Date of Award
Master of Science (MS)
This thesis presents a novel method of indoor localization and autonomous navigation of Unmanned Aerial Vehicles(UAVs) within a building, given a prebuilt Computer Aided Design(CAD) model of the building. The proposed system is novel in that it leverages the support of machine learning and traditional computer vision techniques to provide a robust method of localizing and navigating a drone autonomously in indoor and GPS denied environments leveraging preexisting knowledge of the environment. The goal of this work is to devise a method to enable a UAV to deduce its current pose within a CAD model that is fast and accurate while also maintaining efficient use of resources. A 3-Dimensional CAD model of the building to be navigated through is provided as input to the system along with the required goal position. Initially, the UAV has no idea of its location within the building. The system, comprising a stereo camera system and an Inertial Measurement Unit(IMU) as its sensors, then generates a globally consistent map of its surroundings using a Simultaneous Localization and Mapping (SLAM) algorithm. In addition to the map, it also stores spatially correlated 3D features. These 3D features are then used to generate correspondences between the SLAM map and the 3D CAD model. The correspondences are then used to generate a transformation between the SLAM map and the 3D CAD model, thus effectively localizing the UAV in the 3D CAD model. Our method has been tested to successfully localize the UAV in the test building in an average of 15 seconds in the different scenarios tested contingent upon the abundance of target features in the observed data. Due to the absence of a motion capture system, the results have been verified by the placement of tags on the ground at strategic known locations in the building and measuring the error in the projection of the current UAV location on the ground with the tag.
Haque, Akkas, "Autonomous Localization Of A Uav In A 3d Cad Model" (2018). Theses and Dissertations. 2409.
Available for download on Thursday, January 02, 2020