Analysis of Visual-Inertial Odometry Algorithms for Outdoor Drone Applications
Sähkötekniikan korkeakoulu | Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Control, Robotics and Autonomous Systems
AEE - Master’s Programme in Automation and Electrical Engineering (TS2013)
AbstractVisual-inertial odometry (VIO) and visual-inertial simultaneous localisation and mapping (VISLAM) enables mobile robots to localise without relying on global navigation satellite systems (GNSS) or heavy sensors. They enable mobile robots, especially payload critical robots, such as drones, to perform autonomous tasks with limited resources. Localisation of drones for outdoor applications using visual and inertial sensor fusion is of particular interest, since it widens the use cases and reliability of autonomous drones in different flying conditions and environments. The goal of this thesis is to identify suitable VIO/VISLAM algorithms, and to develop a platform for localising a drone for outdoor applications. A stereo camera and IMU sensor suite was developed to collect visual-inertial data, since suitable off-the-shelf systems were not available. Three state-of-the-art VIO/VISLAM algorithms, FLVIS, ORB-SLAM3 and VINS-Fusion, were evaluated with outdoor drone datasets of varying flight altitudes of 40, 60, 80 and 100 m and speeds of 2, 3 and 4 m/s. The estimation results were compared with the ground truth and were quantitatively evaluated. VINS-Fusion estimated the trajectories most accurately among the three algorithms with an absolute trajectory error of 2.186 m and a relative rotation error of 0.862 deg at an altitude of 60 m for a trajectory of length 800 m. System configurations, algorithm parameters, external conditions, and scene content impacted the estimation results. These factors, further developments and future scopes are discussed along with the obtained results.
Thesis advisorHonkavaara, Eija
vio, vislam, drone, visual-inertial estimation, stereo-vision, localisation