| The goal of this thesis was to develop algorithms that would enable a mobile robot to take advantage of the wealth of information contained in the images acquired by an onboard camera. The thesis presents novel algorithms for solving two very important problems that a mobile robot would encounter, structure and motion from line correspondences and vision-based exploration and navigation. Both of these algorithms could be used as modules in a larger mobile robot control system.; The first problem, structure and motion from line correspondences, can be stated as follows: given a set of line correspondences obtained from multiple images, reconstruct an accurate three dimensional model of the scene. The recovery algorithm is formulated in terms of an objective function which measures the total retinal disparity between the observed edge segments and the projections of the reconstructed lines. This objective function is minimized with respect to the line parameters and the camera positions to obtain an estimate for the structure of the scene.; The main contributions of this research have been the formulation of an appropriate objective function which effectively measures the squared distance between the observed edges and the projected lines and the development of an effective algorithm for minimizing this objective function. The effectiveness of this approach is demonstrated quantitatively through an extensive series of simulations and qualitatively with actual image sequences. This algorithm has proven to be more accurate than any previously proposed technique. The results produced by this method are comparable to those obtained from a calibrated stereo system.; The second part of this thesis focuses on the problem of designing algorithms that would enable a mobile robot equipped with a visual recognition system to carry out a systematic exploration of an unfamiliar environment in search of one or more recognizable targets. As a by-product of this exploration process, the algorithms construct a representation of the environment that the robot can use in future navigation tasks. One of the main advantages of the proposed algorithms is that they do not require a global positioning system. This research demonstrates that it is possible to design provably correct exploration and navigation algorithms that are based on the robot's ability to recognize landmarks. The algorithms have been successfully implemented in simulation and on our mobile robot platform, RJ, and the results from these experiments are presented. |
