Transform-based camera domain adaptive pedestrian re-identification method

Abstract

The invention discloses a camera domain adaptive pedestrian re-identification method based on Transform. The camera domain adaptive pedestrian re-identification method comprises the following steps: encoding an input image into a vector sequence by using a cross patch encoder; inputting the vector sequence into a Transform encoder to learn image features, and constructing identity information loss by using the image features to optimize the network; a cross patch encoder and a Transform encoder are jointly regarded as a feature generator, features generated by the generator are input into a discriminator to judge the type of a camera, and camera classification loss and camera domain adaptation loss are constructed on the basis to optimize the discriminator and the generator respectively; and extracting feature vectors of the pedestrian images by using a generator, calculating Euclidean distances between the feature vectors of the to-be-queried image and the feature vectors of each image, sorting according to the distances from small to large, and selecting the pedestrian identity of the image with the highest sorting as an identification result. The method is high in accuracy, and can effectively solve the problem of image style difference caused by multi-camera image acquisition in a pedestrian re-identification task.

Description

technical field [0001] The invention belongs to the field of computer vision and image retrieval, and in particular relates to a Transformer-based camera domain adaptive pedestrian re-identification method. Background technique [0002] Person re-identification is a technique for retrieving specific pedestrians from large-scale image sets. This technology has important practical significance in the fields of intelligent video surveillance and intelligent security. In recent years, person re-identification technology has been developed rapidly, but the uncontrolled appearance variation of images among multiple cameras makes person re-id still a challenging task. In actual re-identification scenarios, images captured by different cameras often have differences in illumination, background, and resolution, and these differences will have a negative impact on the performance of re-identification. At present, there have been a considerable number of generative models trying to d...

AI Technical Summary

Problems solved by technology

[0003] In recent years, studies have shown that the ability of convolutional neural networks to retain fine-grained information and learn long-distance dependencies is not ideal, but most re-identification methods still choose to use deep convolutional neural networks to extract image features.

Images