Monocular Visual Inertial SLAM for Drone Navigation
Promotor(s) : Béchet, Eric
Date of defense : 24-Jun-2021/25-Jun-2021 • Permalink :
|Monocular Visual Inertial SLAM for Drone Navigation
|Date of defense :
|Committee's member(s) :
|Engineering, computing & technology > Electrical & electronics engineering
|Université de Liège, Liège, Belgique
|Master : ingénieur civil électricien, à finalité spécialisée en "signal processing and intelligent robotics"
|Master thesis of the Faculté des Sciences appliquées
[en] In the past few years, the interest in drones has grown in many domains. However, it turns out that the usual quadrocopters are not the best in terms of power efficiency and maneuverability. This is why the creation of a new coaxial drone has been launched by Christophe Greffe. One major obstacle to the use of drones is the need to train people to fly drones, which is both expensive and time consuming. Moreover, only the trained people can fly drones, and their presence is then required whenever a drone is used. This can be a problem in situations such as earthquake relief or firefighting. The best solution is to have an autonomous drone able to flight alone safely while achieving a given task. A first step for autonomous navigation is to build a representation of the environment around the drone. In this work, the LSD-SLAM algorithm is used to provide a 3D map of the environment suitable for navigation. It uses a single camera, which makes the design easier and cheaper than with multiple cameras. However, this comes at a price: there is an unknown scale factor between the LSD-SLAM map and the reality. An IMU is therefore integrated in the original algorithm to determine this scale factor. Three different methods for scale estimation are implemented and evaluated, as well as a practical way to obtain the orientation of the LSD-SLAM map with respect to the Earth frame. Currently, these scale estimation methods do not provide satisfying results due to the complexity of the problem and the large sensitivity of the methods to the quality of the provided data. However, one of these methods is very promising and would be worth to investigate further. The accuracy and robustness of this method could be improved and a better camera could also enhance the results in complex environments by providing higher-quality data acquisition. With these improvements, the problem of constructing a map of the environment that is suitable for navigation could be solved, which would bring the project of building an autonomous drone one step closer to its achievement.
Size: 54.29 MB
Format: Adobe PDF
Size: 166.45 kB
Format: Adobe PDF
Description: "A few illustrations that reflect my work"
Size: 3.39 MB
Cite this master thesis
The University of Liège does not guarantee the scientific quality of these students' works or the accuracy of all the information they contain.