I am Umer, Final year student of software engineering and deep learning intern at TUKL R&D lab NUST. I am happy to see a project (Machine vision for copter landing in absence of GPS) that aligns perfectly with my final year project which is (autonomous Safe UAV landing using computer vision and deep learning) and we made some decent progress in this project.
we scraped runways from all around the world and created a database of those runways using google satellite api, after that we trained an object detection model to localize runway from aerial view in real time(YOLO) and we are getting 20FPS on TITAN X and mAP of 85%. after that orientation is calculated using hough line transform.
For that matter i believe using my experience in this domain i can make great improvements in this project.
That’s very interesting, looking forward to your proposal!
Our multicopter integration with Companion Computers works fairly well. Copter can accept various commands from a companion computer including 3D velocity commands, target position commands, changing of modes, etc. Our integration with plane is not quite as advanced (yet) because it only accepts target position commands.
I mentioned it in another thread but here are some things that might help with your proposal:
- use open source free tools wherever possible
- the final product should be useful and hopefully a good starting place for follow up work from others
- include documenting the solution as part of the proposal
- add a diagram if possible to explain things
best of luck