A New Computational Approach to Identify Human Social Intention in Action

In this paper, we propose to analyze the trajectories of the human arm to predict social intention (personal or social intention). The trajectories of different 3D markers acquired by Mocap system, are defined in shape spaces of open curves. The results obtained in the experiments on a new dataset show an average recognition of about $68\%$ for the proposed method, which is comparable with the average score produced by human evaluation. The experimental results show also that the classification rate could be used to improve social communication between human and virtual agents. To the best of our knowledge, this is the first paper which uses computer vision techniques to analyze the effect of social intention on motor action for improving the social communication between human and avatar.