Skip to main content

Tune For Detection

This guide is for teams doing first-time tuning for ArUco.

Goal​

You are done when detections are stable under your robot lighting and motion.

Tune Order (Do This, Not Random Sliders)​

  1. Fix the image (focus, blur, lighting).
  2. Set camera controls.
  3. Verify you are viewing the correct output.
  4. Tune pipeline parameters.

Step 0: Confirm Your Stream Is Set Up For Detection​

If you are using the built-in OV9782 and your stream is a mono/detection stream:

  1. Prefer NV12 for the stream format.
  2. Select the luma decoder nv12-luma.

Reason:

  • NV12 with the luma path avoids extra decode work for mono workloads.

If you are unsure where to set this:

Step 1: Camera Controls (Get A Clean, Sharp Image)​

  1. Open the stream.
  2. Go to Controls.
  3. Enable noise reduction and set to Fast (if available).
  4. Reduce motion blur:
    • lower exposure time
    • increase lighting

What you should see:

  • Less speckle noise in low light.
  • Less motion blur when you pan or drive.

If the image is too dark after lowering exposure:

  • Increase lighting first (this is the best fix for ArUco).
  • Then increase gain only as needed.

Step 2: Output Selection (Make Sure You Are Looking At The Right Thing)​

In the stream Pipelines tab:

  • select output frame (annotated output) for preview.

If you do not see frame:

  • Pick another image-like output and watch for overlays.
  • Do not assume the raw camera preview contains overlays.

Guide:

Step 3: Pipeline Tune (Only After Steps 0-2)​

  1. Open OS > Pipelines.
  2. Open your pipeline.
  3. Open Tune.
  4. Select the running stream.

ArUco:

  • set dictionary to match printed tags (commonly 36h11).

What you should verify while you tune:

  • Detections appear when a tag is in view.
  • Detections do not flicker wildly when the tag is steady.
  • Detections still work while you move the robot (lighting and motion).

Step 4: Use Outputs To Confirm (Do Not Trust The Preview Alone)​

Use the Outputs tab to read non-image outputs and confirm detections are actually being produced.

Guide:

If Detections Are Still Bad​

  1. Confirm the dictionary matches your printed tags (36h11 is common, but not automatic).
  2. Improve lighting and reduce motion blur before touching pipeline knobs.
  3. Verify you are not accidentally tuning a different stream than the one you are viewing.
  4. Check Alerts and Systems logs for errors.

See: Tuning