Empirical analysis of continuous affect
Peter Robinson, Tadas Baltrušaitis · 2015
Automatic analysis of affect from facial expressions has been extensively studied, but most work has considered only a small set of discrete emotions, typically Ekman's six basic emotions, or a small number of continuous measures, typically valence and arousal. We have developed a system that accommodates a much larger vocabulary of discrete emotions and links them with continuous measures that have been aggregated over a few seconds. The approach has a sound theoretical basis in multi-dimensional statistics, making it both principled and robust, while a graphical presentation makes it easy to understand.