A geometric approach for recognizing emotions from 3D images with pose variations

Abstract

Emotions are an incredibly important aspect of human life. Research on emotion recognition for the past few decades have resulted in development of several fields. In the current scenario, it is necessary that machines/robots need to identify human emotions and respond accordingly. Applications in this field can be seen in security, entertainment and Human Machine Interface/Human Robot Interface. Recent works on 3D images have gained importance due to its accuracy in real life applications as emotions can be recognised at different head poses. The intention of this work has been to develop an algorithm for recognition of emotion from facial expressions, which recognizes 6 basic emotions, which are anger, fear, happy, disgust, sad and surprise from 3D images in 7 yaw angles (+45° to -45°) and 3 pitch angles (+15°,0°, -15°). Most of the reported work considers + yaw angles. While in the current work, both positive as well as negative pitch and yaw angles are considered. BU3DFE database is used for the implementation. The proposed method resulted in improved accuracy and is comparable with the literature.