Pedestrian re-identification data set expansion method and system based on generative network

Abstract

The invention provides a pedestrian re-identification data set expansion method and system based on a generative network, a medium and a terminal. The method comprises the following steps: obtaining pedestrian images based on a pedestrian re-identification data set and forming a pedestrian image triple; performing feature extraction on the pedestrian image triple to obtain pedestrian features and style features; performing feature fusion on the pedestrian features to obtain multi-person fusion features; performing image reconstruction on the multi-person fusion features, the pedestrian features and the style features to obtain a first generated pedestrian image; performing image discrimination on the first generated pedestrian image to obtain a second generated pedestrian image; performing comparative learning on the second generated pedestrian picture to obtain a new pedestrian image; performing pedestrian identification on the second generated pedestrian picture to obtain a pedestrian image synthesized by the peers; and expanding the pedestrian re-identification data set by using the new pedestrian image and the pedestrian image synthesized by the peers. According to the invention, by generating pedestrians, the breadth of a pedestrian re-identification data set can be greatly expanded, the generalization ability of pedestrian re-identification in each scene is improved, and a good foundation is laid for engineering landing.

Description

technical field [0001] The invention belongs to the technical field of pedestrian re-identification, in particular to a method and system for expanding a pedestrian re-identification data set based on a generative network. Background technique [0002] Person re-identification technology (Person re-identification, Re-ID), also known as pedestrian re-identification, is a technology that associates peers under the cross-probe. Due to its wide application in security and business, this technology has become a computer vision technology. popular research directions. In the process of implementing the current pedestrian re-identification model, the main problems are the lack of data sets and the lack of cross-domain capabilities of the model. The current public datasets for pedestrian re-identification are only Market1501, DukeMTMC, etc., and some datasets have been taken off the shelf due to privacy issues. The data source of the model has become a major difficulty in the proce...

AI Technical Summary

Problems solved by technology

However, the shortcomings of this study are also obvious, that is, the characteristics of the generated pedestrians are quite different, that is, the picture synthesized by pedestrian A's picture 1 and pedestrian B's picture 1, and the pictu

Images