A new framework for automatic 3D scene construction from text description

Generating 3D virtual scene plays an essential role in computer graphics and animation. However, manual 3D scene construction is time-consuming and requires enormous efforts from professionals. Recently, more and more researchers shift their interests into the field of automatic scene generation. Although it hardly yields results as good as manual construction, automatic scene generation largely speeds up the process of creating a virtual scene. In this paper, we propose a new framework capable of constructing 3D scene from text description directly, in which knowledge-based technology is adopted to represent the common sense of the real world. The framework extracts spatial relationships of objects described by texts and translates them into XML file with predefined grammar. A spatial reasoning algorithm is engaged next to decide the object layout in the scene. Several critical issues are taken into consideration in the spatial reasoning process for the purpose of dealing with more complex scenes and pursuing better visual effects. The final results are visualized by Autodesk Maya. Some examples are shown to evaluate the framework and the result verifies that our framework is highly efficient in 3D scene construction.