In this paper, we propose a method for finding intervals of nursing activities from accelerometers and audio on mobile sensors which are attached to nurses in reality. If we can find the intervals of nursing activities correctly, it helps the data to be used for machine learning for activity recognition. We have extracted the times of nursing interactions between nurses and patients by A) recognize walking activity from accelerometers, B) recognize if s/he is in the patient's room or not at each time duration divided by walking activities, from the environmental noise levels of sounds, and, C) for the duration where s/he is assumed to be in the patient's room, apply voice activity detection by fundamental frequencies using Cepstrum method, and extract the duration in which a person speaks. As a result of the experience for 300sec of sensor data, we observed sufficient accuracy for each step of A)-C), and could reduce the time to 11.9 without no missing.
展开▼
