Development of Monocular Visual Odometry Framework for KuupKulgur

Abstract

This thesis presents the development of a monocular Visual Odometry (VO) framework. The work is motivated by the KuupKulgur, the Estonian lunar rover project, for which a extendible monocular VO framework is required as part of the overall localisation system to aid its autonomous capabilities. The developed framework needs to be adaptable to different tunable parameters of VO (including different feature detectors and matchers). A C++ framework has been developed as part of this thesis. A performance analysis of different sub-components of the overall VO framework has been conducted on targeted hardware to identify possible bottlenecks in the implementation. The overall VO implementation has also been tested on the KITTI dataset and a sample dataset collected by the KuupKulgur rover at the lunar analog facility at Tartu Observatory.