CDGaitFusion: a multimodal gait recognition network based on the fusion of commonality patterns and differential features

Abstract

As a behavioral biometric relying on human walking dynamics, gait recognition enables non-contact and long-range identity verification, making it highly valuable for public security and intelligent monitoring applications. However, its performance in real-world environments often deteriorates due to challenges such as clothing variation, illumination changes, and cross-view inconsistencies, which weaken feature stability and discriminability. To address these issues, this paper introduces CDGaitFusion, a multimodal gait recognition framework that integrates shared motion patterns with individual-specific features. The proposed model incorporates two key modules: a Multimodal Hierarchical Mechanism (MHM), which hierarchically focuses on motion-sensitive body regions to enhance local detail perception, and a Commonality–Difference Feature Extraction (CDFE) module, which captures both cross-subject common traits and discriminative differences to strengthen dynamic feature representation. Through mutual interaction between these modules, the network achieves refined gait characterization that balances global semantic understanding with local motion distinctiveness. Experimental results demonstrate that CDGaitFusion attains Rank-1 accuracies of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$83.1\%$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$78.7\%$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$87.9\%$$\end{document} on the SUSTech1K, Gait3D, and GREW datasets, respectively, achieving favorable performance compared with most existing methods. Ablation analyses further verify the effectiveness of its multimodal fusion and motion-aware modeling. Overall, CDGaitFusion provides a reliable and extensible framework for mitigating gait feature degradation under complex real-world conditions.