Pedestrian re-identification method for generating by utilizing target posture

Abstract

The invention provides a pedestrian re-identification method for generating by using a target posture, and belongs to the field of computer vision. According to the method, the generative adversarialnetwork is utilized to generate the target pedestrian in the specific posture under the condition of giving the target posture and the pedestrian, the posture is normalized, and then the influence ofthe posture on the pedestrian is eliminated. Specifically, the method is divided into two parts: (1) generating pedestrians in a given posture by utilizing a generative adversarial network under the condition of the given target posture and the pedestrians; (2) performing feature extraction through a feature extraction network by utilizing the generated pedestrian with the target posture and the pedestrian with the real target posture, and then performing feature fusion on the pedestrian with the target posture and the pedestrian with the real target posture to form a feature vector; and carrying out the same operation on the real target posture pedestrian, so that two feature vectors are obtained; finally, performing distance measurement on the obtained feature vector, and combining information of a pedestrian camera. Further the identification performance is improved.

Description

technical field [0001] The invention relates to a pedestrian re-identification method generated by using a target posture, which belongs to the field of computer vision. Background technique [0002] With the rapid development of artificial intelligence and the wide application of deep learning. The research on pedestrian re-identification has attracted more and more attention and has become a research hotspot. However, there may be great differences between the postures of different pedestrians and the posture of the same pedestrian. Also, it is difficult to extract stable features that are not affected by pose changes in data with a large number of different poses. Limitations of existing pedestrian re-identification algorithms: it is difficult to expand to large-scale camera networks, because large-scale monitoring areas contain a large number of pedestrian information with different angles, different postures and different identities. Existing models require sufficien...

AI Technical Summary

Problems solved by technology

However, since the poses of different pedestrians may differ greatly from the poses of the same pedestrian

Moreover, it is difficult to extract stable features that are not affected by pose changes in data with a large number of different poses

Limitations of existing pedestrian re-identification algorithms: it is difficult to expand to large-scale camera networks, because large-scale monitoring areas contain a large number of pedestrian information with different angles, different postures and different identities

The performance of many previous methods is not ideal. Therefore, it is particularly important to design a good method to extract pedestrian poses without discrimination and features.

Images