Skip to main content

Extrinsic calibration

Recover the 6-DoF pose of every camera in a single shared coordinate frame.

Work in progress

Scaffold page — to be expanded with anchor selection and bundle-adjustment review.

The flow​

Like intrinsics, extrinsics are computed on demand, in four phases:

  1. Capture — record a synchronized sweep while presenting the board so that pairs of cameras see it at the same instant.
  2. Prepare — replay the synchronized frame groups (every camera's frame of the same instant, side by side) and tune the compute: a sampling stride; the minimum shared sightings a camera pair must have — how many synchronized instants two cameras must both see the board before their relative pose is estimated directly (below it the geometry is too weak, so that pair is instead linked through other cameras in the co-visibility graph); and the maximum sync spread (ms) allowed within a group.
  3. Computing — pairwise poses, transitive chaining, then bundle adjustment.
  4. Result — inspect the reconstructed rig in the 3D review. The world frame starts on the anchor camera (index 0); you can rebase it onto a board ("set frame on board") — for example, a board laid on the floor becomes the world origin and ground plane — snap-rotate the axes by ±90°, and optionally re-run the bundle adjustment ("minimize").

Under the hood​

  • Pairwise relative poses via cv2.stereoCalibrate on each pair's shared board views.
  • A co-visibility graph (with bridge-filling) links the cameras; poses are chained transitively from the anchor (camera index 0).
  • A final bundle adjustment (scipy.optimize.least_squares, trf, sparse Jacobian) jointly refines every non-anchor pose and the 3D points. The anchor stays fixed (identity), which removes the gauge freedom and conditions the solve.

→ Explanation & sources: Methodology

→ See also: Calibration best practices — shared views and multi-camera tips.