An arbitrary modal pedestrian re-identification method based on modal perception distance optimization

By optimizing the modality perception distance, the problem of decreased recognition accuracy caused by modal differences in cross-modal pedestrian re-identification is solved, improving recognition accuracy and robustness, and adapting to pedestrian re-identification in various complex scenarios.

CN122223751APending Publication Date: 2026-06-16WUHAN UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
WUHAN UNIV OF SCI & TECH
Filing Date
2026-03-12
Publication Date
2026-06-16

AI Technical Summary

⚠Technical Problem

Existing cross-modal pedestrian re-identification methods suffer from decreased recognition accuracy in practical applications due to the assumption that the query image and the sample in the image library have different modalities. Furthermore, they suffer from problems such as intra-modal interference and insufficient modal alignment capabilities.

⚗Method used

A modality-aware distance optimization method is adopted. By constructing a multimodal pedestrian image sample dataset, calculating feature centers, constructing modality-aware center interaction and aggregation loss, optimizing the feature extraction network, and optimizing the matching distance by using modality mask matrix and cross-modal bridge sample-driven path propagation.

🎯Benefits of technology

It effectively adapts to arbitrary modal inputs, improves recognition accuracy and generalization ability, enhances cross-modal consistency and identity differentiation, suppresses intra-modal interference, improves retrieval robustness, and achieves accurate pedestrian re-identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122223751A_ABST
    Figure CN122223751A_ABST
Patent Text Reader

Abstract

The application provides an arbitrary modal pedestrian re-identification method based on modal perception distance optimization, and relates to the field of image retrieval.The method comprises the following steps: step S1, constructing a multi-modal pedestrian image sample dataset, including a training set and a test set; step S2, based on the training set, extracting modal features through a feature extraction network, and calculating feature centers; step S3, based on the feature centers, constructing modal perception center interaction and aggregation loss, and optimizing the feature extraction network; step S4, using the optimized feature extraction network to extract features of the test set, and constructing a modal mask matrix; step S5, based on the modal mask matrix, optimizing the matching distance through the cross-modal bridge sample driving path propagation, and completing pedestrian re-identification according to the optimized matching distance.The application adopts the above-mentioned arbitrary modal pedestrian re-identification method based on modal perception distance optimization, effectively suppresses the intra-modal interference, and improves the retrieval robustness of the mixed modal database.
Need to check novelty before this filing date? Find Prior Art