Enhancing automatic stress detection through activity-related behavioural features.
In the context of the European funded (FP7), research project INTERSTRESS, we have developed CBAR, a Camera and accelerometer –Based Activity Recognition tool. This tool monitors in real-time the subject’s activities through a video and an accelerometer modality, extracting behavioural features that correlate with stress. The CBAR video modality is based on a low-cost camera (Kinect), monitoring the subject’s upper body. From the video images sequence, Motion History Images (MHIs) are extracted, allowing the calculation of parameters related to upper-body activity, such as frequency of specific activities occurrence (e.g. a hand raised to the head), average upper body activity level, deviation, symmetry, etc. The accelerometer modality is based on two tri-axial accelerometers placed at the subject’s knees, for monitoring lower body activity such as foot trembling. All extracted behavioural parameters are provided to the INTERSTRESS platform, so as to enhance its automatic stress detection system, which is further based on monitoring of physiological signals. Moreover, by taking into account the stress correlates of the monitored parameters, the tool is capable to provide estimates, regarding the psychological stress of the monitored person. This video demonstrates how the developed tool translates the incoming signals from the video and accelerometer modalities into a stress level estimation and provides this estimate as input to the NeuroVR application that is used in the INTERSTRESS platform.