Sensor-based identification of human stress levels

In this work we present a mobile stress recognition system based on an existing activity recognition system using a hip-worn inertial measurement unit and a chest belt. Integrating activity knowledge, the prediction of different human stress levels in a mobile environment can be enabled while the state of the art is focussed on stress recognition in static environments. Our system has been implemented on an Android mobile phone and evaluated for different Bayesian networks as classifiers. Our implementation is able to operate in real-time with a stress inference rate of 1 Hz. The results of this work indicate that the implemented system is able to differentiate between the states 'No Stress' and 'Stress' in a mobile context. A more detailed distinction of stress in five substates has not been possible in a reliable way to date. With our results, the proposed system can serve as a basis for further improvements with larger data sets and for in-situ testing during disaster assessment.