Absolute positioning of drones using aerial images
Organisatsiooni nimi
Autonomous Driving Lab
Kokkuvõte
In war zones the GPS is not usable and drones need to rely on other means for navigation. Typical solution is to use a combination of inertial measurement unit (IMU) and visual odometry to relatively position the drone with respect to the starting point. Such relative positioning systems tend to drift over time, the positioning error can be hundreds of meters after flying a few kilometers. To fly longer distances such drones need additional absolute positioning - the drone camera image needs to be matched with an aerial map of the location.
Your tasks will be:
* Validate the use of simple pre-trained convolutional neural networks for creating a heatmap of the potential locations of the drone camera image with respect to the aerial image. Estonian Land Board orthophotos can be used as the aerial images.
* Train a custom neural network for creating such a heatmap. Contrastive learning will be used to learn representation where two images of the same place from different sources have similar representation and two images of a different place have different representation.
Links:
* https://cs231n.stanford.edu/reports/2017/pdfs/817.pdf
* https://github.com/XingruiWang/wheres-waldo
* https://www.researchgate.net/publication/350061509_PatchNet_--_Short-range_Template_Matching_for_Efficient_Video_Processing
* https://amslaurea.unibo.it/26836/1/Convolutional%20Neural%20Network%20Architectures%20for%20Template%20Matching.pdf
Your tasks will be:
* Validate the use of simple pre-trained convolutional neural networks for creating a heatmap of the potential locations of the drone camera image with respect to the aerial image. Estonian Land Board orthophotos can be used as the aerial images.
* Train a custom neural network for creating such a heatmap. Contrastive learning will be used to learn representation where two images of the same place from different sources have similar representation and two images of a different place have different representation.
Links:
* https://cs231n.stanford.edu/reports/2017/pdfs/817.pdf
* https://github.com/XingruiWang/wheres-waldo
* https://www.researchgate.net/publication/350061509_PatchNet_--_Short-range_Template_Matching_for_Efficient_Video_Processing
* https://amslaurea.unibo.it/26836/1/Convolutional%20Neural%20Network%20Architectures%20for%20Template%20Matching.pdf
Lõputöö kaitsmise aasta
2024-2025
Juhendaja
Tambet Matiisen, Edgar Sepp, Karl-Johan Pilve
Suhtlemiskeel(ed)
eesti keel, inglise keel
Nõuded kandideerijale
Tase
Magister
Märksõnad
Kandideerimise kontakt
Nimi
Tambet Matiisen
Tel
E-mail
Vaata lähemalt