Abstracting And Fitting Reference Plane For Scattered Point Cloud Data Of Ancient Architecture

The Laser Scanning technology has been used in cultural relic protection widely because of its speed, accuracy, non-touch, real-time and automatism. There are many characters in the point cloud data of ancient architectures such as large scale in space, complexity and huge number of data, so it’s urgent and difficult to resolve the problems of point cloud data’s storage, management, display, features extraction and3D modeling. Now every section is considered separately, not as the whole processing strategy. It takes point cloud data of ancient architecture as the research objects in this paper. A whole idea of processing point cloud data was proposed, in which plane, cylinder and sphere were used as reference surfaces to generate triangulation and deep image. This idea can translate three dimension’s problems into two dimension’s problems, so the complexity of storage, display and processing point cloud data can be reduced distinctly. In order to search nearest neighbor of point cloud data, a new high efficiency spatial index-MultyGrid-KD tree was raised. How to segment scattered point cloud data automatically and abstracted reference surface is key point. A series of solution including Gauss Map, curve map, clustering analyzing and predefined model were proposed to overcome the problems, such as huge number and many noises of point cloud data of ancient architecture. At the same time, the characters of features can be extracted too so as to complete segmentation and fitting integrative. It mainly includes the following contents:(1) A series of effective algorithms of extracting reference surfaces were proposed. Based on predefine model and clustering analyzing in Gauss map and curve map, geometric characters were Statistical Analysis from overall point of view to distinguish noise and core points. According to the known characters of ancient architecture components, the shape can be analyzed and the character can be extract. Then combined region growing algorithm all point cloud data can be segmented. The model of top-to-bottom was adopted in these algorithms. Segmentation algorithms combined with Gauss Domain and Spatial Domain. It was good anti-noise and self-adapted. At the same time, the characters of features can be extracted too so as to complete segmentation and fitting integrative.(2) An improved DBSCAN clustering algorithm-AQDBSCAN algorithm was proposed according the characters of point cloud data of ancient architecture. The parameter o of spatial ball radius can be calculated automatically and the speed of clustering can be improved with the same effect of DBSCAN in this algorithm. This method can resolve the problems in traditional DBSCAN algorithm such as low efficiency because of choosing parameter σ manually and only based on the relation of density and points’ numbers so as to divide some core point to noise.(3) A new rapid spatial index-MultyGrid-KD tree was proposed. The MultyGrid-KD tree combined the advantages of Octree and KD tree and overcame the disadvantages of Octree and KD tree. Experiments certificated that it is a high efficiency spatial index and the base of the nearest neighbor search.(4) Using parallel computing and multi-core technology, the main algorithm such as the construction of MultyGrId-KD tree, nearest neighbor search, calculation of differential geometry and AQ-DBSCAN clustering was optimized based on OpenMP Parallel programming model. The efficiency of them is compared.Using VC++2005, the ancient architecture point cloud data processing system was programmed based on Visualization Toolkit according to the theories proposed in this article. The system concluded input and output model,3D scene model, editing model, preprocessing model, segmentation model, feature fitting model and triangulation construction model. Using the data of Forbidden City the algorithms were certificated on this software.