Comparison of LiDAR Building Point Cloud with Reference Model for Deep Comprehension of Cloud Structure

Abstract

This paper studies the fidelity level of the extracted Lidar (LIght Detection And Ranging) building point cloud in relation to the original building.In this context, the building point cloud is compared with a reference model. This comparison allows a deep understanding of the point cloud structure with respect to both the actual building and the constructed model. Consequently, the source of the incompatibility in a (reference or constructed) building model is determined and described. Thus, this study considers four aspects of the building point cloud.First, the errors of building point cloud extraction and the undesirable points are quantified. Hence, it is found that the percentages of undesirable points are sometimes considerable (between 15% and 70%). Second, the evaluation of the fit of the altimetry with the reference model shows that the roof plane equations calculated from Lidar data can be more precise than those of reference model, Third, it is noted that the segmentability level between different point densities and building typologies are variable. Finally, the per plane comparison mentions the incompatibility of the plane boundaries of point cloud with reference model. Moreover, considerable differences are noted between the theoretical and the true point densities.

Full Series Algorithm of Automatic Building Extraction and Modelling From LiDAR Data

“…In fact, the undesirable points sometimes represent an obstacle to the modelling algorithms, especially when their percentage exceeds certain limits. Sometimes they are considered as the main reason for automatic modelling failure or the source for some deformations in the obtained building model [21]. Second, it considers an important case, which is the case of high trees associated with the building roof.…”

“…The outcome of airborne laser scanning is a 3D point cloud. This point cloud provides the input data for a series of operations that aims to classify and then model the points into their appropriate classes such as terrain, buildings and vegetation [21]. In literature, from 1995 until the present, hundreds of research papers suggested automatic approaches for achieving the automatic LiDAR data classification and modelling [28].…”

“…Secondly, separation of building points (e.g., roof, wall, and façade points) from the noise-free point cloud [4]. Finally, generation of 3D building models from the extracted building points [5,21].…”

“…The reconstructed 3D building models must be followed by an accuracy estimation process to measure the reconstructed buildings' fidelity to the original or reference building. Several techniques have been proposed and used for this which includes both visual and geometric estimation processes [21,22,23]. The visual estimation techniques use several parameters, such as voting from human experts, fuzzy set theories, and the quality of estimated dominant directions [23,24].…”

“…The visual estimation techniques use several parameters, such as voting from human experts, fuzzy set theories, and the quality of estimated dominant directions [23,24]. The geometric accuracy estimation methods take the reference and reconstructed building models and calculate feature-based and LiDAR-based accuracy measures [21].…”