Invisible-Guardian: Using Ensemble Learning to Classify Sound Events in Healthcare Scenes

Abstract

The dramatic increase of senior population worldwide is challenging the existing healthcare and support systems. Whereas traditional home health monitoring systems (like video-based method) often cause users’ discomfort with severe feeling of privacy intrusion, which impedes the popularity of home health monitoring systems. Then Invisible-Guardian is proposed, an ensemble learning (EL)-based high-efficiency auditory information centered healthcare system for the elderly at home. The system collects everyday sound, like door open and close through a sound sensor array arranged indoors. Information such as the current event type is obtained after preprocessing, extracting features, classifying with the EL and supplemented by the inertial sensing units. If an emergencies event occurs, such as fall, the information will be sent to the doctor and relatives immediately, which greatly protect the daily life safety of the elderly living alone. The feasibility of the system is verified by human activity recognition experiments, and the recognition accuracy reaches 94.17%. Compared with the traditional home health monitoring method, it not only reduces the cost burden of hardware equipment, but also monitoring elders’ daily life with less feeling of privacy intrusion. what’s more, it minimizes the amount of data and ease of calculation, which improves the efficiency of home health monitoring for the elders and more favorable to the safety of the elderly living alone.