Autonomous visual-inertial navigation and absolute visual scale estimation

In this thesis, we present a system that uses a single camera and an inertial measurement unit (IMU) to navigate an Unmanned Aerial Vehicle (UAV) in a previously unknown environment. The approach consists of two parts. First, we apply a state-of-the-art simultaneous localization and mapping (SLAM) method to the video stream of a onboard camera. From the SLAM system, an up-to-a-scale pose of the camera is estimated, because the absolute size of the environment cannot be estimated with a single camera. Second, the estimated pose is fused with the data from IMU to resolve the scale ambiguity. While analyzing the performance of the system, we find that the conver- gence rate of scale decreases when the magnitude of scale increases. This relationship has not been demonstrated and explained before. In this thesis, we present an analysis and explanation of this phenomenon