Rekonstrukcija 3D oblike iz premikov

Danov, Kristijan (author), Slavič, Janko (mentor)

V zakljčni nalogi obravnavamo, kako iz slik izdelati hitro, natančno in skalabilno 3D rekonstrukcijo. Kot rešitev uporabimo metodo strukturo iz gibanja (ang. Structure from Motion), pri čemer predstavimo postopek kalibracije kamere ter odprtokodni program Meshroom. Za to nalogo smo razvili odprtokodni Python paket OracleDepth, ki združuje OpenCV in Meshroom z namenom izboljšanja hitrosti in kakovosti rekonstrukcij. Na koncu podamo analizo rezultatov, pridobljenih z OracleDepth.

• Language: Slovenian
• Keywords: računalniški vid, 3D rekonstrukcija, numerične metode, kalibracija kamere
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FS - Faculty of Mechanical Engineering
• Year: 2025
• Number of pages: XI, 32 f.
• PID: 20.500.12556/RUL-172022
• UDC: 531.718:004.93:004.925.8(043.2)
• COBISS.SI-ID: 249221635
• Publication date in RUL: 05.09.2025

Secondary language

• Language: English
• Title: Reconstructing 3D geometry from motion

Abstract

In this thesis, we examine how to obtain a fast, precise, and scalable 3D reconstruction from images. We use a method called Structure from Motion and describe the process of camera calibration as well as the open source software Meshroom. As part of this work, we developed an open source Python package called OracleDepth, which combines OpenCV and Meshroom with the aim of achieving faster and higher quality reconstructions. Finally, we analyze the reconstruction results obtained with OracleDepth.

• Keywords: computer vision, 3D reconstruction, numerical methods, camera calibration