Hey all! For any #frc folks, are there any good resources for vision? I'd like to use the apriltags to align and score, but I don't know much and was hoping to learn more.
To anyone, if you know of resources, papers, or anything else relating to vision, and how that ends up making stuff happen in the code, I'd appreciate anything you could share.

#programming #robot #robotics #code #apriltags #help #firstrobotics

@aberrant68 the limelights have built in April tag support, so the coding is essentially reduced to "where is April tag 1", "it's left X pixels and up Y pixels on my video"

@aberrant68 I wrote a blog post on the AprilTags a little while ago (https://blog.fixermark.com/posts/2022/april-tags-python-recognizer/).

For recognizing them in competition, the team I’m working with is using Raspberry Pis with PhotonVision (https://photonvision.org/). This solution is pretty plug-and-play (modulo the challenge of getting a Raspberry Pi and networking switch mounted and powered on the robot chassis)—once you get it working, it’ll auto-recognize the tags and dump all the relevant data (either via NetworkTables or directly via a library for the robot code).

Playing with AprilTags: Python Recognizer

An announcement by FIRST Robotics earlier this year revealed an interesting bit of news: next year’s kit would include AprilTags as vision targets. I’m completely unfamiliar with these, so I figured it’d be fun to dive in and see what they’re all about. Generating some tags AprilTags are a so-called “fiducial” system that identifies not only a position and orientation, but an identifier code as well via a block encoding scheme, conceptually similar to a barcode but with some design features useful to making them easy to comprehend from multiple angles.

@aberrant68 We have been hoping to use Apriltags for localization via Limelight. Initial tests show a lot of noise: even under pretty easy lighting conditions apriltag-only localization jumps around a lot. Investigating `SwerveDrivePoseEstimator` to filter things. Competition field lighting (and other bots with lights and occlusion!) is _SO_ much harder than our test conditions, I'm not super-hopeful about apriltags being better than good ol' retro' tape, esp. inside 8' or so.
@aberrant68 could try photon vision. Haven't had the chance to test localisation performance yet but targeting is reasonably good
@sebasptsch That might be useful, thank you! I saw a mention of Photon Vision on CD but I didn't realize it could track apriltags as well.
@aberrant68 their documentation is really helpful as well. Thankfully a lot of different hardware options are supported including limelight or even a low power NUC. Though our team went for a raspberry pi.