Recent robots are employed not only in manufacturing industries but also in living environment for supporting daily human activities. For intelligent support appropriate for the role of the place, robots need to recognize it. Although it is proposed in previous work to recognize the role of each place from its appearance, places with the same role could have different appearances. Since similar roles should cause similar human behavior, it is attempted in this article to recognize places with different roles by clustering human motions observed in various places in indoor space with RGB-D cameras.