I am using photon vision for pose estimation, although the Y coordinates is switching between 4.45 and 5.45 rapidly as well as the bounding boxes on the April tags being super distorted. I would love some insight on this. The X coordinate is completely stable, I think I have narrowed it down to the angles being detected being wrong but I have no clue what is wrong.
2 Likes
I’ve had this issue today too! I tried recalibrating my camera and still didn’t fix it.
1 Like
I’m using version 2024.3.1 not the 2025 beta version.
Can you go to the camera calibration settings, and check the error table there? There should be a table with Resolution, Mean Error, and Standard Deviation (and the FOV numbers).
Our numbers on the robot were 1280 X 720, 0.37px, 0.13px, showing that our pixel errors were within a pixel. I don’t know what is considered bad numbers, but I have seen it be as bad as 57px with a camera with a fish eye. If that number is too bad, you probably need to change to a different camera.
You can actually see with this sample image, the fish eye caused the bounding box to miss the tag, and you can see how not straight the black to green section is.
Note, I wrote this thinking it was an odometry error, but saying the bounding box was distorted, this note is 2-3 steps down the road…
Is your camera mounted straight vertical, or at an angle?
If it’s mounted at an angle, try inverting the angle. I know we were off by an inconsistent 7-12" looking at tags until we blindly put a minus sign and it got us to ~1" error.
1 Like
What’s your resolution? We used to run 480p, which had a lot of jittering. It became much better when we switched to 720p.
My resolution is 1280x720, error is 1px. It doesn’t jump just a foot or so it jumps literally from source side to amp side.
It’s mounted at an angle. The crazy thing is that it didn’t do any of this last week.
If it helps, it only gets screwed up when I enable multi tag estimation in output, I don’t think it is my camera as it is the one recommended on the photon docs.
Blind guess, are your apriltags in the correct positions? Both flipping which one is left/right, or having the wrong gap between them could cause radically different results, especially if you said it’s only happening on multi-tag scenarios.
Edit, if you have the ~1px error, I think your camera is fine. We use a camera not on their recommended camera list (the one with those results), and it worked fine all season.
1 Like
Why would the April tags being in the wrong place cause jittering, wouldn’t it yield and incorrect estimate, rather than switching between 4.45 and 5.45?
Gap probably shouldn’t make it jump, but swapping them might. The solving algorithm gets confused on what side of the tags you are seeing and jumps around. It could also be a rotated tag too, that would freak out the solving algorithm.
I will try that thank you.