Over recent years, the possibility of reconstructing three dimensional models from images has led to a very active research field. This has culminated in systems that are capable of largely automated reconstruction of 3D from image sequences even when nothing at all is known about the cameras or scene except for the images themselves. This thesis attempts to improve the state of the art by refining and adding to the techniques used for this process.
The main novel contributions are a method for tracking image features, a means of selecting from video sequences so as to reduce computational load and reduce degeneracy and a new approach to projective reconstruction and rectification. These techniques are combined with existing techniques to create a complete system capable of reconstructing dense 3D models from image sequences alone. Examples of the systems use with real data are given.
NOTE: This research was carried out with a grant obtained as part of the REVEAL project in the Advanced Interfaces Group of Manchester university. Later work led to the notable ICARUS project.
Keywords: Projective Reconstruction, Calibration, Rectification, Point Matching
Download as pdf (7.94Mb)Some of the amazing reviews my thesis has had:
'Awesome, amazing, brilliant, hilarious. I almost died with laughter during the section on rectification' The List
'I'll never be able to look at Tensor algebra again without smiling' The Metro
'top-notch comedy that's more cheeky than nasty' Chortle
'Clever, subtle and original with a gently iconoclastic twist' Sunday Times