Research Article
Federated Learning with Sensing Enhancement for CSI-Based Gait Recognition
@INPROCEEDINGS{10.1007/978-3-032-33662-0_35, author={Haoyu Liang and Jiayuan Lei and Juntao Yang and Zuan Qin and Dongqi Hou}, title={Federated Learning with Sensing Enhancement for CSI-Based Gait Recognition}, proceedings={Communications and Networking. 20th EAI International Conference, ChinaCom 2025, Xi’an, China, November 8--9, 2025, Proceedings}, proceedings_a={CHINACOM}, year={2026}, month={10}, keywords={CSI Gait Recognition Model Heterogeneity Cross-Scene Federated Learning}, doi={10.1007/978-3-032-33662-0_35} }- Haoyu Liang
Jiayuan Lei
Juntao Yang
Zuan Qin
Dongqi Hou
Year: 2026
Federated Learning with Sensing Enhancement for CSI-Based Gait Recognition
CHINACOM
Springer
DOI: 10.1007/978-3-032-33662-0_35
Abstract
Gait recognition systems based on CSI (Channel State Information) measurements face dual challenges in practical applications: individual differences among users and scene diversity. Individual differences make it difficult for a unified model to accurately capture personalized features, while environmental variations cause systematic shifts in the CSI measurement distribution of the same user across different scenes. We propose UniSense, a universal sensing enhancement module that ensures prototype alignment and cross-scene consistency through collaborative class feature prototype modeling. UniSense integrates into local models agnostically, mitigating environmental interference and enhancing robustness via prototype sharing in a federated learning manner, without requiring raw data upload or centralized model training. Experiments in two real-world environments (S1, S2) show UniSense improves recognition accuracy by 3.48% points under scene and model heterogeneity. Intra-scene experiments with non-overlapping user sets further show a 3.25% point accuracy improvement, highlighting UniSense’s adaptability and generalizability.