This video shows a mobile activity recognition system using accelerometer data from sensors strapped onto body parts.

The mobile phone collects the acceleration data from the sensors via Bluetooth wireless protocol and evaluates them using artificial neural networks (ANNs) in real-time. The weight in the networks were adjusted in the learning phase which was carried out on a workstation beforehand.

Each activity has an associated ANN, all of them are evaluated and the one with the highest score is selected as the recognized activity as shown at the bottom of the mobile screen.

Switching from one activity to another or unknown activities can throw off the recognition temporarily. This can be filtered by setting threshold rates and only letting an activity to be recognized if it scores higher by some margin than the others.