Self-Localization of Unmanned Aerial Vehicles Using Visual Inertial Odometry


Jan Bednář

Supporting material for the master's thesis



This thesis is concerned with visual-inertial odometry (VIO) algorithms and their usability and suitability for integration into the feedback loop of the unmanned aerial vehicle (UAV) control system. The main part of the thesis is focused on the comparison of chosen VIO algorithms in terms of pose estimation precision for chosen camera mounting orientations, camera frame rate, UAV velocity and the feedback suitability. According to prior survey of VIO algorithms precision, availability and fitness on UAV deployment, three VIO algorithms are chosen for this thesis, namely S-MSCKF, SVO, and VINS-Fusion. The VIO algorithms performance is evaluated in both the simulation environment and the real environment to prove the suitability for feedback loop integration. The trajectory-shaping filter was implemented to smooth the trajectory by constraining the accelerations according to the UAV dynamics.
Such filter improves not only the precision of VIO pose estimation but also the similarity of the control reference generated from the tracked trajectory. Lastly, the feedback integration for tested algorithms is presented for all used VIO algorithms in the simulator and partially in the real deployment.


Video material from the deployment on the real UAV


Video material from the deployment on the UAV in the Gazebo simulation environment