Optimization based visual-inertial slam and odometry systems in GPS-denied environments

Official URL: https://risc01.sabanciuniv.edu/record=b2718349_(table of contents)

Visual Inertial Simultaneous Localization and Mapping (VI-SLAM) and Visual Inertial Odometry (VIO) systems are widely used in various areas such as augmented reality, autonomous cars and aerial vehicles’ navigation systems where there is need for navigating the platform in the absence of the GPS information. There are many different configurations of the VI-SLAM and VIO systems in the literature in terms of the sensor types, methods that are used in estimating the states of the platform, sensor fusion methods and the front and back end structures of the visual-inertial systems. In this thesis, the focus will be on monocular graph optimization based VISLAM and VIO systems. For this purpose, end-to-end VI-SLAM and VIO structures have been built and the trajectory results have been evaluated using the Euroc-Mav dataset. Moreover, as a contribution to the current studies, a solution to the problem of neglecting dynamic objects in the environment has been proposed to increase the robustness of the visual-inertial navigation systems.