Unsupervised cross-domain pedestrian re-identification method

Abstract

The invention relates to an unsupervised cross-domain pedestrian re-identification method, and the method comprises the following steps: performing pre-training by using a labeled source domain training image to obtain a baseline network weight, and taking the baseline network weight as a baseline network initial weight of a multi-loss optimization learning training process; performing multi-lossoptimization learning training by using the label-free target domain training image, and performing multi-time multi-loss optimization learning training on the basis of the initial weight of the baseline network to obtain a baseline network after multi-loss optimization learning training; performing an unsupervised cross-domain pedestrian re-identification test by using the label-free target domain test image, and inputting the label-free target domain test image into the baseline network subjected to multi-loss optimization learning training for testing to obtain an identification result. According to the method, the natural similarity in the target domain image is concerned, complete dependence on a pseudo label is avoided, and compared with other methods in the same field, the method has higher recognition accuracy.

Description

technical field [0001] The invention relates to the field of pattern recognition and image retrieval in computer vision, in particular to an unsupervised cross-domain pedestrian re-identification method. In particular, it refers to the unsupervised cross-domain person re-identification method using deep learning. [0002] The use of deep learning refers in particular to learning based on multi-loss optimization. Background technique [0003] Pedestrian re-identification technology is usually used to solve the problem of people matching in non-overlapping fields of view. This technology is an important part of intelligent video analysis technology. It can be used to track criminal suspects, find lost people, etc., and has broad application prospects. [0004] In recent years, person re-identification technology has gained widespread attention and has become a research hotspot in the field of computer vision. [0005] Early person re-ID research mainly relied on traditional ...

AI Technical Summary

Problems solved by technology

However, these methods rely too much on pseudo-labels, and pseudo-label noise may degrade the performance of person re-identification

Images