An Appearance-preserving Simplification Method For Complex 3D Building Models

In virtual geographic environments and 3D city landscapes,3D building models are an important component and research target which provide good visual effects and interactive experiences.Over recent years,3D building models have been applied in a wide range of fields,including urban planning,3D navigation,and disaster managements.These visual applications all require that the rendering of 3D building models should not only guarantee strong fidelity,but also meet the demands of real-time transmission and dynamic interaction.With the rapid development of 3D acquisition techniques,especially LiDAR and oblique photogrammetry,the data amount of 3D building models has increased dramatically.This has made it more difficult for computer graphics hardware with limited performance to process 3D data,leaving a significant challenge to the real-time rendering of 3D models.To solve this problem,the concept of levels of details(LODs)are proposed.The key to generating reasonable LODs for 3D building models is to simplify the models,thus reducing their geometric complexity and decreasing the amount of data required.Compared with free-form surface models,3D building models usually have stricter geometric constraints,such as perpendicular,parallel,and coplanar relationships.Besides,some building models may contain complex texture images and semantic information.These characteristics present more limits and challenges to the simplification of 3D building models.General simplification algorithms proposed in computer graphics area can obtain good simplification results for free-form surface models,while their application to building models may violate geometric and constraints and topological structures.In recent years,some 3D building simplification algorithms mainly consider geometric complexity which aim to create simplified models whose silhouettes are similar to the original models.Other simplification methods take texture into consideration,but the models they used are mostly regular and simple.In general,there has been litter research on the simplification of complex 3D building models considering both geometry and texture.Supported by National Natural Science Foundation of China"Study on Construction and Visualization of Continuous LODs of 3D Building Model",this paper proposed a simplification method for complex 3D building model to maintain the geometric and texture features of buildings.The main content is as below:(1)The data model of 3D building model and surface mesh segmentation.Vertex,Edge,and Triangle are chosen as the fundamental elements to represent the geometric structures of buildings.Considering the texture discontinuities of building models,the neighboring triangles of one vertex and its corresponding texture coordinates are stored in the form of mapping.A mesh segmentation algorithm based on Breadth-First Search is proposed.This algorithm can segment the original mesh into multiple regions with inner consistence using the information of topology and color deviation,and can provide important guidelines for subsequent model simplification.(2)A 3D model simplification method with both geometry and texture taken into consideration.Conventional quadric error metric only evaluate the geometric error during simplification,while this research presented an improved simplification error metric,in which the geometric and texture information are combined with mesh segmentation information to better maintain the original appearance of 3D building models.After each simplification operation,texture adjustment is conducted to repair the resulting texture distortion,thus improving visual quality of the simplified models.(3)Determining the optimal simplification rate and generating discrete LODs for 3D building models based on pixel error analysis.To calculate the optimal simplification rate under a certain viewing distance,in this study,simplification error is projected onto the screen to obtain the pixel error.By setting the pixel error threshold under a certain viewing distance,the simplification of a building model can be acquired.Furtherly,getting corresponding simplified models under multiple viewing distances can generate a discrete LOD sequence.The pixel error analysis method can effectively eliminate the differences in spatial scale of different 3D building models,and incorporate the calculation of the optimal simplification rate into a unified framework.The experiments show that the proposed method can significantly reduce texture distortions.Compared with traditional simplification algorithms,the proposed method can achieve better simplification quality under the same simplification rate,and produce simplified models whose important geometric and texture features are well preserved,thus presenting good visual effects for 3D city landscapes.