A Pedestrian Re-Identification Method Based on Joint Local and Global Similarity Metric Learning

Abstract

A pedestrian re-identification method that combines local and global similarity measurement learning. First, all pedestrian images in the pedestrian re-identification database are preprocessed to extract the local features and global features of each pedestrian image; secondly, in the training module, the joint The local similarity and global similarity of pedestrian samples, measure the overall similarity of two pedestrian samples, and learn the measurement matrix; finally, in the test module, import the measurement matrix learned by the training module, and measure the relationship between each pedestrian sample to be tested and the pedestrian sample in the library. Similarity, sorted according to the similarity, and identify the pedestrian samples to be tested.

Description

Technical field: [0001] The invention proposes a method for pedestrian re-identification combined with local and global similarity measurement learning, which relates to the fields of computer vision and pattern recognition. Background technique: [0002] With the development of social economy, people have higher and higher requirements for safety precautions, especially in public places such as railway stations, airports, and large shopping malls. Video surveillance is a product of technological development and is widely used in the field of security protection. However, in the traditional video surveillance system, the monitoring task is done manually, which requires the monitoring personnel to stare at the surveillance video non-stop, which is a great challenge for the monitoring personnel. Pedestrian re-identification is a cross-monitoring pedestrian target recognition technology, which can quickly identify human targets of interest in the monitoring system. [0003] I...

AI Technical Summary

Problems solved by technology

[0003] In most monitoring scenarios, the captured images of pedestrians often have problems such as occlusion, pose changes, illumination changes, and low resolution, making pedestrian re-identification one of the most challenging problems in the field of intelligent video surveillance.

Images