基于NeRF的树木数字化重建

The digital reconstruction of trees based on NeRF

我国有数量巨大的古树和城市大树,如何合理构建树木数字化模型,是评估树木抗风能力和安全性的重要前提。鉴于风向的多变性,需要基于该模型能够从任意角度获得树木树冠的各项数据。由于激光雷达价格高昂、专业性强,不便于大规模推广,采用便宜、简便的手机拍摄树木图像。鉴于传统的图像三维重建技术不适合树木这样结构复杂、高反射率、多孔的物体,采用NeRF(神经辐射场)技术,利用体素进行模型体渲染,对树木进行三维数字化重建,为从任意角度获得树木图像和树冠参数打下基础。同时,采用NeRF的instant-ngp加速算法,分析研究了背景、光照条件、图像数量等因素对建模质量的影响。研究结果表明,基于NeRF的树木数字化重建方法能有效避免传统三维重建的空洞、变形等问题,是一种可行的树木三维数字化模型构建方法。

There are a large number of ancient trees and urban trees in China,so how to build a reasonable digital model of trees is an important prerequisite for evaluating the wind resistance and safety of trees.In view of the variability of wind direction,it is necessary to obtain the different data of tree crown from any angles based on this model.Due to the high price and strong professionalism of Lidar,it is not convenient for its widely using,and a cheap and convenient mobile phone to capture tree images was used in this paper.In view of the fact that the traditional three-dimensional image reconstruction technology was not suitable for trees with complex structure,high reflectivity and porous objects,the researchers used nerf(neural radiation field)technology,which uses voxels to render the model volume and carry out three-dimensional digital reconstruction of trees,which can lay the foundation for obtaining tree images and canopy parameters from any angle.At the same time,the paper used instant ngp acceleration algorithm to analyze the influence of background,lighting conditions,image number and other factors on the quality of modeling.The results show that the tree digital reconstruction method based on nerf can effectively avoid the problems of cavity and deformation in traditional three-dimensional reconstruction,which is a feasible method for building three-dimensional digital model of trees.