Authors:
Alan Kunz Cechinel
1
;
Juha Röning
2
;
Antti Tikanmaki
2
;
Edson DePieri
1
and
Patricia Della Méa Plentz
3
Affiliations:
1
Automation and Systems Department, Federal University of Santa Catarina, Florianópolis, Brazil
;
2
Biomimetics and Intelligent Systems Group, Faculty of Information Technology and Electrical Engineering, University of Oulu, Oulu, Finland
;
3
Informatics and Statistics Department, Federal University of Santa Catarina, Florianópolis, Brazil
Keyword(s):
Autonomous Landing, Unnamed Aerial Vehicle (UAV), Computer Vision, Wildfire Monitoring.
Abstract:
Wildfire has been an environmental, economic, and health problem worldwide. Technological advances have led to the popularization of Unmanned Aerial Vehicles (UAVs) for personal and business use. One of the Unmanned Aerial Vehicle (UAV) applications is monitoring. However, UAVs still have payload and battery limitations. UAVs can be an ally for wildfire management, but their use is challenging considering their restraints and the large size of monitored areas. Therefore, it is necessary to develop approaches to circumvent UAV limitations. This work’s approach allows a drone to land in strategic locations for data acquisition, resulting in significantly less battery consumption. The method uses principles from stereo vision through a monocular camera motion to estimate the relative position of a selected landing site, allowing a drone to hang itself by a hook in an artificial (e.g., aluminum frame, power line) or natural (e.g., tree branch) location. However, the system is limited to
static landing sites where the FAST feature detector algorithm can detect features. The results showed that the landing site estimation system achieves over 90% accuracy in controlled scenarios. Moreover, the Landing Site Estimation System (LSES) allied with navigation controllers achieved 95% success in landing attempts with light and wind under control.
(More)