Estimating Human Activities in Bathroom Through Sound Event Detection in Embedded Systems

Because bathroom surfaces are prone to becoming slippery due to water and soap, the risk of falls in bathrooms becomes considerable, especially for elderly people with weak legs and hips. Furthermore, the bathroom environment is characterized to significant temperature fluctuations, potentially leading to heat shock. To mitigate these risks, an event detection system must be in place to quickly detect accidents and call for rescue, requiring the estimation of human activities in the bathroom. Accordingly, we propose a method that estimates human behavior in the bathroom using sound event detection (SED). To ensure reliability, we compiled a dataset encompassing five people for model training. Subsequently, we implemented the proposed model on a microcontroller and found that its accuracy did not degrade significantly compared to that of Conformer, the state-of-the-art model in DCASE2022 Task3.