Research Behavior Analysis & Understanding
Detecting Conversational Group Associates from Static
We propose a novel method for estimating levels of conversational involvement. With the recent development of single-person analysis in computer vision, social group analysis has received growing attention, and many group de- tection methods have been proposed. Most of previous stud- ies consider each individual person in an image as binary lev- els of involvement (group member or not), but actually each person has various status in social space. These complexity of social status sometimes causes to decrease group detection accuracy. Our approach expresses each individual person as social involvement features which represent the relationship to surrounding person. And involvement level classifier is trained by using machine learning algorithm. We evaluated our proposed method by comparing with previous method and confirmed the advantage of our method.