Abstract: Aiming at the raster image of architecture floor plan, this paper presents a lightweight algorithm for auto-
matically generating 3D building models. Firstly, IPF (integral projection function) is applied to split up the raster image, and effectively extract the sub-regions which contain the walls. Then, based on the improved SPV (sparse pixel vectorization) algorithm, this paper obtains information such as the location and the size of these walls. In order to identify the architectural components in the wall, like windows, doors and holes, the problem is transformed into solving image multiple classifications. And this paper proposes an efficient method to locate each architectural component in the drawing accurately, then insert them into the walls correctly. Finally, based on the identification result, the 3D building model can be generated rapidly. With the help of CEMO (conceptual and expressive modeling), a quick modeling system is developed, the 3D data will be processed conveniently as well. A large number of experiments prove that the method in this paper is robust and efficient. The method has many useful applications such as digital city, virtual reality authoring and public security planning.

Key words: floor plan, procedure modeling, vectorization, linear discriminant analysis (LDA), LDA classification

摘要： 针对建筑平面图的栅格图像，提出了一种全自动生成三维建筑模型的轻量计算方法。应用平均积分投影函数（integral projection function，IPF）方法对光栅图像进行区域分块，有效地抽取包含墙体对象的子区域。改进了基于非细化的稀疏点像素矢量化（sparse pixel vectorization，SPV）方法，用于抽取墙体的位置和尺寸等信息。为识别出墙体上的门窗和孔洞等建筑部件，将问题转化为图像多分类问题进行求解，同时设计了高效计算方案，精确地定位建筑部件在图纸中的位置。基于识别结果，进行三维建筑模型的快速生成，并集成于笔者所开发的三维建筑快速建模原型系统，方便了三维数据的加工和处理。通过大量实例，验证了所述方法的性能和效率。该方法可用于数字城市、虚拟现实内容创作和公共安全等领域。

关键词: 建筑平面图, 过程式建模, 矢量化, 线性辨别分析（LDA）, LDA分类