// the find
AprilRobotics/apriltag
AprilTag is a visual fiducial system popular for robotics research.
AprilTag is the standard visual fiducial library for robotics — print a tag, point a camera at it, get a 6-DOF pose back. It's the go-to for drone localization, robot arm calibration, and AR marker work. If you're doing anything where a camera needs to know where a physical object is, this is what most people reach for first.
The detector is genuinely fast for what it does — multi-threaded quad detection with configurable decimation lets you tune the speed/range tradeoff at runtime. No external dependencies beyond libc means it drops into embedded and bare-metal builds without a dependency fight. ArUco family support was added natively, so you're not forced to choose between ecosystems. The Python wrapper ships with the library and doesn't require a separate install step.
Linux-only officially, and the Windows situation is 'users have had success' — not a confidence-inspiring phrase if you're building a production pipeline on Windows. The pose estimation has a known ambiguity problem for tags that aren't square-on to the camera; the README points you to OpenCV's SOLVEPNP_IPPE_SQUARE as an alternative but doesn't explain when you should prefer it. The test suite is thin — three test files covering detection counts and pose error, nothing that exercises the multi-threaded paths or stress-tests the decoder under noise. Documentation for the flexible layout system is sparse given how much of the codebase is dedicated to it.