A gait recognition method based on a generative adversarial network

Abstract

The invention discloses a gait recognition method based on a generative adversarial network, and the method comprises the steps: carrying out the training based on an improved generative adversarial network through employing a reverse rebroadcast algorithm in a feature extraction stage, and obtaining a multi-field conversion model which can carry out the conversion among a plurality of different types of images; in the Gait recognition phase, Firstly, gait videos in a test set and a verification set are converted into gait energy graphs, then the gait energy graphs of test samples are converted into gait energy graphs consistent with a state domain and a visual angle domain in the verification set through the multi-field conversion model, and an identification result is obtained by comparing the similarity between the test samples and all verification samples. According to the method, three interference factors of view angles, clothes and carried objects can be processed at the same time, high robustness is achieved in the aspect of processing the cross-view-angle gait recognition problem, and the defect that an existing gait recognition technology is not high in accuracy in cross-view-angle recognition is overcome. The method can be widely applied to the fields of access control systems, social safety, judicial criminal investigation and the like, and is suitable for most scenes with monitoring videos.

Description

Technical field [0001] The invention belongs to computer vision and pattern recognition technology, and specifically relates to a gait recognition method based on a generative confrontation network. Background technique [0002] As an important part of biometric recognition technology, gait recognition is being widely used. The main advantages of using gait for identity recognition are as follows: 1) Gait recognition is a long-distance non-contact identity authentication technology with great flexibility and convenience in operation; 2) Gait recognition can be observed without affecting the observation In the case of the object, complete the feature extraction and recognition process, which is non-intrusive and strong concealment. 3) Gait recognition does not require high video resolution. Even in the case of relatively low resolution, identity recognition can be completed task. [0003] In practical applications, there will always be some gait noise, such as the observer carryin...

AI Technical Summary

Problems solved by technology

Most of the previous gait recognition methods regard the recognition of the same viewing angle as the main research object, or only focus on solving one of the many interference factors, while ignoring that in actual situations, the interference factors that affect the recognition accuracy often exist at the same time

In addition, it is difficult for surveillance to capture gait videos from the same viewpoint as those in existing databases, which creates the so-called cross-view recognition problem

Images