This thesis provides a comprehensive description of the computer vision based simultaneous localization and mapping algorithms, as well as a method to detect and reconstruct the 3D position of edges and planar surfaces in the scene.
I describe the basic concepts of both feature based and direct simultaneous localization and mapping methods. For each method, a state-of-the-art algorithm is described and evaluated on two datasets from a handheld monocular camera. I also compare these algorithms with one another. Furthermore, I present methods that were developed to extract additional information and geometry from images. The thesis discusses the opportunities and problems of the utilization of SLAM algorithms on self-driving cars.
I propose a method to detect closed planar surfaces and calculate their positions in 3D space using only images and egomotion. This method is using one of the simultaneous localization and mapping algorithms to determine polygon correspondances and the camera motion.