Fast3D Building Modeling Based On Sparse Pixel Vectorization

This paper proposes an efficient method for the vectorization and the recognition of the architecture floor plan raster image, and based on this, implements a lightweight algorithm for automatically generating3D building models, in order to meet the needs of the actual requirements. First of all, Average Integral Projection Function (AIPF) is applied to split up the raster image, effectively extract the sub-regions which contain the walls. Then, based on the Sparse Pixel Vectorization (SPV) core algorithm, we get the wall’s properties such as its location and size accurately. Besides, the improved algorithm is fast enough to allow for the real system. In order to identify the architectural components in the wall, like windows, doors and holes, the problem is transformed into solving image multiple classifications. We adopt the Linear Discriminant Analysis (LDA), which is among the most optimal dimension reduction methods for multi-classification. However the traditional LDA has the undersampled problem, with the help of the generalized singular value (GSVD) and the QR decomposition, we can solve this problem perfectly. Without any sacrifice in the main information, this efficient approach reduces data dimension, saves computational time and memory, then recognizes and locates each architectural component accurately. Finally, based on the identification result, the3D building model can be generated rapidly. With the help of CEMO (Conceptual and Expressive Modeling), which is a quick modeling system developed by our lab, the3D data will be processed conveniently as well. A large number of experiments will prove that the performance of the method in this paper is both robust and efficient. The method in this paper has many useful applications such as digital city, virtual reality authoring and public security planning.