Programming · Vision
Vision that works on the field
Camera setup, OpenCV basics, AprilTags overview, and how to survive common vision failures.
What this is
- Quick path to get a camera feed and simple pipelines running.
- Overview of AprilTags usage for pose/ID.
- Checks to keep vision stable under match lighting.
When you need it
- First time adding a webcam or pipeline to FTC code.
- Detection drops under bright or dim venues.
- AprilTags pose is noisy or IDs flicker.
How it works
- Camera: use UVC webcams; set resolution modest (e.g., 640x480) to keep FPS.
- OpenCV: create a pipeline class; process frames, output telemetry; keep operations lightweight.
- AprilTags: calibrate camera if using pose; set correct tag size and FX/FY/CX/CY; stabilize by averaging a few frames.
- Lighting: prefer auto exposure off if unstable; clamp brightness with simple thresholds.
Common mistakes
- Running high resolution → low FPS and timeouts.
- No camera calibration for pose → wrong distances/angles.
- Blocking pipelines (heavy loops) that starve OpMode.
- Hardcoded tag size/focal lengths from sample that don’t match your setup.
How to test it
- Print FPS and detection latency in telemetry.
- Move tag or target slowly; ensure IDs stay stable and pose doesn’t jump wildly.
- Test under venue-like lighting; adjust exposure/white balance if needed.
- Check CPU load; keep control loop responsive while vision runs.
Related problems
- Pipeline crashes → add null checks; guard empty detections; keep processing lightweight.
- Pose noise → average a few frames; ensure correct camera intrinsics and tag size.
- FPS drops mid-match → reduce resolution or complexity; avoid heavy allocations per frame.