Cross-modal pedestrian re-identification method based on multi-modal image style conversion

Abstract

The invention discloses a cross-modal pedestrian re-identification method based on multi-modal image style conversion, and the method comprises: carrying out the collection and preprocessing of training images according to a pedestrian image data set, and obtaining a training sample; constructing an automatic coding network model for decoupling image features, performing feature decoupling on thepedestrian image input after processing by the model, and dividing the pedestrian image into domain-independent content features and domain-related style features; and constructing a generation network and discrimination network model based on modal transformation, the model realizing style attribute information exchange of pedestrian images in different domains, and realizing style-transformed sample generation. Aiming at the problem that a pedestrian re-identification algorithm is easily influenced by different illumination, the similarity matrix of different images is learned by extractingthe characteristics of the pedestrian images through the neural network, the metric matrix learned by the method has flexibility compared with a manually selected matrix, and the similarity between the image characteristics can be better obtained.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a cross-modal pedestrian re-identification method based on multi-modal image style conversion. Background technique [0002] With the rapid development of artificial intelligence technology and the improvement of people's requirements for social security, more and more cameras are put into use and appear in every corner of life, and people pay more and more attention to the analysis and research of camera videos. Multi-camera monitoring has a wide field of view, which overcomes the problem of limited monitoring range of a single camera, but also brings a large amount of video and image information. Reasonable use of camera monitoring information combined with good video tracking technology can filter out useful information from massive amounts of data. Video tracking technology involves many research directions, including image processing, computer vision, patte...

AI Technical Summary

Problems solved by technology

[0009] The present invention researches a method based on deep learning, uses generative confrontation network to realize the exchange of style attribute information of pedestrian images in the infrared domain and visible light domain, and solves the problem that the pedestrian re-identification method is affected by different illuminations in practice, which makes it difficult to identify pedestrians, and the design A neural network-based similarity metric learning method improves the robustness of cross-modal person re-identification methods based on multi-modal image style transfer

Images