Identification-driven emotion recognition system for a social robot

This paper presents a design of identification-driven visual emotion recognition system for a social robot. The proposed modification includes identification step prior to emotion classification in order to provide a personalized emotion recognition. The facial detection and parametrization is done by FaceTracker application. The feature extraction step is based on authorial feature set definition, that will work on any given mesh model, which guarantees ease of modification. The evaluation of the system uses two discriminant classifiers, decision tree and four variants of nearest neighbour classifier. The system is able to recognize 7 emotion classes. The results of classification were compared to previous research concerning interpersonal emotion recognition and had shown a significant improvement in performance. The usage of personal data increased the recognition rate to 99% with 1NN classifier, while the previous interpersonal system achieved the rate of 82%. The addition of identification provided a boost in recognition to discriminant analysis classifiers. Linear discriminant analysis recognition rate increased from 69% to 87%, while quadratic discriminant analysis increased from 81% to 92%. The modification had no effect on, already high, nearest neighbour classifiers performance.