Hello,
I am making a thesis about a system to auto land a quadcopter on a station based on a visual homing system.
I have already made some studies on the problem, and used Youtube footage to test some hypothesis. The problem is that visual homing is too computationaly expensive to be done in the air without some conditioning of the images with telemetry information.
The Youtube videos I have found are below. The problem I found with the Youtube data is that I think measuring the parameters by hand would a bit tedious. I also thought of using satelite data which has scale and orientation data but I am not sure if I could get the accuracy I needed near the ground.
Vertical FPV Highland and Yucaipa CA
Gopro looking straight down on the skywalker fpv
As an example of my approach:
* In a certain position I take a vertical picture of the terrain below.
* Correct the scale of the feature points I am looking for with altitude data.
* Correct the orientation of the image with the orientation data.
* Detect and match the feature points
* Take a course of action based on the matched points, like point correlation
Without this corrections the algorithms necessary would be unwieldy in the air.
So what I am asking is:Does any body know, or have footage of a flight with the camera looking down and the associated timestamped telemetry?
How could I simulate something like this without acquiring all the equipment beforehand. If somebody is available to do this as a job I am open to offers.
Thank you a lot and I am open to suggestions.
Paulo Neves
Replies
I think I can get the recognition going but i would really like a log of a flight with the photos or video to validate the recognition.
Thanks for the reply.
you can take as many photos/video you like, but that's another story, the flowmeter or is still the way to do the job. px4flow could store images too.
Thank you a good lot!
It should not be a problem to land with precision of <+/-5cm. the resolution is better than needed, but the final landing will be affected by ground effect /downwash turbulence, which will make it harder to compensate for movement in the final moment.