RGB-D calibration aligns measured depth with scene geometry and, where an RGB camera is present, with the color image. A ToF camera may also need product-specific offset, temperature, and cover-window checks. There is no universal set of coefficients or calibration order that fits every camera, so start with the manufacturer’s method and record the exact hardware, firmware, and operating mode.
1. Capture a reference data set
Fix the camera in its final or representative enclosure. Collect repeated images of a measured planar target or another traceable reference at several distances across the intended field of view. Repeat at relevant target reflectivities and ambient-light levels. If the device will operate across a broad temperature range, include stabilized temperature points. Save raw depth, confidence or amplitude, RGB frames where available, and the reference measurements.
2. Check intrinsics and depth scale
Estimate or verify focal length, principal point, and lens distortion using the camera vendor’s calibration procedure or a suitable calibration target. Compare reported depth with the physical reference to assess scale, offset, nonlinearity, and spatial variation. Keep invalid pixels separate from a measured zero or maximum range. Correct only errors supported by enough observations; an attractive fit to one plane can fail at other distances.
3. Align RGB and depth
If RGB and depth come from separate optical paths, estimate their rigid transform and verify registration at multiple depths. Report the coordinate convention and whether depth is registered into RGB coordinates or vice versa. A ToF-only camera has no RGB-to-depth extrinsic calibration to perform.
4. Validate the finished assembly
Re-test after adding the cover window and final mounting. Internal reflections, contamination, thermal drift, multipath, and poor returns from dark, transparent, or shiny surfaces can change results. Report mean or median error, spread, invalid-return rate, and the distances and scene conditions tested. Recalibrate after a hardware, optics, firmware, or enclosure change if the vendor’s procedure requires it.
For a practical choice of ToF camera or single-point sensor, see how a ToF camera works and compare the final device against the task’s error budget.
