A progressive granular pedestrian retrieval method and apparatus

CN118587734BActive Publication Date: 2026-07-14TSINGHUA UNIVERSITY

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Patents(China)
Current Assignee / Owner: TSINGHUA UNIVERSITY
Filing Date: 2024-04-23
Publication Date: 2026-07-14

AI Technical Summary

Technical Problem

Existing pedestrian retrieval methods lack the ability to distinguish pedestrian features under occlusion conditions, resulting in insufficient identity verification information.

Method used

A progressive granularity pedestrian retrieval method is adopted, which provides rich identity identification information by extracting features at three levels: global, component, and pixel. This includes determining the feature map of the training sample, performing pixel feature classification, obtaining the probability map, and performing progressive granularity feature extraction based on the feature map and the probability map, and finally training the pedestrian retrieval network model.

Benefits of technology

It improves the performance of pedestrian retrieval, providing more detailed descriptions and discriminative capabilities even under occlusion conditions, and solves the problem of insufficient feature discriminative capabilities in existing technologies.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN118587734B_ABST

Patent Text Reader

Abstract

The application provides a progressive granularity pedestrian retrieval method and device, which comprises the following steps: determining a feature map of a training sample; classifying pixel features of the feature map to obtain a probability map; performing progressive granularity feature extraction based on the feature map and the probability map to obtain a progressive granularity pedestrian retrieval feature; the progressive granularity pedestrian retrieval feature comprises a global granularity feature, a component granularity feature and a pixel granularity feature; training a pedestrian retrieval network model according to the progressive granularity pedestrian retrieval feature to obtain a trained pedestrian retrieval network model; and performing progressive granularity pedestrian retrieval according to a preset pedestrian retrieval algorithm based on the trained pedestrian retrieval network model. The application extracts three levels of progressive granularity features, i.e., global, component and pixel, to provide rich identity identification information and effectively improve the performance of pedestrian retrieval.

Need to check novelty before this filing date? Find Prior Art