Frame rate enhanced gait identification method based on generative adversarial network and device thereof

Abstract

The invention belongs to the field of computer vision, and particularly relates to a frame rate enhanced gait identification method based on a generative adversarial network and a device thereof. Themethod and the device aim to reduce noise of an identified image and improve accuracy in gait identification. The method comprises the steps of generating a frame between two continuous frames in a data set through a generative adversarial network, combining the generated frame with an original frame for calculating a gait energy diagram, and then identifying an individual through a gait energy diagram identification network. The generative adversarial network in the method can remarkably improve the frame rate of the original image sequence, and furthermore the generated image has relativelyhigh robustness to noise, thereby performing a noise reduction function on the gait energy diagram. Furthermore a novel boundary ratio loss function is added into the gait energy diagram identification network, thereby well balancing magnitudes between different loss function and greatly improving model training stability. The method and the device can remarkably improve gait identification rate on the condition of cross visual angle or no cross visual angle.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a frame rate enhanced gait recognition method and device based on a generative confrontation network. Background technique [0002] Gait recognition is one of the most important problems in the field of biometrics. Gait information is highly representative, difficult to camouflage, and does not require the cooperation of individual subjects. The current mainstream gait recognition method is to segment the original image first, and then store the human body outline as a grayscale image for subsequent processing. Due to the performance of the image acquisition device itself, the frame rate we can obtain is very limited. In addition, in this process, the segmentation algorithm cannot segment the original image 100% accurately, and sometimes even gets serious noise. In this case, it is particularly important to deploy a frame rate enhancement algorithm that is robust to noi...

AI Technical Summary

Problems solved by technology

The current mainstream gait recognition method is to segment the original image first, and then store the human body outline as a grayscale image for subsequent processing. Due to the performance of the image acquisition device itself, the frame rate we can obtain is very limited.

Images