Extrinsic Calibration Reference

This page covers the mechanics behind extrinsic calibration: intrinsic refinement, the depth-ratio gate, skipping intrinsic calibration, and quality metrics. For the step-by-step workflow, see Extrinsic Calibration.

Intrinsic Refinement

The Extrinsics tab has a "Refine camera intrinsics" checkbox, on by default. When checked, bundle adjustment re-estimates each camera's focal length and leading distortion coefficients (k1, k2) jointly with the camera poses, starting from whatever intrinsics you provided. The principal point, tangential distortion (p1, p2), and k3 stay fixed.

Refinement works when the calibration target moves toward and away from the cameras enough that its change in size in the image improves the focal length estimate. Uncheck the box to lock your provided intrinsics when you trust a careful prior calibration more than the extrinsic footage.

Fisheye cameras are excluded from refinement. The equidistant distortion model stays locked during bundle adjustment regardless of the checkbox.

When one or more cameras have no intrinsics at all (see Skipping Intrinsic Calibration), the checkbox is forced on: the solver must recover those parameters.

The Depth-Ratio Gate

A target at 3 meters with a 50mm lens produces the same image as a target at 6 meters with a 100mm lens. If the target stays at roughly the same distance throughout the capture, the solver cannot separate focal length from target distance. Moving the target toward and away from the cameras breaks this ambiguity.

Caliscope measures this before refining intrinsics. For each camera, it computes the ratio between the farthest and nearest observed target distances. If any camera's ratio falls below 2.0, Caliscope disables intrinsic refinement for the entire rig. One camera with flat depth coverage gates all the others.

The gate can silently override the refine-intrinsics checkbox. The log records per-camera depth ratios when this happens.

Skipping Intrinsic Calibration

Experimental

This path is supported by the pipeline and passes synthetic tests, but has not been validated on real-world data to the same degree as the standard intrinsic-then-extrinsic workflow. The recommended path is to calibrate intrinsics first.

If a project has extrinsic videos but no intrinsic videos, Caliscope starts each camera from a rough guess (focal length from resolution, zero distortion) and refines focal length, k1, and k2 during bundle adjustment. The principal point, tangential distortion, and k3 stay at their assumed values.

Prerequisites: depth variation (move the target toward and away from the cameras) and no fisheye cameras. The depth-ratio gate checks depth variation automatically. Fisheye cameras require prior intrinsic calibration.

For the file layout, see Project Setup.

Quality Metrics

After setting the origin, Caliscope computes volumetric scale accuracy:

Pooled RMSE (mm): The root-mean-square error of pairwise distances between reconstructed board corners, compared to the known board geometry. Measures how accurately the reconstruction preserves physical scale.

Bias: Consistently positive errors (reconstructed distances larger than known) suggest the entered board size is too small. Consistently negative errors suggest it is too large. This catches measurement errors in the board dimensions you entered during setup.