Wei, Xijia;
Fang, Yuan;
Shi, Fangzhan;
Wu, Shuang;
Williams, Amanda;
Gold, Nicolas;
Chetty, Kevin;
... Berthouze, Nadia; + view all
(2025)
WiProt: A WiFi-RGBD Fusion System for Robust
Protective Behaviour Recognition.
In:
Proceedings of the Second International Workshop on Radio Frequency computing (RFCom '25).
ACM (Association for Computing Machinery): New York, NY, USA.
(In press).
Preview |
Text
WiProt__A_WiFi_RGBD_Fusion_System_for_Robust_Protective_Behaviour_Recognition_RFCOM25.pdf - Accepted Version Download (2MB) | Preview |
Abstract
People with chronic pain often exhibit protective behaviours, such as reduced movement, low self-efficacy, or fear of movement. Automatic protective behaviour recognition is increasingly important in healthcare, with the potential to aid in chronic pain rehabilitation. Traditional single-modality systems, relying on sensors like WiFi or RGB, have limitations due to occlusion, privacy concerns, and environmental variations. To address these issues, we propose WiProt system that leverages WiFi Channel State Information (CSI) and RGBDepth (RGBD) extracted skeleton data to recognize protective behaviours. Our system integrates modality-specific pretrained encoders and an attention-based multimodal learning framework, enabling robust performance under various modality-missing situations. Experimental results demonstrate the system’s robustness, achieving over 73.9% accuracy with 75% missing data, significantly outperforming singlemodality models. WiProt offers a fault-tolerant solution for real-world applications, where sensor availability is often imperfect. In addition, we present the WiPRB, the first-ofits-kind WiFi-RGBD Fusion dataset focusing on leveraging WiFi-optical sensing for protective behaviour recognition.
Archive Staff Only
 |
View Item |
