Detecting Conversational Group Associates from Static Images

In this paper, we present a novel method for estimating levels of group involvement. With the recent development of single-person analysis in computer vision, social group analysis has received growing attention and many group detection methods have been proposed. Most of the previous studies considered each person in an image in terms of binary levels of involvement (group member or not), but actually each person can have various statuses in a social space. These complexity of social status sometimes causes a decrease in the group detection accuracy. Our approach expresses each person in terms of social involvement features representing the relationship to the surrounding people. An involvement level classifier is trained by using a machine learning algorithm. We evaluated our proposed method by comparison with a previous method and confirmed the advantageousness of our method.