Gait recognition method based on attention 3D frequency convolutional neural network

Abstract

The invention provides a pedestrian recognition method for a gait image. The attention mechanism is used for enhancing learning of important features, meanwhile, the frequency convolution mechanism isintroduced, the high-resolution size is given to high-frequency feature information, and the low-resolution size is given to low-frequency feature information. According to the method, on one hand, the calculated data volume can be greatly reduced, on the other hand, high-resolution high-frequency features have better expression ability, gait representation is more accurate. Further, the model effect is further improved by combining 3D-CNN to extract spatio-temporal information. Due to the fact that different gait angles have a huge influence on the accuracy of model prediction, the angle recognition network is trained firstly, then the corresponding gait recognition network is trained for each angle, and finally during testing, the gait angle is predicted through the angle recognition network, and then the corresponding gait recognition network is selected for gait recognition. Experimental results prove that the method can effectively identify the gait of the pedestrian.

Description

technical field [0001] The present invention relates to the field of biological feature recognition based on deep learning, in particular to a method for video pedestrian gait recognition. Background technique [0002] As a key feature in the field of machine vision and biometric recognition, gait features have opened up a new field of long-distance target recognition. Its main advantages are: the uniqueness of gait features, easy to perceive, difficult to camouflage and non-invasive. Based on the above advantages, gait recognition has a wide application prospect in video surveillance. [0003] The video image is a common way of displaying pedestrian gait in life scenes, which reflects the walking characteristics of pedestrians, but it is not likely that the image has defects such as light changes and noise disturbances, so it is very difficult to extract the features of the image. At the same time, pedestrian images have different angles, different clothes, and different d...

AI Technical Summary

Problems solved by technology

This method can learn the information of the entire video frame, and the information extraction is more comprehensive. At the same time, it can learn the space-time dimension information, and the gait description for different environments is more accurate, but the model has a large amount of calculation and complex parameter tuning.

Images