Emotion Recognition from Facial Images of People in a Mixed Reality Helmet

The research presented in the paper demonstrates how different machine learning methods handle the task of emotion recognition (emotion classification) on facial images of people in a mixed reality helmet. We developed different types of features extracted from facial image to be used as input for machine learning methods. Features were based on the location of facial anthropometric points. Different types of features allowed for the creation of multiple datasets. To optimise the hyperparameters of the proposed methods, we used various strategies in order to maximise classification metrics. The paper shows results of conducted experiments using different machine learning methods, boosting algorithms and also ensembling methods.