Abstract

This paper presents an automatic reconstruction algorithm of surfaces of revolution (SORs) with a self-adaptive method for generatrix line extraction from point clouds. The proposed method does not need to calculate the normal of point clouds, which can greatly improve the efficiency and accuracy of SORs reconstruction. Firstly, the rotation axis of a SOR is automatically extracted by a minimum relative deviation among the three axial directions for both tall-thin and short-wide SORs. Secondly, the projection profile of a SOR is extracted by the triangulated irregular network (TIN) model and random sample consensus (RANSAC) algorithm. Thirdly, the point set of a generatrix line of a SOR is determined by searching for the extremum of coordinate Z, together with overflow points processing, and further determines the type of generatrix line by the smaller RMS errors between linear fitting and quadratic curve fitting. In order to validate the efficiency and accuracy of the proposed method, two kinds of SORs, simple SORs with a straight generatrix line and complex SORs with a curved generatrix line are selected for comparison analysis in the paper. The results demonstrate that the proposed method is robust and can reconstruct SORs with a higher accuracy and efficiency based on the point clouds.

Highlights

  • Three-dimensional (3D) surface reconstruction of objects has been a popular research area in computer vision and remote sensing for a long time [1]

  • With the purpose of reconstructing the initial 3D model of surface of revolution (SOR) for ancient artifacts and important structural components of ancient architectures from point cloud by terrestrial laser scanning (TLS), this paper explores an automatic reconstruction algorithm of SORs with a self-adaptive method for generatrix line extraction from point clouds

  • The percentage improvement of time efficiency is 25.1% and 28.1% for the reconstructed short-wide SORs of a pot and a ceramic, respectively, which is similar to reconstructed tall-thin SORs. This indicates that the percentage improvement of time efficiency is almost identical for both tall-thin SORs and short-wide SORs by as shown in Figure 13c, it is a sloping reconstructed SOR compared with the original point cloud, which is caused by the poor quality of the acquired normal of the point cloud of the pot by the curvature computation method

Read more

Summary

Introduction

Three-dimensional (3D) surface reconstruction of objects has been a popular research area in computer vision and remote sensing for a long time [1]. To reconstruct a SOR is to determine its rotation axis and generatrix line from point clouds by TLS [18]. For the determination of generatrix lines for SORs reconstruction, the most commonly used method is to estimate the surface normal at the data points by calculating the curvatures of point clouds of geometric objects and solve certain approximation problems in the space of lines or the space of planes [13]. The most commonly used method to obtain the surface normal from point clouds is total linear least squares, which is relatively cheap to compute and easy to implement It becomes more computationally expensive with the increase of measurements per second generated by TLS sensors [25]. 33ooff 2210 rotation axis of a SOR by a minimum relative deviation among the three axial directions for both tall tshhionrta-nwdidsehoSrOt-RwsidauetSoOmRastiacaultloym; (a2t)iceaxltlrya;c(t2t)heexptrraocjtecthtieonprporjoefictlieonofparoSfOilRe obfyatrSiaOnRgublyatterdianirgrueglautleadr inreretwguolrakr(nTeItNw)ocrokn(sTtrIuNc)ticoonnasntrducrtainodnoamndsarmanpdleomconsasmenpsluesc(oRnAseNnSsAusC()RaAlgNorSiAthCm);aalngdor(i3th) mde;taenrmd i(n3e) dtheetegremnienreattrhixe lgineneeorfaatrSixOlRinbeyosfeaaSrcOhRinbgyfosreathrceheixntgrefmorutmheoefxctoreomrduinmatoefZc,otoorgdeitnhaetrewZi,thtoagnetohveerrwfloitwh apnoionvtserpfrloocwespsioning,tsanpdrofcuerstshienrgd, eatnedrmfuinrtehtehredteytpeermofingeenthereattyripxelionfegbeyntehreatsrmixallilneer RbyMtSheersrmoraslbleertwReMenS elirnreoarrs fibtettiwngeeanndlinqeuaardfritattiincgcuanrvdeqfiutatidnrga.tic curve fitting

102 2. Methods
Extraction of Rotation Axis μ
Extraction of the Point Set of Boundary X
Generatrix Line Fitting
Experiments and Analysis
Comparison with the Curvature Computation Method
Tall-Thin SORs
Short-Wide SORs
Comparison with Surface Reconstruction Methods
Accuracy Analysis
Findings
Conclusions
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call