基于地面LiDAR数据的建筑物立面提取及建模研究

发布时间：2017-12-28 11:19

本文关键词：基于地面LiDAR数据的建筑物立面提取及建模研究　出处：《东华理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：建筑物作为城市三维建模的重要目标,其位置边界信息在目前地图更新与导航、房产规划等应用方面起着重要的作用。地面三维激光扫描仪能够获取大面积高分辨率的被测对象表面海量三维数据,但原始三维激光扫描数据中除了建筑物立面目标点云外,还包含一些地面、树木、行人、车辆和城市的部分基础设施点云等非建筑物点云数据及其他噪声点,需要进一步提取建筑物立面信息,以便对目标建筑进行三维建模。因此,如何高效的从大量点云数据中快速准确的提取出建筑物立面,成为了许多学者研究的热点方向,是“数字城市”建设工作中的重要环节,提取准确度决定着建筑物后期模型的重建精度。随机抽样一致性算法(Random Sample Consensus)作为一种鲁棒性强的算法,在数据错误率超过50%时仍然能够得到理想的处理结果,能有效抑制噪声点的影响,提取出正确的特征线和特征面,是一种稳健、高效的从样本集中拟合数学要素的方法,在基础矩阵估计、特征匹配、运动模型选择等计算机视觉领域内有着广泛应用。目前RANSAC算法多用于机载和车载LiDAR数据处理上,在地面三维激光扫描仪获取数据中的研究较少。传统的RANSAC算法需要事先确定阈值,在进行建筑物立面信息的平面点云提取过程中,阈值的选取对平面提取准确性有很大影响。论文在简要介绍地面三维激光扫描技术的基础上,阐述了RANSAC算法的原理,针对传统RANSAC算法的特点和不足,对传统的基于RANSAC方法提取建筑物立面进行了改进。采用基于点云密度值和半径密度优化了RANSAC算法,通过实验完成了一建筑物立面提取。实验结果表明改进后的方法在建筑物立面提取的准确度上有明显的提高。研究了曲面重构法和参数法两种不同的建模方法,分别对获取的两个点云数据进行了模型重建,阐述了两种模型重建方法具体的操作步骤,并将两种建模技术进行了结合使用,最后对两种建模技术进行对比分析。
[Abstract]:As an important goal of urban 3D modeling, the location and boundary information of buildings plays an important role in map updating and navigation, real estate planning and other applications. The measured data of 3D object surface terrestrial 3D laser scanner to obtain a large area of high resolution, but the original 3D laser scanning data in addition to building facade object point cloud, also contains some ground, trees, pedestrians, vehicles and parts of the city infrastructure and other non point cloud building point cloud and other noise, need further extraction of building facade information for 3D modeling of buildings. Therefore, how to extract building facade quickly and accurately from a large number of cloud data has become a hot research direction of many scholars. It is an important link in the construction of "digital city". The accuracy of extraction determines the accuracy of reconstruction. Random consistency algorithm (Random Sample Consensus) as a robust algorithm, still can get ideal results rate of more than 50% in the data error, can effectively suppress the influence of noise, feature extraction and feature line is correct, is a robust and efficient method on mathematical fitting elements from the sample, based on matrix estimation, feature matching and motion model selection in the field of computer vision has been widely used. At present, the RANSAC algorithm is mostly used in airborne and vehicle LiDAR data processing, and there is less research in the acquisition of data in the ground three-dimensional laser scanner. The traditional RANSAC algorithm needs to determine the threshold beforehand. During the process of building elevation information extraction, the threshold selection has great influence on the accuracy of plane extraction. Based on the brief introduction of the terrestrial 3D laser scanning technology, this paper expounds the principle of RANSAC algorithm, and improves the traditional RANSAC algorithm based on the RANSAC method. The RANSAC algorithm is optimized based on the point cloud density value and the radius density, and a building facade is extracted by experiment. The experimental results show that the improved method has an obvious improvement in the accuracy of the building elevation. The study of two different modeling methods of surface reconstruction method and parameter method, respectively on the two point cloud data acquisition of model reconstruction, expounds the operation steps of two kinds of model reconstruction method, and two kinds of modeling techniques were used in combination, at the end of the two kinds of modeling techniques were compared and analyzed.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P225.2;TU198

【参考文献】