Menelaos - Premiere H2020
Structure from Motion (SfM) evolved in Computer Vision and Robotics where has supported the development of visual Simultaneous Localization and Mapping systems (SLAM). It has constantly progressed, a great leap forward being the use of Deep Learning. The Deep Neural Networks have the capacity to learn complex models that are more far reaching than man crafted features.
Supervised learning, while effective, requires large amounts of images with corresponding ground truth depth for each application. The solution to reduce the effort of collecting depth maps has been the use of unsupervised learning. This is the most recent trend in SLAM. Known as self-supervising, this form of learning consists in using indirect information for training.
In this project, we recast SfM and self-learning for drone flight. The foreseen applications are obstacle detection and collision avoidance, flying and landing in absence of GPS signal and/or failure of onboard sensors, photogrammetry.
OBJECTIVES
- Enlarge the research area of Computational Imaging at CEOSpaceTech.
- Development of competences on Computational Imaging by training a new generation of researchers in view of a sustainable development of CEOSpaceTech.
- Initiating research activities at master level by proposing dissertation subjects to the students pursuing a M.Sc. degree in Electronics, Telecommunications and Information Technology, a pool for recruiting future PhDs at CEOSpaceTech.
ACKNOWLEDGEMENT
This work was supported by the grant of the Romanian Ministry of Education and Research, CNCS - UEFISCDI, project number 31/01.01.2021, program International and European Cooperation, Support – Awarding of the participation in Horizon 2020, PNCDI III.