面向高分辨率激光扫描点云的三维建筑物风格化渲染方法  

A method on stylized rendering of 3-D building via high-resolution laser scanning point clouds

作　　者：唐翔[1] 王伟鹏 罗欢 郭迎亚 TANG Xiang;WANG Weipeng;LUO Huan;GUO Yingya(College of Architecture and Urban-rural Planning,Fuzhou University,Fuzhou,Fujian 350108,China;College of Computer and Data Science,Fuzhou University,Fuzhou,Fujian 350108,China)

机构地区：[1]福州大学建筑与城乡规划学院,福建福州350108 [2]福州大学计算机与大数据学院,福建福州350108

出　　处：《福州大学学报（自然科学版）》2021年第6期761-766,共6页Journal of Fuzhou University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61801121);福建省自然科学基金资助项目(2019J05034)。

摘　　要：针对点云的非真实感渲染问题,提出一种面向高分辨率点云数据的风格化渲染方法,以提高点云数据展示的艺术性.首先,针对高分辨率点云中海量三维点所带来的计算量问题,提出远距离点采样方法,在实现精简数据的同时保留其三维结构特征;然后,针对跨维度的风格迁移问题,引入基于深度神经网络的风格化渲染方法,实现二维图像至三维网格的风格迁移;最后,为了快速且高效地将可视化风格信息从三维网格向三维点云进行传递,提出基于K-D树的网格数据与点云数据的融合方法,以建立两种数据之间的映射关系.实验结果表明,所提出的方法可有效地将给定二维图像的风格转化为目标三维建筑物的风格,增强了高分辨率三维建筑物点云数据的展示效果.Aiming at solving the non-photorealistic rendering problem of high-resolution laser scanning point clouds,a stylized rendering method is proposed to render the point clouds to improve the visual effect of high-resolution 3 D building point clouds.Firstly,to handle computational burden brought by the large amount points in the high-resolution point clouds,a farthest point sampling method is proposed to reduce the number of points while preserving its 3 D structural features.Then,to conduct the cross-dimensional style transfer,a neural network-based style rendering method is introduced to achieve the visual style transferred from 2 D images to 3 D mesh.Finally,in order to transfer visualization information from 3 D mesh to 3 D point clouds quickly and efficiently,we proposed a K-D tree based fusion method to establish the mapping relationship between 3 D mesh and 3 D point cloud.Experiments demonstrate that the proposed framework can effectively transform the style of a given 2 D image into a 3 D building style,which promotes the visual effects of the high-resolution laser scanning point clouds of buildings.

关 键 词：风格化渲染 非真实感渲染 高分辨率扫描点云数据 三维建筑物模型 点云采样 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]