Seven-step handwashing recognition based on multi-angle information fusion from dual millimeter-wave radars

Hand hygiene is essential for reducing healthcare-associated infections (HAIs), yet compliance remains suboptimal. Although monitoring can significantly improve hand hygiene compliance and quality, existing methods often entail high labor costs, privacy concerns, or intrusive sensors. To address these limitations, this study proposes a non-contact seven-step handwashing recognition method using millimeter-wave radar. A dual-radar collaborative architecture is designed, in which front-view and up-view radars jointly capture hand-motion scattering signals. A dual-branch convolutional neural network is developed to extract and fuse multi-view spatiotemporal features for accurate action discrimination. Six representative models are further compared under different radar configurations to evaluate overall performance. Experimental results show that the number and placement of radar sensors critically affect recognition accuracy. The front-view radar achieved the highest accuracy on the independent test set (87.9%), substantially outperforming the up-view radar (55.7%). The dual-radar system exhibited significantly higher accuracy and stability than any single-radar configuration. Among all evaluated models, the bidirectional long short-term memory network achieved the best performance, with an average accuracy of 97.7% in five-fold cross-validation and 97.9% on the independent test set, a 10.0% improvement over the best single-radar solution. For four handwashing actions, Finger webs, Thumbs, Finger tips, and Wrists, the system reached 100% accuracy. These results suggest a promising cross-subject recognition trend under a small-scale subject-independent evaluation setting, while further validation with larger and more diverse populations is required before concluding large-scale generalization. This study demonstrates accurate, low-power, and privacy-preserving recognition of seven-step handwashing actions, providing a promising foundation for intelligent hand hygiene monitoring in healthcare environments.