Solutions and their ambiguities for structure and motion problems

This thesis revolves around the problem of solving the structure and motion problem in computer vision. This is the problem of reconstructing the world, the structure, from images taken by a number of cameras undergoing an unknown motion. Much of the thesis is centred on the problem of finding initial estimates of the structure and the motion using a small set of image data. These initial estimates can then be extended and optimized using non-linear bundle adjustment. These initial estimates can be found by solving minimal problems, i.e. problems with exactly enough data to constrain a solution. Chapters 3-7 are devoted to the analysis, classification and solution of minimal structure and motion problems for different types of cameras and different types of image features. The aim is to classify and solve all possible structure and motion problems.Large parts of the thesis concerns vision based on cameras with one-dimensional retinae. A one-dimensional camera takes one-dimensional images of a two- or three-dimensional scene.Chapters 8-11 describe research and development of an automatic system for structure and motion estimation for a laser guided vehicle. The imaging system of a laser guided vehicle can be modelled by a one-dimensional camera. Chapter 9 and 10 covers the problem of automatic geometric reasoning which is an essential part of making the system described in chapter 11 more robust