Gait-DenseNet: A Hybrid Convolutional Neural Network for Gait Recognition

Citation

Mogan, Jashila Nair and Lee, Chin Poo and Sonai Muthu Anbananthen, Kalaiarasi and Lim, Kian Ming (2022) Gait-DenseNet: A Hybrid Convolutional Neural Network for Gait Recognition. IAENG International Journal of Computer Science, 49 (2). pp. 1-8. ISSN 1819-9224

[img] Text
14.pdf - Published Version
Restricted to Repository staff only

Download (1MB)

Abstract

Gait is the walking posture of a human, which involves movements of joints at upper limbs and lower limbs of the body. In gait recognition, the human appearance changes are taken into account, which makes it easier to differentiate every individual. However, covariates such as viewing angle, clothing and carrying condition act as the crucial factors that affect the gait recognition process. In this work, a hybrid model that integrates pre-trained DenseNet-201 and multilayer perceptron is presented. The method first extracts the gait energy image by windowing the gait binary images. Subsequently, transfer learning of the pre-trained DenseNet-201 model is leveraged to learn the representative features of the gait energy image. A multilayer perceptron is then used to further capture the relationships between these features. Finally, a classification layer assigns the features to the associated class label. The performance of the proposed method is evaluated on CASIA-B dataset, OU-ISIR D dataset and OU-ISIR Large Population dataset. The experimental results show significant improvements on all the datasets compared to the state-of-theart methods.

Item Type: Article
Uncontrolled Keywords: Neural Network, Gait recognition
Subjects: Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions: Faculty of Information Science and Technology (FIST)
Depositing User: Ms Nurul Iqtiani Ahmad
Date Deposited: 04 Jul 2022 01:11
Last Modified: 27 Apr 2023 13:16
URII: http://shdl.mmu.edu.my/id/eprint/10123

Downloads

Downloads per month over past year

View ItemEdit (login required)