MoGEI-HashNet: Motion-Enhanced Gait Energy Image Hashing for Fast and Scalable Gait Retrieval
IEEE Access · 2026 · 14 · 37307-37320
DOI: 10.1109/access.2026.3671827 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von P. Hrudya und 1 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Centre for Internet Studies and Artificial Intelligence.
Abstract (Originalsprache)
Gait recognition has become an increasingly vital biometric modality due to its capacity to recognize individuals at a distance, non-intrusively, and under covert surveillance conditions. Nevertheless, real-world gait recognition remains challenging because appearance-altering co-variates such as clothing, carried objects, and viewpoint differences will significantly degrade the accuracy of the system. In addition, large-scale surveillance systems require compact and computationally efficient features to retrieve information quickly in large galleries. To mitigate these challenges, this study introduces MoGEI-HashNet, a motion-enhanced deep hashing framework that combines static and dynamic gait features within a scalable retrieval pipeline. The proposed MoGEI-HashNet, consists of a two-channel gait descriptor comprising the Gait Energy Image and a motion map derived from frame-to-frame silhouette differences, which will help the model to capture both shape and temporal dynamics. A lightweight CNN enhanced with Squeeze-and-Excitation attention blocks extracts discriminative features, which are further projected into low-dimensional binary hash codes using a learned hashing layer. The model is optimized through cross-entropy classification, quantization regularization, and a pairwise contrastive loss to ensure within-class feature distributions and enhanced between-class discrimination in the Hamming space. Experiments on CASIA-B dataset demonstrate that the proposed MoGEI-HashNet model achieves a Rank-1 accuracy of 98.83% while producing highly compact 64-bit representations, allowing large-scale real-time retrieval. The results demonstrate that the combination of SE-enhanced CNN features and binary hashing provides an efficient and effective solution for large-scale gait retrieval. The proposed method is further validated on additional benchmark datasets, including OU-ISIR LP and OU-ISIR MVLP.
Förderung
- Centre for Internet Studies and Artificial Intelligence
Zitieren
APA
Hrudya, P., & Poornachandran, P. (2026). MoGEI-HashNet: Motion-Enhanced Gait Energy Image Hashing for Fast and Scalable Gait Retrieval. IEEE Access, 14, 37307-37320. https://doi.org/10.1109/access.2026.3671827
Vancouver
Hrudya P, Poornachandran P. MoGEI-HashNet: Motion-Enhanced Gait Energy Image Hashing for Fast and Scalable Gait Retrieval. IEEE Access. 2026;14:37307-37320. doi:10.1109/access.2026.3671827
BibTeX
@article{Hrudya2026,
author = {Hrudya, P. and Poornachandran, Prabaharan},
title = {MoGEI-HashNet: Motion-Enhanced Gait Energy Image Hashing for Fast and Scalable Gait Retrieval},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {37307--37320},
doi = {10.1109/access.2026.3671827}
}
EUVIMED