基于高噪声深度图像的树木模型重建

Tree reconstruction based on range image with strong noise

对现实世界中的植物模型进行真实准确的重建一直是计算机图形学的一个重要目标。基于激光扫描仪提供的深度图像的树木重建近年得到越来越多的关注。当扫描距离过远或者扫描精度较低的情况下深度图像带有较大的噪声,这给重建带来了很大的困难。本文提出了一种针对高噪声深度图像的树木自动重建技术,本算法以主曲率分析为基础,首先通过二次曲面拟合计算各个点对应的主曲率方向,之后利用各个点云与周围点云的主曲率相似度将噪声点云去除,利用点云分割技术将属于不同树枝的点云初步分离,最后从分割数据中提取树枝的骨架点以及半径,完成重建。利用本方法 ,我们对一个具有高噪声的深度图像进行了处理,从中提取出了树木的三维模型。本算法能够利用之前算法无法处理的高噪声的深度图像重建树木模型,降低了树木重建算法对输入数据精度的要求,同时减少了手工交互。

Accurately modeling of real plants is an important goal of computer graphics. Tree reconstruction based on range image obtained by laser scanner is getting more and more attention in recent years. When the distance between trees and laser scanner is large or the scan accuracy is low, there is strong noise in the range image, which brings great difficulty to the reconstruction. An automatic tree reconstruction method based on range image is presented in this paper. This method is based on analyzing principal directions. Firstly, principal directions and principal curvatures are estimated with quadric surface fitting. Noisy points are removed based on similarity between neighbor points. Then points from different branches are segmented and main skeleton nodes and corresponding radius are computed from each branch. We tested our algorithm with a range image with big noise and obtained an accurate 3D model. This method can reconstruct trees’ models from range image with strong noise which can not be processed by previous available methods. Moreover, it also reduces the requirement for range image accuracy and needs less interaction.