Detection and prediction of falls among elderly people using walkers

Abstract

Falls of elderly people are big health burden, especially for long-term consequence. Yet we already have research, describing how exactly elderly fall and reasons of falls. We aimed to develop means that could not only detect falls and send alerts to relatives and doctors to conquer one of the biggest fears of elderly to fall and do not have the ability to call for help, but also tried to implement fall prevention system. This system based on “relatively safe walking patterns” that our system tries to detect during the walk. During the work we used SensorTag 2.0 CC2650 sensors, iPhone and Apple Watch to collect motion data (Gyroscope, Accelerometer and Magnetometer) and compared the accuracy of each device. As we chosen iPhone and Apple Watch to use Core ML framework to integrate the neural network model we generated using Keras into prototype app. The iPhone app perfectly detects falls, but it needs to collect data more accurately, to improve the machine learning model to improve the work of prediction falls. The Apple Watch app does not work acceptable, despite well prepared Keras model and requires revision.