that the model proposed in this paper can reduce the impact of occlusion phenomena on re-identification, and the experimental results are better than other state-of-the-art methods.