Citation

Tat, Jessen Tan Kean and Mogan, Jashila Nair and Ganesan, Thinesh (2026) Gait Recognition Using EfficientNetV2-S. In: 22nd IEEE International Colloquium on Signal Processing and Its Applications, CSPA 2026, 1 May 2026 - 2 May 2026, Kuala Lumpur.

Text 11.pdf - Published Version Restricted to Repository staff only Download (362kB)

Abstract

Gait recognition represents a biometric method of identification since it identifies people by their gait, or walking style, usually non-invasively and without being obtrusive. Hand-designed features Traditional techniques have the problem of being reliant on handcrafted features or need large, labeled datasets, which constrain their flexibility and efficiency over non-stationary, complex, real-world settings. To overcome these shortcomings, this paper suggests a gait recognition framework using deep learning with EfficientNetV2S in transfer learning. The model is refined using CASIA-B dataset consisting of various gait conditions and variations in e.g. clothing, objects carried, and camera poses. The EfficientNetV2S architecture achieves a desirable combination between accuracy and computation efficiency by using compound scaling, Fused-MBConv blocks and enhanced activation and attention mechanisms. The proposed system was able to deliver a high score of 98.24% accuracy compared to several available methods. It performs well with different conditions showing its great generalization ability, thus able to implement it in real world applications like biometric surveillance, access control and health monitoring. This contribution shows how EfficientNetV2S can be used as one of the practical and efficient methods to overcome contemporary gait recognition problems.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Deep Learning, CASIA-B
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Engineering and Technology (FET) Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	30 Jun 2026 07:45
Last Modified:	30 Jun 2026 07:45
URII:	http://shdl.mmu.edu.my/id/eprint/16155

Downloads

Edit (login required)