Cross-View Gait Recognition Via Graph-Attention-Based Deep Relational Learning with CNN Feature Encoding
Computer and Decision Making: An International Journal · 02.11.2026 · 3 (6S) · 1147-1166
DOI: 10.1922/comdem.v3i6s.18898 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 02.11.2026. Verfasst von Pranesh Vijayakumar und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Gait recognition is a technique that identifies individuals based on their walking patterns. Traditional gait recognition approaches face difficulties in addressing viewpoint variations and do not fully utilise the temporal-spatial information among body parts. To overcome these challenges, we propose a novel deep learning architecture that integrates Convolutional Neural Networks (CNNs) and Graph Neural Networks (GNNs). CNNs are employed for efficient feature abstraction, while GNNs represent relational dependencies among body joints from various perspectives. Experiments on the CASIA-B and OU-ISIR datasets shows that our method achieves 95.8% accuracy and an F1-score of 0.923 under normal conditions, 89.4% accuracy and 0.894 F1-score for bag carrying, 88.1% accuracy and 0.881 F1-score for clothing variations, and 90.3% accuracy with an F1-score of 0.903 on OU-ISIR. These results outperform existing baselines by margins of up to 6.9%, demonstrating balanced precision and recall across diverse conditions. Ablation studies confirm the importance of integrating CNNs, GNNs, and the proposed fusion strategy. Visualisation of graph embeddings further shows that the model learns discriminative features that separate individual clusters effectively. Overall, the integrated CNN-GNN model offers a robust, generalisable, and scalable solution for multi-view gait recognition.
Zitieren
APA
Vijayakumar, P., & Sivakumar, N. (2026). Cross-View Gait Recognition Via Graph-Attention-Based Deep Relational Learning with CNN Feature Encoding. Computer and Decision Making: An International Journal, 3(6S), 1147-1166. https://doi.org/10.1922/comdem.v3i6s.18898
Vancouver
Vijayakumar P, Sivakumar N. Cross-View Gait Recognition Via Graph-Attention-Based Deep Relational Learning with CNN Feature Encoding. Computer and Decision Making: An International Journal. 2026;3(6S):1147-1166. doi:10.1922/comdem.v3i6s.18898
BibTeX
@article{Vijayakumar2026,
author = {Vijayakumar, Pranesh and Sivakumar, N.},
title = {Cross-View Gait Recognition Via Graph-Attention-Based Deep Relational Learning with CNN Feature Encoding},
journal = {Computer and Decision Making: An International Journal},
year = {2026},
volume = {3},
number = {6S},
pages = {1147--1166},
doi = {10.1922/comdem.v3i6s.18898}
}
EUVIMED