This paper discusses about an automatic 3D scene generation system. Recently 3D graphics applications has become in great demand in various fields, especially in the video game industry year by year. The development of 3D graphics applications is still not easy and needs much time. Many tools or development systems for 3D graphics applications have been proposed so far. Even using such a development system, we have to prepare 3D scenes by manually laying out various 3D objects using a mouse device on a computer screen when developing 3D graphics applications. It also takes much time. We need any tool or system that can automatically generates 3D scenes. The authors have already proposed an automatic 3D scene generation system based on contact constraints. The contact constraints among 3D objects are defined as semantic database records by manually. It is useful to generate such semantic database records automatically from the information about layouts of real objects in the real world. So, this paper discusses about the usability of the robot vision technology to automatically generate such semantic database records.