Fall Detection Algorithm Using a Smart Wearable System for Remote Health Monitoring

Abstract

Nowadays more people prefer to live independently, especially the elderly, leaving them prone to incidents that they might not be able to report. Falls, for instance, are responsible for over 3 million emergency hospitalizations for head injuries and hip fractures each year in the U.S. In addition, other cases often go unreported, leading to further complications including chronic disabilities and even fatality. Therefore, the detection of such incidents has become of urgent necessity. The purpose of this paper is to develop and propose a machine learning support vector classification (SVC) algorithm for fall detection using accelerometer, gyroscope, and magnetometer sensors embedded in a smart wearable system for remote health monitoring. The device is placed on the subject’s wrist to collect data on various motion activities in real-time, such as walking, running, jogging, waving, and stair-climbing in addition to other static postures like standing, lying, and sitting. The constructed dataset comprises 30 subjects with over 1200 data frames. The model achieved an overall accuracy of 98.3% and a specificity of 98.2% in separating falls from other daily-life activities.