Abstract:

The problem of emotion recognition is becoming an important tool for profiling human interactions with technology. Utilisation of smartphones and wearables delivers personal tools which can be used for emotions data acquisition and processing. In widely discussed publications we can find many algorithms and mechanisms for physiological signals analysis applied in many areas such health care, gamming, advertisement, learning. Beforehand a thorough study of available approaches for emotions sensing have been reviewed and selected for further inspection based on the application of photoplethysmography PPG and skin conductance GSR data channels. The developed and compared in the paper AI methods implement multi-modal original approach, concentrating on wearable devices application.  Following comprehensive review on physiological signal-based emotion recognition, we presented our practical approach for emotions description, sensing, data channels, sensor implementation and algorithmic support implemented within mobile application and server-side classification services. Our proposition of original approach utilizes data fusion methods for selected sensor data streams (IMU, PPG, GSR) and a set of developed classifiers. A summary and comparation of obtained results conclude the paper and presented quantitative capabilities of implemented classification methods.