Automatic Dense Reconstruction from Uncalibrated Video Sequences

Abstract: This thesis describes a system that completely automaticallybuilds a three-dimensional model of a scene given a sequence ofimages of the scene. The system also estimates the internalparameters of the camera and the poses from where the originalimages were taken. Results that have been produced from realworld sequences acquired with a handheld video camera arepresented.The main contribution of the thesis is in building acomplete system and applying it to full-scale real worldproblems, thereby facing the practical difficulties of far fromideal imagery. Contributions are also made to several systemcomponents, most notably in dealing with variable amounts ofmotion between frames, auto-calibration and densereconstruction from a large number of images. Thesecontributions are presented as appended papers to enable theexperienced reader to easily study the novelty of the thesis.The main text gives a detailed coherent account of thetheoretical foundation for the system and its components.There are several motivations for constructing systems ofthe proposed type. One motivation is to make it possible forany amateur photographer to produce graphical models of theworld with the use of a computer. The viewer of the materialcan then navigate through the model and view it from any point.Another application is the insertion of synthetic objects intoan existing video sequence. This task is frequently carried outin movie making but is then performed with a great deal ofexpensive manual work. A quite futuristic but highlyinteresting application is augmented reality where theuser?s view of the world is augmented by the insertion ofsynthetic objects.