Hand hygiene duration and technique recognition using wrist-worn sensors

Hand washing is an effective countermeasure to the spread of many types of infection. Recently, sensing technology has automated the sampling and study of hand hygiene rates. Surprisingly, many questions about the area are unresolved, motivating further exploration based on wrist-worn commodity sensors (accelerometer and MEMS gyroscope). This paper describes initial work on techniques for measuring the duration of washing events and classifying different scrubbing motions. The work compares different sensor types and their fusion, compares sensing from one wrist to measuring both wrists, and explains results of experiments on a range of hand washing motions in a variety of subject populations, some in clinics of a teaching hospital. Machine learning is used to explore such questions: the paper investigates numerous features extracted from sensor data, looking at sampling rates, windowing, and platform details that affect classification. In training and classification experiments, data collection starts on the wrist, activated by a message from a disinfectant dispenser; data is then transferred by radio to a base station for subsequent reduction, analysis and characterization. Results show that hand hygiene motions can be classified with up to 93% accuracy.