A Novel Real-Time Fall Detection System Based on Real-Time Video and Mobile Phones

Abstract

In recent years, due to the growing population of the elderly, falls of elderly people have aroused wide public concern. Detecting timely falls of the elderly is significant to their safety. Numerous challenges exist in real-time fall detection systems because some features of normal human activities are greatly similar to the characteristics of falls. To address these problems, we propose a novel fall detection scheme and build a health-care system to detect falls of the elderly based on a real-time video surveillance system and a smart phone. The system contains two major modules. The first module is a feature extraction module. We adopt the Gaussian mixture model, tracking learning detecting algorithm and logpolar histogram to extract the characteristics of falls from the video surveillance system and the sensors embedded in mobile phones. The main purpose of the second module is to detect a fall-based on the features obtained in the first module. The experimental results show that every module is significant. Besides, our system is effective to separate falls from other similar actions such as bend down with an accuracy rate of more than 98% and performs better than other state-of-the-art fall detection systems.