三维场景是虚拟现实、元宇宙、电子游戏等领域的重要基础，其负责提供渲染所需的必要配置与内容。然而，三维数据本身的设计难度、复杂结构与版权维护导致了三维数据并无法大量获取。又随着用户终端与三维图形应用的飞速发展，三维场景的需求进一步增高。故场景的智能高效构建迫在眉睫。另一方面，三维场景不仅是用户与图形应用交互的载体，其自身也可作为交互中的一部分。所以我们不仅要引导场景本身的构建，也要辅助用户与场景的交互编辑。本文针对三维场景的构建与交互展开深入研究，提出了一系列解决方案。从如何让计算机全自动布局场景入手，进一步引入用户交互，即让用户可以控制场景自动构建的过程。本文的主要贡献如下：（1）提出了一种基于空间关联关系的三维场景自动布局方法。通过CSR测试使物体间的关联关系自动解耦，即在计算中可以直接忽略大量无上下文关联物体间的位置调整并只关注有上下文关系的物体，从而不仅提高了结果的合理性，也提高了算法的计算效率。进一步，本文提出了基于离散布局模式来自动布局三维场景物体的方法和一个用于将功能区自动布局的算法。（2）提出了“魔法指针”，一种实时、交互构造场景的方法。完全自动的场景构建经常不能满足用户需求，在实际图形应用（如互联网家装）中往往用户需要定制自己的场景，然而手动定制的场景却又会耗费大量人力。故本方法基于用户指定的位置，可以让用户控制自动布局的流程。在鼠标指针在场景中漫游时，符合当前指针处上下文的家具会自动连同合理的变换出现在对应的位置，且整个过程是实时的。在此过程中，用户可用最短的时间探索房间中有哪些适合放置的物体和它们的变换，进而决定布局的走向。同时，为解决单次交互仅能针对单一物体的问题，本文还衍生出了组物体智能编辑的方法。（3）提出了两个面向三维场景的智能应用。其一，室内三维场景的自动摄影，即自动生成一系列满足摄影学规则的视点用于向观众介绍当前室内场景。其二，场景调色盘，即让用户可以基于上下文探索与当前布局场景相关的模型库。
The 3D scenes are an essential foundation in virtual reality, metauniverse, electronic games, etc. 3D scenes are responsible for providing the necessary configurations and contents for rendering. However, the design difficulty, complex structure and copyright maintenance of 3D scenes lead to the inability to obtain a large amount of 3D data. With the rapid development of devices and 3D graphics applications, the demand for 3D scenes is increasing rapidly. Therefore, it is urgent to make the construction of computer-aided scenes.On the other hand, 3D scenes are not only the carrier for users to interact with graphical applications but also a part of the interaction. We should assist in constructing the scenes and in interactive editing between users and the scene. In this thesis, the construction and interaction of 3D scenes are studied thoroughly, and a series of solutions are proposed. Starting from how to make computers entirely automate the scene‘s layout, this thesis further introduces interactive synthesis, i.e., letting users control the process of crafting 3D scenes. The main contributions of this paper are as follows:(1) An automatic layout method of 3D scenes based on spatial relations between/among objects is proposed. Through CSR testing, the relations of objects‘ contexts are automatically decoupled. In other words, optimizing transformations among many objects with no relative contexts can be directly ignored, and only the objects with contexts are concerned, which not only improves the plausibility of the results but also improves the computational efficiency. Furthermore, this paper proposes a method to automatically arrange objects in 3D scenes based on discrete spatial layout pattern extraction and an algorithm to arrange groups of objects automatically.(2) A real-time and interactive method for constructing scenes is proposed. Fully automatic scene construction often fails to meet users‘ needs. Users often customize their scenes in actual graphic applications, such as internet home decoration. However, manually customized scenes will consume much workforce. Therefore, this method is based on the mouse cursor and allows users to control the automatic layout process. When the mouse hovers in the scene, the furniture that conforms to the context of the current cursor will automatically appear in the corresponding position with a reasonable transformation, and the whole process is in real time. In this process, users can explore what objects are suitable for placement in the room and where to transform them in the shortest time. Then users determine how the layout is arranged with objects added. At the same time, to solve the problem that a single interaction can only target a single object, this paper also derives the group object intelligent editing method.(3) Two intelligent applications for 3D scenes are proposed. First, the automatic photography of 3D scenes, i.e., the automatic generation of a series of photos that meet the rules of photography, is used to introduce the current scenes to the audience. Second, the scene palette allows users to explore the model library related to the current scene based on the context.
