3D model reconstruction from 2D images

• Main Author: Fong, Yi Qing
• Other Authors: Gwee Bah Hwee
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/77883
• Institution: Nanyang Technological University

Three-dimension (3D) reconstruction is an important and challenging area of study that has been actively researched on over the last few decades. It has brought a wide range of applications that could be applied across various fields. The objective of this Final Year Project is to understand the 3D reconstruction process and experiment with surface and sub-surface reconstruction. The project also aims to explore the possibilities of enhancing its usage for real-life applications and seek to minimize the likelihood of workplace hazards from occurring. This report will evaluate 1) the Structure from Motion technique for surface 3D reconstruction and 2) the use of Ground Penetrating Radar (GPR) microwave imaging for sub-surface reconstruction to detect and recreate objects buried in an underground simulation. Additionally, the extraction of spatial information from two-dimension (2D) images would be attempted to provide useful information about the actual object such as estimating its physical measurement (e.g. length) in images through object detection. This allows distance measurements to be obtained without physically measuring the object. Experiments for Structure from Motion were conducted for various objects. An observation was the two views technique producing results that were more desirable and accurate as compared to multiple views’. However, multiple views technique managed to capture more colours and construct more realistic looking models. As for sub-surface reconstruction, GPR depth slices acquired through microwave imaging were used in attempt to recreate the buried objects. A model was constructed based on the reflections captured in the depth slices. Apart from 3D reconstruction, object distance measurement was performed on 2D images and the results were relatively accurate. The best results attained only differed 1.21mm and 3.75mm from the actual object size and camera distance respectively.