A Fall Detection and Emergency Notification System for Elderly

Elderly are the fastest growing segment of the world population. Most elders prefer to live independently in their own homes. This makes them more vulnerable to emergency incidents such as fall and unconsciousness. Out of them, falling is the most dominant cause leading to both fatal and non-fatal injuries among elders. Timely detection and notification of falls can minimise the adversity caused by the impact. This paper discusses a non-obtrusive vision-based fall detection system using a service robot with a Kinect sensor attached and an automatic fall notification system based on Q-Learning. The proposed system is capable of distinguishing a fall from an activity of daily living by analysing body joint velocities and orientation of selected limbs using the RGB and depth data generated from the Kinect Sensor. The system can also differentiate between three types of falls referred to as Prone Position, Crawl Position and Kneel Position in the paper. Upon fall detection, the system can automatically make successive phone calls to a list of emergency contacts by considering the contact person's probability of answering a call and level of being busy. Furthermore this system has the ability to reduce the confusion and anxiety raised within an elder during a fall by ensuring quick notification and thereby timely assistance. An experiment was conducted to observe the effectiveness of the proposed system and it was able to achieve an accuracy of 92.5% and a sensitivity of 95.45%. The other critical observations made have also been highlighted in the paper.