Machine Learning for Single and Complex 3D Head Gestures: Classification in Human-Computer Interaction

Abstract

This paper presents a new Hidden Markov Model based approach for fast and automatic detection and classification of head movements in real time dynamic videos. The model has been developed to utilize human-computer interaction applications by using only the laptop webcam. The proposed model has the ability to predict single head and combined simultaneously in fast responses. Other models paid more attention to classify head nod and shake only, but our model contribute the role of other head movements. The model proposed here doesn’t need any user intervention or previous knowledge of its environment. In addition, there is no limitation on illumination changes and occlusions, as well as no restrictions on head movements ranges. The model achieved significant results and efficient performances when tested on unseen data. As the model accuracies were 94%, 99%, 83%, 87%, 93%, 96% for all head gestures (rest, nod, turn, shake, tilt and tilting) respectively. On the other hand, the model accuracy was 99% and 88% for combined and single cues respectively. The aim of this model is to provide a fast application to infer and predict human emotions and affective states in real time through head gestures.