Research And Development Of Rapid Modeling System Based On 3D Point Cloud Data

In recent years,with the increasing popularity of 3D laser scanning technology,automatic modeling of 3D point clouds of industrial scenes is crucial for applications like digitalization of real scenes combining with virtual reality technology for navigation,information inquiry and supervisory control.However,this problem is challenging due to poor-quality input industrial scenes data suffering from incompleteness,noise and anisotropy in density.In response to these challenges,we take large-scale electrical substation scenes as the specific research targets and propose a new framework to fast reconstruct these scenes.Our framework is characterized by automatical scene analysis process and model retrieval technique based on our own electrical device database which significantly improves the accuracy and efficiency of electrical substation scene reconstruction.In order to achieve the goal of developing the rapid modeling system,pre-processing,repeated structure detection and shape description of point cloud are studied in this paper,sequentially.First,Principal Component Analysis and Gauss hemisphere mapping method are applied to detect the ground data,which improves the integrity of the ground detection.We next propose a power line extractor by reasonably integrating some common data processing techniques.For substation scenarios with ground and power lines removed,the improved DBSCAN clustering algorithm is used to separate the device data in the scene.Subsequently,we propose a graph-based pattern recognition algorithm,which recognizes various repeated patterns in the input scene and derives a flexible and hierarchical representation followed by extracting various device individuals from the scene.For each electrical device point set,a mesh model from our database is retrieved based on the similarity of object descriptors which well balance the descriptiveness,the robustness and the efficiency.With the high-level representation of the scene,we place all retrieved models into the input data to achieve the faithful reconstruction.Finally,we research the visualization technology for reconstruction,and develop the rapid modeling system based on 3D laser point clouds by integrating all techniques presented above.The modeling system makes full use of the structural features of the equipment in industrial scenes,which reduces the complexity of modeling while improving the efficiency of scene modeling,and is of great significance to the digitization of practical industrial scenes.