Gait recognition method based on 3D dense convolutional neural network

Abstract

The invention discloses a gait recognition method based on a 3D dense convolutional neural network. A network in the method employs 3D convolution to extract the transformation features of a gait at atime dimension, and has the feature retaining capability of a DenseNet structure. According to the invention, the method achieves the training of a classification and recognition model with the excellent performances under the conditions that a network depth is smaller and the number of training samples is smaller, wherein the classification and recognition model can recognize the identity according to the gait in a video. The testing on Dataset A of a CASIA gait database indicates that the method can achieve the training of the gait recognition model under the condition that the training samples are not sufficient. Moreover, the method is high in training speed, is small in model parameters, is high in recognition efficiency, and has the considerable recognition capability under the single-view-angle or cross-angle conditions.

Description

technical field [0001] The invention relates to the fields of computer vision and pattern recognition, in particular to a gait recognition method based on a 3D dense convolutional neural network. Background technique [0002] Compared with other biometric technologies (such as fingerprints, irises, faces, palm prints, etc.), gait recognition is non-invasive, non-contact, easy to perceive, difficult to hide, and difficult to camouflage. It has been widely used in intelligent video surveillance technology. attention and research. [0003] Gait recognition methods generally fall into two categories: model-based and appearance-based techniques. In the former, the parameters of a predefined model are tuned, while in the latter handcrafted gait features are extracted from images or videos. The establishment of model-based gait recognition method and parameter estimation have high computational complexity, large amount of data storage, and low real-time performance. The appearan...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the complex steps of video data preprocessing of current gait recognition technology (for example, GEI-based gait recognition needs to include pedestrian contour extraction, gait periodicity detection, GEI generation and other processing processes) and recognition, especially under the condition of cross-view The problem of low precision

Images