Pedestrian re-identification method based on data synthesis and color space learning

Through the method based on data synthesis and color space learning, the problems of insufficient data diversity and difficulty in color space adaptation in pedestrian re-identification technology are solved, and more efficient and robust pedestrian re-identification performance is achieved.

CN119785382APending Publication Date: 2025-04-08CHINA TELECOM DIGITAL INTELLIGENCE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411859488.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-17
Publication Date
2025-04-08

AI Technical Summary

Technical Problem

The existing pedestrian re-identification technology faces the problems of insufficient data diversity and difficulty in color space adaptation, which leads to poor performance of the model in complex and variable real environments, limiting the further development of the technology.

Method used

Using a method based on data synthesis and color space learning, data synthesis is carried out through diffusion model, new properties that have not appeared have been introduced, and color distribution is adjusted through channel replacement, exchange and splicing, nonlinear activation functions and convolution processing to generate training data that is suitable for multi-camera and complex lighting conditions.

Benefits of technology

It improves the diversity and authenticity of the data set, enhances the recognition accuracy and robustness of the model in different scenarios, reduces feature offsets under cross-modal and cross-device conditions, and achieves more efficient pedestrian re-identification performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119785382A_ABST
    Figure CN119785382A_ABST
Patent Text Reader

Abstract

The invention discloses a pedestrian re-identification method based on data synthesis and color space learning, and the method comprises the following steps: collecting and marking pedestrian data under multiple cameras, and constructing a pedestrian data set; inputting the pedestrian data set into the diffusion model for data synthesis; the image is input to a color space learning module, and a color conversion image mapped to a new color space is generated; training the backbone network; performing feature extraction on an input image by using the trained backbone network, and storing the feature in a feature library; feature extraction is carried out on the pedestrian images, cosine distances between extracted features and features in the feature library are calculated and sorted, and a candidate pedestrian list is generated. Based on a diffusion model and a color space learning technology, a multi-attribute amplification data set is constructed, backbone network training and feature retrieval are optimized, and the precision, robustness and cross-camera adaptive capacity of pedestrian re-identification are remarkably improved.
Need to check novelty before this filing date? Find Prior Art