טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
Ph.D Thesis
Ph.D StudentMiropolsky Alexander
SubjectAn Adaptive 3D Reconstruction Method Utilizing Diverse Data
Types Inherent To Emerging Scanning Technologies
DepartmentDepartment of Mechanical Engineering
Supervisor Professor Anath Fischer
Full Thesis textFull thesis text - English Version


Abstract

3D reconstruction of computerized models using Reverse Engineering (RE) is an important issue in modern industry that has the potential to significantly improve design, inspection and analysis applications. To meet the demands of modern industry, reconstruction must be automatic, fast, accurate and robust. A computerized model is usually reconstructed from scan data in the form of digitized points provided by 3D scanners. Scan data, however, is typically very large scale (i.e. many points), unorganized, noisy and incomplete. For these reasons, surface reconstruction is a difficult but challenging task. Scanning technologies that have emerged in recent decades are capable of capturing additional information about the sampled object, such as surface normals, color and other geometrical and physical properties. Most existing reconstruction methods, however, do not utilize this additional information. Thus, one way of overcoming the problems outlined above is to develop reconstruction methods that exploit diverse feature data, that is, diverse information about the properties of the scanned object. This research develops this concept and proposes a new reconstruction approach that can inherently handle a variety of scan information, thus providing simple and effective solutions for surface reconstruction. Based on this approach, this work proposes a new Diverse Data Surface Reconstruction (DDSR) method, comprising the following stages: 1) scanning; 2) scanned point denoising by Geometric Denoising Filter (GDF); 3) creating Hierarchical Space Decomposition Model (HSDM); 4) detecting sharp features by Sharp Feature Detection (SFD) method; 5) reducing data by Geometric Reduction Filter (GRF) and reconstructing missing points by Geometric Prediction Filter (GPF); 6) creating connectivity graph; 7) mesh generation. The proposed method is automatic, fast and accurate. The method is capable of completing the missing diverse data and can therefore be applied for different scanning technologies. The feasibility of the proposed method is demonstrated on a number of complex engineering and medical objects.