Enhancing Generalised Real-Time Hand Hygiene Recognition

Effective hand hygiene remains one of the most costefficient yet inconsistently enforced interventions for preventing the spread of infectious diseases worldwide. This paper presents the development of a generalised hand hygiene recognition system for real-time monitoring using a single-camera setup. Motivated by the global impact of infectious diseases and the critical importance of proper handwashing, the system aims to improve hygiene compliance across diverse users and environments. Using a dataset of WHO-compliant handwashing sequences, the system employs MediaPipe for hand landmark detection and a Long Short-Term Memory (LSTM) model to capture temporal dependencies in hand gestures. The processed data are normalised and reshaped for sequential model input, achieving a recognition accuracy of 93.89%. The system is deployed on an edge device (Jetson Nano) for live inference, providing immediate visual feedback during handwashing. Experimental results demonstrate robust generalisation, while limitations related to environmental variability and real-time latency are discussed. Future work includes performance optimisation, dataset expansion, and the integration of additional sensors to enhance recognition capabilities and practical deployment.