Abstract
Human hands play an important role in nonverbal communication. Some motions including gestures and sign language are used for giving a message. However, other motions such as habit which human makes involuntarily also exist and they are considered as motions which characterize the person. In this paper, we consider that such motions would appear frequently independent of a conversation topic and we therefore detect frequent hand motions in natural conversation. Since habits vary according to individual, we cannot determine the motion patterns in advance. So motions which have similar trajectories are clustered based on LCSS similarity measure and large clusters are extracted.
