Abstract:
Monocular Visual SLAM refers to the process of determining an agents pose using a single camera as a sensory input. Extensive research in the field for the past decade ensued a number of systems that found their ways into various applications, such as robotics and augmented reality. Although filter-based (e.g., Kalman Filter, Particle Filter) Visual SLAM systems were common at some time, non-filter based (i.e., using optimization) solutions, which are more efficient, have become the de facto methodologies for building any Visual SLAM system. The major contribution of this thesis is a comparative assessment of the state of the art in open source non-filter based mononocular Visual SLAM systems, namely PTAM, SVO, DT SLAM, LSD SLAM and ORB SLAM. Detailed experiments are presented for the SLAM comparison. To motivate this comparison, we present at the beginning of the thesis a case study, of a Visual SLAM application in an outdoor scene, in which the major problems of Visual SLAM are unearthed. The second major contribution of this thesis is the development of a scaled monocular SLAM in which depth from focus is used to determine the correct scale and maintain it through a SLAM trajectory. Real experiments are also performed and the obtained results prove the viability of the proposed method.
Description:
Thesis. M.E. American University of Beirut. Department of Mechanical Engineering, 2016. ET:6351
Advisor : Dr. Daniel Asmar, Associate Professor, Mechanical Engineering ; Members of Committee : Dr. Imad Elhajj, Associate Professor, Electric and Computer Engineering ; Dr. Elie Shammas, Assistant Professor, Mechanical Engineering.
Includes bibliographical references (leaves 81-87)