Motion capturing machine learning. Bluetooth motion sensors

Abstract

Good and efficient care for our elderly is a major concern in our ever aging popu- lation. The elderly’s greatest fear is falling to the ground, not being able to get up or get help. Use of Bluetooth motion sensors and appropriate agents can help alert for in the event of a person falling. In this master thesis an experimental sensor rig has been created together with an android application which can capture and save the data from the rig. This data has then been transfered to a feed-forward backpropagating network in order to train it to recogise falls and separate them from fi e daily activities. We have used three sensors to accomplish this: an ac- celerometer; a gyroscope; and a magnetometer. Sensory data from these has been given to three separate networks which performs their own prediction before being sent to an overarching network to perform the fi prediction. Results show that it is possible to use such a system to a high degree of recognition, however the system is vulnerable to overfitting so care must be taken during training.