Multi-scale attention pedestrian re-identification deep learning system based on guidance

Abstract

The invention provides a guidance-based multi-scale attention pedestrian re-identification deep learning system with strong feature representation capability and generalization capability, and the system is characterized in that the system comprises a pedestrian picture acquisition part for acquiring pedestrian pictures including pedestrian to-be-detected pictures and pedestrian candidate pictures; a picture feature extraction part which is used for extracting global features and local features of the pedestrian picture under each scale according to a pre-trained multi-scale feature extractionmodel based on guidance; a picture feature splicing part which is used for respectively splicing the global features and the local features of each pedestrian picture to serve as pedestrian picture features corresponding to each pedestrian picture; a feature distance calculation part which is used for calculating feature distances between different pedestrian picture features according to the pedestrian picture features; and a similarity degree judgment part which is used for judging the similarity degree of the pedestrian to-be-detected picture and the pedestrian candidate picture as a pedestrian re-identification result according to the distance between the feature distances.

Description

technical field [0001] The invention belongs to the technical field of computer image recognition, and in particular relates to a guidance-based multi-scale attention pedestrian re-identification deep learning system. Background technique [0002] The pedestrian re-identification task aims to identify and match pedestrians through two disjoint cameras. Usually, the appearance of pedestrians can change greatly due to changes in posture, illumination, occlusion, viewing angle, etc.; and in public, different pedestrians may wear very similar clothing, such as wearing dark thick coats in winter. These phenomena will bring huge challenges and difficulties to the pedestrian re-identification problem to a large extent. Under these severe interference factors, it is often necessary to rely on some subtle differences to achieve pedestrian re-identification. These subtle factors can be global, such as body shape and gender; they can also be local, such as shoes and hairstyles. Coars...

AI Technical Summary

Problems solved by technology

However, this method has the disadvantages of time-consuming calculation, large model parameters, and independent multi-scale feature learning.

Images