You Can Wash Hands Better: Accurate Daily Handwashing Assessment with a Smartwatch

TL;DR

This work tackles low adherence to WHO handwashing guidelines by presenting UWash, a smartwatch-only system that automatically assesses handwashing technique using inertial measurement unit data. Framing handwashing as a gesture semantic segmentation problem, it employs a lightweight dual-stream Temporal U‑Net with post-processing and a WHO-based scoring rule to provide fine-grained gesture timing and a final quality score. The approach achieves high gesture recognition accuracy (up to 92.27% after smoothing) and robust cross-domain, cross-time performance across 51 participants in real-world settings, with a publicly released dataset and code. The practical impact lies in enabling accessible, ongoing feedback to cultivate high-quality daily hand hygiene without external sensors or manual auditing.

Abstract

Hand hygiene is among the most effective daily practices for preventing infectious diseases such as influenza, malaria, and skin infections. While professional guidelines emphasize proper handwashing to reduce the risk of viral infections, surveys reveal that adherence to these recommendations remains low. To address this gap, we propose UWash, a wearable solution leveraging smartwatches to evaluate handwashing procedures, aiming to raise awareness and cultivate high-quality handwashing habits. We frame the task of handwashing assessment as an action segmentation problem, similar to those in computer vision, and introduce a simple yet efficient two-stream UNet-like network to achieve this goal. Experiments involving 51 subjects demonstrate that UWash achieves 92.27% accuracy in handwashing gesture recognition, an error of <0.5 seconds in onset/offset detection, and an error of <5 points in gesture scoring under user-dependent settings. The system also performs robustly in user-independent and user-independent-location-independent evaluations. Remarkably, UWash maintains high performance in real-world tests, including evaluations with 10 random passersby at a hospital 9 months later and 10 passersby in an in-the-wild test conducted 2 years later. UWash is the first system to score handwashing quality based on gesture sequences, offering actionable guidance for improving daily hand hygiene. The code and dataset are publicly available at https://github.com/aiotgroup/UWash

Figures (22)

• Figure 1: UWash utilizes records of motion sensors of smartwatches, i.e., accelerometers and gyroscopes, to segment handwashing gestures, estimate the duration of each gesture, and score the quality of each gesture as well as the entire procedure with WHO guidelines.
• Figure 2: Uwash can automatically detect the event of handwashing, not requiring to work along with Bluetooth sensors in dispensers mondol2015harmony or to awaken the monitoring system manually samyoun2021iwash. We attribute this ability partly to the unique characteristics explained in Section \ref{['sec:character']}, i.e., successive, periodical, and being with expectable pre-activities and post-activities.
• Figure 3: Handwashing patterns emerge in a larger field of view (left). However, a larger field of view may lead to a large prediction error under window-grained prediction when handwashing gestures switch (right). Therefore, we apply the semantic segmentation method for sample-wise prediction in the window to bypass the dilemma.
• Figure 4: Deep Network Architecture of UWash. The dual-branch U-Nets ronneberger2015u take data from two modality sensors, i.e., accelerometers and gyroscopes, as inputs, respectively. Feature maps from two branches are further concatenated in high-level layers for sample-wise gesture recognition. Since accelerometers and gyroscopes have different modality measures of handwashing gestures, before concatenating, the squeeze-and-excitation modules hu2018squeeze automatically normalize their feature maps.
• Figure 5: A post smoothing example. UWash can do sample-wise recognition well directly, while jitter errors cause the majority of the false recognition. To reduce these errors, we propose post smoothing methods including multiple test voting and the mode filter, which are simple but work effectively.