Development of Wearable Device and Clustering Based Method for Detecting Falls in the Elderly

This work presents a wearable inertial sensor-based system for detecting falls in the elderly. The system is fixed on the elderly’s head to obtain stable human body acceleration. To detect a variety of falls, a K-means model based on abundant daily activity data is constructed, and then the fall samples are detected by comparing the similarity with the other non-fall samples. The system requires no fall samples for training, and achieves an accuracy of 96.8%.