Hand Hygiene Noncompliance (in hospitals)
Reliable use of hand hygiene is associated with large reductions in hospital-acquired infections, which account for a significant fraction of hospital complications, mortality and healthcare expenditures. We are designing and demonstrating privacy-protective depth sensors and refining computer vision technology to make it easier for all clinicians and staff to perfect hand hygiene.
Detecting "the way"
Devil is in the details. These days it is possible to use RFID (or other marker-less) technology to detect if a caregiver has spend time near a hand wash dispenser. Yet few things can be said about "the way" a washing episode was performed. We work on pushing those boundaries forward by providing a fine grained understanding of the hand washing events. The value of our proposal allows going beyond an invasive "surveillance/monitoring" system by enabling an engaging reward framework (i.e. a score is provided after each episode allowing the user to compete with others). To proof the concept, we detect automatically and in real time 5 of the WHO protocol hand hygiene poses.
We have trained an AI algorithm capable of classifying the hand hygiene poses automatically from thermal images. The algorithm was trained with 287 exemplar images of each pose. More training data will allow achieving better results. Below, we show examples of the automatically detected poses, where the label at the top left is the detected hand pose and the green bar is the algorithm's confidence. The full test sequence can be find in the video below.