Tehnologije za zajem podatkov snovnega sveta v informacijska okolja za gradbeništvo : magistrsko delo

Jotanović, Uroš (author), Turk, Žiga (mentor), Klinc, Robert (co-mentor), Kregar, Klemen (co-mentor)

Uporaba oblakov točk pri pridobivanju podatkov za informacijsko modeliranje stavb je v zadnjem času vedno bolj pogosto. Upravljavci in lastniki starejših stavb in objektov si prizadevajo za centralizacijo informacij. Dokumentacija o strojnih inštalacijah, cevovodih, električni opremi in preteklih posegih je pogosto shranjena na dislociranih medijih, v večini primerov pa še vedno na papirju. Oblak točk v procesu pretvorbe snovnega sveta v digitalni predstavlja izhodišče, v katerem so zajeti podatki o snovnem svetu, ki jih pridobimo na več različnih načinov, kot so fotogrametrija, terestrično lasersko skeniranje in zračni lidar. Ročno modeliranje BIM za upravljanje, vzdrževanje in nadaljnjo uporabo je zamuden postopek, nagnjen k napakam, zato želimo ta proces avtomatizirati in se tem napakam izogniti. V zadnjem času se pojavlja vedno več samostojnih programov in dodatkov za obstoječe programe, ki omogočajo avtomatizirano, hitro in bolj natančno modeliranje na osnovi surovih podatkov iz oblakov točk. Princip delovanja teh programov in dodatkov izhaja iz modelov računalniškega vida, ki omogočajo poenostavitev dela in zmanjšanje vloge uporabnika v celotnem procesu. V magistrski nalogi smo raziskovali možnosti za avtomatizacijo procesa izdelave BIM modelov iz surovih oblakov točk. V prvem delu naloge smo primerjali izmenjevalne formate za zapis oblakov točk, v katerih so na različne lahko načine zapisani prostorski podatki. Primerjali smo različno programsko opremo, ki omogoča delo z oblaki točk in modeliranje na njihovi osnovi. Raziskali smo trenutne algoritme računalniškega vida, ki so osnova teh programskih rešitev. V drugem delu smo na konkretnih primerih oblakov točk prikazali postopek modeliranja posameznih BIM elementov. V postopku modeliranja smo rezultate za posamezne elemente (stene, cevi in stebre) grafično prikazali in dokazali visoko stopnjo natančnosti avtomatiziranega modeliranja. Kljub avtomatizaciji celotnega procesa je za doseganje dobrih rezultatov še vedno potrebna uporabniška interakcija, ki pa se z algoritmi globokega učenja v zadnjih letih zmanjšuje.

• Language: Slovenian
• Keywords: oblak točk, segmentacija, klasifikacija, fotogrametrija, terestrično lasersko skeniranje, BIM
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FGG - Faculty of Civil and Geodetic Engineering
• Place of publishing: Ljubljana
• Publisher: [U. Jotanović]
• Year: 2021
• PID: 20.500.12556/RUL-127193
• UDC: 004.6:004.76:624(043.3)
• COBISS.SI-ID: 77394691
• Publication date in RUL: 24.05.2021

Secondary language

• Language: English
• Title: Technologies for capturing material world data into civil engineering information systems : master thesis

Abstract

The use of point clouds in data acquisition for building information modeling has become common practice in recent years. Managers and owners of older buildings strive to centralize information about their assets. Current documentation regarding mechanical installations, pipelines, electrical equipment, and past interventions is often stored on dislocated media or on paper in many cases. Point cloud represents the starting point in the process of transforming the material world into digital, which includes raw data about the material world, that is obtained in several different ways, such as photogrammetry, terrestrial laser scanning and air lidar. Manual BIM modeling for management, maintenance and further use is a time-consuming, error-prone process, that we strive to automate. Recently, more stand-alone programs and add-ons for existing programs are emerging, allowing for fast, automated, and more accurate modeling based on raw point cloud data. The principles of these programs and add-ons are based on computer vision models that simplify the work and reduce the role of the user in the entire process. In the master's thesis, we explored the possibilities of automating the creation process of BIM models from raw point clouds. In the first part of the thesis, we compared point cloud exchange formats, in which the same spatial data can be written in different ways. We compared different software that allows working with point clouds and modeling. We explored the current computer vision algorithms that serve as basis for these software solutions. In the second part, we presented the process of modeling individual BIM elements on point cloud examples. During the modeling process, we graphically presented the results for individual elements (walls, pipes and columns) and proved the high level of accuracy for presented process. Despite the high level of automation during the proess, user interaction is still required to achieve good results. The need for user interactions is lowering with the advancements of the deep learning algorithms in recent years.

• Keywords: point cloud, segmentation, classification, photogrammetry, terrestric laser scanning, BIM