OPTIMIZATION IN MONOCULAR VISUAL ODOMETRY

• Main Author: GILDRON, DAVID
• Format: Final Year Project Report
• Language: English; English
• Published: Universiti Malaysia Sarawak, (UNIMAS) 2022
• Subjects: Q Science (General)
• Online Access: http://ir.unimas.my/id/eprint/40104/1/Gildron%20David%2024pgs.pdf; http://ir.unimas.my/id/eprint/40104/7/Gildron%20David.pdf; http://ir.unimas.my/id/eprint/40104/

Presently, the world of autonomous vehicle is rapidly advancing as day by day, the features in many aspects of the car are made smarter and improved for safety, convenience and comfort. Apart of it, localization is one particular aspect of an autonomous vehicle which is especially important as it works to determine the precise location of the vehicle in the map, as maps are vitals nowadays to navigate. In details, the work of localization in the autonomous vehicle are done with visual odometry. As for this project, the focus is on monocular visual odometry (VO) as mono cameras are more preferred as for their low cost and easy to handle. The framework of the monocular visual odometry applied incorporated Oriented FAST and Rotated BRIEF (ORB) for the feature detection and the Flann-based matcher for feature matching. Optimizations done in this project was to tweak and adjust the system parameters which includes the value of features detected and distance between the matches, that is known as good matches. Overall, 3 different sequences from KITTI dataset are utilized in this simulation, the results were compared with the original system configuration and the ground truth. The optimized results show overall improvement of each individually sequence from their original ground truth. Entirely, the project aims to optimize the mono VO system which could help in the development of the mono VO in the future.