Greetings! I’m currently working on an attachment for model aircraft and drones. I’m hoping that tensorflow will come in handy for this project. The goal is to create a fully functional targeting system that aids RC pilots in fpv dogfighting. Is it possible to create an fpv hud that displays targeting data to the operator, using tensorflow? The current script targets bright objects. Any suggestions? I’m a newbie at this. The picture below is an example of what I’d like to accomplish. Thank you.
Hi, we’re about to release bounding box detection support for TensorFlow. This will do what you want. The FPS isn’t going to be real-time though.
Anyway, other techniques would be something called cam-shift. We don’t have this algorithm onboard… but, essentially, the camera can track an object by finding the odd thing out in a flat background… I.e sky. And then track the plan in the sky. This can be done at the max frame rate since it’s very simple.
You’d have to edit our firmware and write C code to add this algorithm. But, it’s quite trivial to implement as it’s from the 80s.
Actually, just use color tracking for a color not equal to the sky.
This is very easy. Set the color thresholds to the sky and then pass the invert parameter for find _blobs() so it tracked everything that’s not the sky.
I will definitely give this a try. thank you. I may also try attaching bright LEDs of different colors to each plane to exaggerate the effect of the find_blobs(). I’m hoping that will solve the possible issue of terrain interference.