Automated Tracking of Hand Hygiene Stages

The European Centre for Disease Prevention and Control (ECDC) estimates that 2.5 millioncases of Hospital Acquired Infections (HAIs) occur each year in the European Union. Handhygiene is regarded as one of the most important preventive measures for HAIs. If it is implemented properly, hand hygiene can reduce the risk of cross-transmission of an infection in the healthcare environment. Good hand hygiene is not only important for healthcare settings. Therecent ongoing coronavirus pandemic has highlighted the importance of hand hygiene practices in our daily lives, with governments and health authorities around the world promoting goodhand hygiene practices. The WHO has published guidelines of hand hygiene stages to promotegood hand washing practices. A significant amount of existing research has focused on theproblem of tracking hands to enable hand gesture recognition. In this work, gesture trackingdevices and image processing are explored in the context of the hand washing environment.Hand washing videos of professional healthcare workers were carefully observed and analyzedin order to recognize hand features associated with hand hygiene stages that could be extractedautomatically. Selected hand features such as palm shape (flat or curved); palm orientation(palms facing or not); hand trajectory (linear or circular movement) were then extracted andtracked with the help of a 3D gesture tracking device - the Leap Motion Controller. These fea-tures were further coupled together to detect the execution of a required WHO - hand hygienestage,Rub hands palm to palm, with the help of the Leap sensor in real time. In certain conditions, the Leap Motion Controller enables a clear distinction to be made between the left andright hands. However, whenever the two hands came into contact with each other, sensor data from the Leap, such as palm position and palm orientation was lost for one of the two hands.Hand occlusion was found to be a major drawback with the application of the device to this usecase. Therefore, RGB digital cameras were selected for further processing and tracking of the hands. An image processing technique, using a skin detection algorithm, was applied to extractinstantaneous hand positions for further processing, to enable various hand hygiene poses to be detected. Contour and centroid detection algorithms were further applied to track the handtrajectory in hand hygiene video recordings. In addition, feature detection algorithms wereapplied to a hand hygiene pose to extract the useful hand features. The video recordings did not suffer from occlusion as is the case for the Leap sensor, but the segmentation of one handfrom another was identified as a major challenge with images because the contour detectionresulted in a continuous mass when the two hands were in contact. For future work, the datafrom gesture trackers, such as the Leap Motion Controller and cameras (with image processing)could be combined to make a robust hand hygiene gesture classification system