Learning effective gait features for gait recognition using deep convolutional autoencoder

Abstract

Gait recognition is important for identifying individuals at a distance but becomes challenging by covariates such as clothing and carrying conditions. Recent methods, often reliant on supervised learning and extensive labeled data, may not be feasible in applications with large datasets. To address this, a new method using a deep convolutional autoencoder—an unsupervised learning technique—has been developed to extract distinctive gait features resilient to these variables. This technique reduces the dimensionality of feature space representing the gait data and these features are classified using softmax classifier. Experimented on the CASIA-B dataset, this approach demonstrates superior performance in gait recognition.

More information

Identifiers

Catalogue record

Export record