Set up the calibration board

Use this page to pick, measure and describe the board. You need a printed board or a monitor. You get the board: block every later step uses.

Running example

The guides use one rig as the running example: the VI-Rig recording set virig-4imus-4cams. It has 4 IMUs, a T265 fisheye pair and an ELP stereo pair. It is on the Calibench Hub (calibench-vi/datasets) under CC BY-NC 4.0.

Hands on: you leave this page with a board: block (and, if you have one, a heights file). Step 2 records with that board. Step 3 puts the block in its config.

1. Pick a board type

AprilGrid 6 × 6

Chessboard 10 × 7 squares

These are the two boards we used to record the calibration motion. Both are in section 5 as images and configs. Pick one. Each needs these fields in the config:

Board	board_type	Required fields
AprilGrid	apriltag	rows, cols, tag_size (m), tag_spacing (ratio), tag_family
Chessboard	chessboard	rows, cols (inner corners), square_size (m)

• AprilGrid follows the Kalibr 2-bit border convention.
• Kalibr target YAML: MVIS can read one in Python (AprilTagBoard.from_kalibr_yaml). No config key points at a Kalibr file, so write the board: block yourself.

2. The board in the running example

The VI-Rig recordings use a printed AprilGrid:

• Grid: 6 × 6 tags, family t36h11
• Tag: 0.088 m, edge to edge
• Spacing: 0.3, the ratio of the gap to the tag

Its Kalibr file, april.yaml, is on the Hub next to the bags. It says target_type: 'aprilgrid', tagCols: 6, tagRows: 6, tagSize: 0.088, tagSpacing: 0.3.

The same board in MVIS:

config.yaml

board:
  board_type: apriltag
  rows: 6
  cols: 6
  tag_size: 0.088
  tag_spacing: 0.3
  tag_family: t36h11

3. Measure what you printed

Measure the printed board

Print at 100% and measure a tag with calipers. If the true size differs from tag_size, every translation in the rig scales by that same factor. This error is silent and systematic: no residual will reveal it.

4. A board is never flat

A printed board bends and twists a little. A solve that assumes it is flat pushes that error into the calibration.

The VI-Rig board, estimated from the camera residuals:

• Out of plane: 10–11 mm peak to peak, 2.3–2.4 mm rms.
• Example run: the 4 IMUs + 4 cameras example re-estimates it at 2.21 mm rms and 8.94 mm peak to peak.
• Effect: estimating the shape cut the camera-only residual by 30–45 %. On the ELP left camera alone, 0.771 px became 0.430 px.

Turn it on with one key. MVIS then solves one height per corner, in camera-only and camera–IMU runs:

config.yaml

board:
  # ... the fields above ...
  heights_file: board_shape_4imus_4cams.yaml   # optional start values, from an earlier run
estimate:
  board_points: true

• estimate.board_points: true: solve the board shape with everything else.
• board.heights_file: optional. The heights start from this file, or flat without it. A camera–IMU run writes its estimate next to its output as <output>_board_shape.yaml, which you can use here. The example’s board_shape_4imus_4cams.yaml is such an earlier estimate of the VI-Rig board.
• Prior: each height has a weak prior of 3 mm (solver default board_height_sigma: 0.003, in m).

We turn it on for every VI-Rig solve.

5. No printer? Use a screen board

You can show the board on a monitor instead. We made two full-screen images for a 2560 × 1440 monitor, with configs for MVIS and Kalibr:

Board	Size	Downloads
AprilGrid 6 × 6	tag36h11, tag 43.47 mm, spacing 0.3	image (PNG, 75 KB) · MVIS config · Kalibr config
Chessboard 10 × 7 squares	9 × 6 inner corners, square 43.47 mm	image (PNG, 30 KB) · MVIS config · Kalibr config

How to show them:

1. Open the image full screen at 100 % on a 2560 × 1440 monitor.
2. Use no display scaling and no zoom, so one image pixel is one screen pixel.

On our monitor 160 px (one tag, one square) measure 43.47 mm, the size in the configs. On another screen, measure one tag or square with a ruler. Put that size in the config:

• MVIS: tag_size / square_size
• Kalibr: tagSize / rowSpacingMeters and colSpacingMeters

Setup:

• Board upright and flat.
• Enough light for a short exposure (about 2 ms).

Next

Step 2: record the dataset. It shows how to move the rig in front of the board.