Vision-based UAV autonomous landing in natural scenario
Kalliola, Jussi (2023)
2023
Master's Programme in Computing Sciences
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-05-16
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202305024980
https://urn.fi/URN:NBN:fi:tuni-202305024980
Tiivistelmä
In this thesis, the topics of safe landing area determination and vision-based localization in the context of UAV autonomous landing in natural scenario are studied. Topics were investigated by answering the following research questions: ”Is a customer-grade mobile phone equipped with LiDAR sensor suitable for geometry-based safe landing area determination?”, ”How robust to extreme scale and viewpoint differences are the detector-based visual localization methods for localizing a UAV in relation to the landing site?”, and ”Can the mentioned technologies and methods in the previous research questions be used to perform an autonomous landing in a natural scenario?”. All of the questions were answered by experimenting the developed systems using static datasets collected from real-life natural scenarios.
The results of the experiments showed that proposed safe landing area determination algorithm implemented on iPhone 13 pro mobile phone is able to propose the optimal and alternative suitable landing sites that satisfy geometric requirements. In addition, the mobile device was able to run the data collection, preprocessing, and landing area determination methods in near real-time.
The experimentation on vision-based localization of a UAV in relation to the landing site revealed that detector-based methods can localize the landing site with a good accuracy from high altitudes in difficult natural scenarios. Good accuracy and localization rate indicates high robustness towards extreme scale and viewpoint changes. Additionally, the classical SIFT+NN-ratio methods outperformed modern
SuperPoint+SuperGlue methods in almost all metrics.
The combined analysis of the results indicates that the safe landing area determination system with vision-based localization of a UAV satisfies all the requirements, and therefore, can be used to perform an autonomous landing in natural scenario.
The results of the experiments showed that proposed safe landing area determination algorithm implemented on iPhone 13 pro mobile phone is able to propose the optimal and alternative suitable landing sites that satisfy geometric requirements. In addition, the mobile device was able to run the data collection, preprocessing, and landing area determination methods in near real-time.
The experimentation on vision-based localization of a UAV in relation to the landing site revealed that detector-based methods can localize the landing site with a good accuracy from high altitudes in difficult natural scenarios. Good accuracy and localization rate indicates high robustness towards extreme scale and viewpoint changes. Additionally, the classical SIFT+NN-ratio methods outperformed modern
SuperPoint+SuperGlue methods in almost all metrics.
The combined analysis of the results indicates that the safe landing area determination system with vision-based localization of a UAV satisfies all the requirements, and therefore, can be used to perform an autonomous landing in natural scenario.