Quantitative measures for classification of human upper body posture in video signal to improve online learning

Abstract

In an era dominated by video conferencing and distance learning, technologies that can increase the quality of education have become exceptionally important. Moreover, these unprecedented times can be used to introduce new and improve the traditional paradigms of classroom teaching. In this regard, it would be highly beneficial to develop methods for efficient estimation and classification of student attention and engagement based on readily available visual information. Today webcams and other different optical sensors have become ubiquitous and are used in online lecturing on regular basis. It was already demonstrated that the visual signal they provide is of sufficient quality for the prediction of student's learning curves. In our previous work, a software application for the recognition of the head, arms, and upper-body orientation and position was developed and initially tested on a limited dataset achieving a mean accuracy of 96.88%. In this research, a catalogue of upper body posture positions characteristic in online lectures is identified and described. A catalogue of quantitative measures for the classification of upper body posture positions, based on the set, are proposed. The measures are divided into spatial, temporal, and functional groups. Subsequent studies should verify the validity of proposed measures.