Using multiple view geometry for transmission tower reconstruction

Automated platforms that conduct power line inspections need to have a vision system which is robust for their harsh working environment. State-of-the-art work in this field focuses on detecting primitive shapes in 2D images in order to isolate power line hardware. Recent trends are starting to explore 3D vision for autonomous platforms, both for navigation and inspection. However, expensive options in the form of specialised hardware is being researched. A cost effective approach would begin with multiple view geometry. Therefore, this study aims to provide a 3D context in the form of a reconstructed transmission pylon that arises from image data. To this end, structure from motion techniques are used to understand multiple view geometry and extract camera extrinsics. Thereafter, a state-of-art line reconstruction algorithm is applied to produce a tower. The pipeline designed is capable of reconstructing a tower up to scale, provided that a known measurement of the scene is provided. Both 2D and 3D hypotheses are formed and scored using edge detection methods before being clustered into a final model. The process of matching 2D lines is based on an exploitation of epipolar geometry, where such 2D lines are detected via the Line Segment Detection (LSD) algorithm. The transmission tower reconstructions contrast their point cloud counterparts, in that no specialised tools or software is required. Instead, this work exploits the wiry nature of the tower and uses camera geometry to evaluate algorithms that are suitable for offline tower reconstruction. [Please note: the fulltext has been deferred until 9 December 2016]

Grantor dc:publisher.institution

Department of Electrical Engineering

Year dc:date.issued

2016

Author dc:creator

• Morarjee, Bhavani

Advisors dc:contributor.advisor

• Nicolls, Fred C
• Boje, Edward

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/11427/20463

OAI identifier oai:identifier

oai:open.uct.ac.za:11427/20463