The Regularization Of Building Roof Top Structure Based On Topology Optimization

With the acceleration of urbanization,the high accuracy and semantic-rich virtualization of 3D building assets have become more and more important to support various applications such as urban planning,emergency response,and location services.The use of remote sensing data to automatically reconstruct urban-scale building models has become a research hotspot at home and abroad.However,it is still a challenging task to develop a fully automated photogrammetric computer vision system that can generate high-precision building models on a large scale.One of the most challenging tasks for 3D reconstruction is to standardize the noise introduced by the boundary of a building that lacks shape knowledge with raw data.Even if only the roof of the building is reconstructed,the reconstruction process will face great difficulties.This article mainly discusses the key issues in the process of building roof reconstruction.The specific content is as follows:1.The 3D point cloud data for the central area of a city is often blocked by tall trees or vegetation,making it difficult to recognize man-made objects such as buildings.The direct quadratic polynomial fitting method is used to extract the regional information of typical local areas of vegetation and building targets such as tall trees,and construct the sensitive structural features of the regional targets.Furthermore,the building target-sensitive classification task of three-dimensional point cloud data can be completed through fuzzy logic,and the classification of LiDAR point cloud data in a large scale can be quickly and effectively realized.2.A method of building roof segmentation based on airborne lidar data is proposed.In order to obtain the optimal results at the target level and the pixel level,the energy minimization process was carried out 3 times in the order of hierarchy.First,use the active multi-plane fitting method to obtain a reliable initial segmentation;then,minimize the coarsest energy function composed of the pixel-level plane fitting error and the number of target-level plane hypotheses to obtain the optimal label space;secondly,minimize The energy function formed by the plane fitting error and the spatial smoothness between adjacent planes obtains the optimal segmentation result.Finally,considering the prior knowledge of the building roof structure,by minimizing the energy function of the structural adjustment plane fitting error,the optimal plane parameters of the segmented plane hypothesis are obtained.Experimental results show that this method can accurately segment building roofs from airborne LiDAR data quickly,stably and reliably.3.A data-driven modeling method is proposed to reconstruct a three-dimensional roof model at an urban scale from airborne laser scanning(ALS)data.The focus of this method is to implicitly derive the shape of the three-dimensional building roof from the given building boundary noise information in a gradual manner.The binary space partition(BSP)technology is used to recover the topological elements in the modeling clues.Under the framework of minimum description length(MDL),the implicit regularization process combined with hypothesis and testing(HAT)realizes the regularity of the building roof model.Automatically estimate the parameters of MDL optimization based on Min-Max optimization and entropy weight method.The results show that this method can generate accurate and regularized 3D building roof models.