Resources
Egocentric camera calibration and synchronization checklist
Audit geometry and time independently, then connect the evidence to camera settings, sessions and delivery files.
Published 2026-10-27 · 6 min read
| Evidence group | Minimum inspection record |
|---|---|
| Camera identity | IDs, lenses, settings, mount and session applicability |
| Geometry | Inputs, units, intrinsics, distortion, transforms and review results |
| Time | Clock domains, frame times, alignment method, measured drift and gaps |
| Delivery | Versioned references, corrected versus original files and unresolved issues |
Calibration describes the geometry of observations; synchronization describes their relationship in time. A project can pass one and fail the other. Request separate evidence and tie both to the exact camera settings and sessions being delivered.
This is an inspection checklist, not a universal certification test. The buyer must set tolerances for the intended task, motion, distances and downstream processing. A statement that a rig is calibrated or synchronized is not the evidence itself.
1. Identify the cameras and settings
Record stable camera IDs, model and lens, resolution, frame rate, focus or zoom settings, exposure-related settings where relevant and the physical mount. Mark settings that changed during capture. A calibration file must identify the configuration it describes.
OpenCV describes intrinsics, distortion and coordinate transforms. Store the target type, target dimensions and units, calibration inputs, software version and resulting parameters. Do not discard supporting images after producing a final calibration file.
2. Check geometry and applicability
Inspect reprojection results and rectification on more than one convenient view. Include relevant distances and image regions in the review. Record the method and actual results, then compare with project criteria rather than attaching an unexplained pass label.
Tie the calibration to sessions and document mount changes. A head-to-wrist transform changes as the person moves unless tracked; a rigid stereo transform can become invalid if the mount shifts. Version corrected parameters and keep the earlier record.
3. Record clock domains and exposure timing
For every stream, state timestamp units, clock domain, frame IDs and whether time represents exposure, device generation or host arrival. Explain hardware triggering, shared clock or post-alignment procedures actually used. Never infer the method from the filename alone.
Inspect the beginning and end of a representative session for offsets and drift, and examine gaps or duplicate frames. Equal nominal frame rates and matching start times do not establish continued alignment. Record measured limitations rather than describing all frames as perfectly synchronized.
4. Validate the actual device configuration
RealSense’s multi-camera guidance includes hardware synchronization and validation for specified devices and configurations. Apply vendor instructions to the exact model, firmware and enabled streams. A depth-stream synchronization claim should not silently be extended to every color stream or unrelated camera.
Check bandwidth, storage, dropped frames and stream restart behavior under the full configuration. A successful two-camera desk test does not prove a longer capture with additional streams will remain intact. Preserve logs and identify sessions affected by resets.
5. Link evidence to the delivery
The manifest should map session and camera IDs to original files, timestamp logs, calibration versions and quality findings. Explain any resampling, rectification or timestamp correction, keeping original values available where agreed. Missing evidence should be marked missing, not populated with assumed values.
The pilot acceptance record should name the reviewer, tested subset, observed results, applicable limits and unresolved issues. Agree whether failures trigger partial acceptance, reprocessing or recapture, and who bears that work. Calibration review is separate from permissions and commercial acceptance.
Frequently asked questions
Does rectification prove exposure synchronization?
No. Rectification addresses image geometry. Timing needs independent evidence.
Can one threshold apply to every buyer?
No universal threshold is asserted here. Agree tolerances for the task and test conditions.
Can file creation times substitute for frame timestamps?
Do not assume so. Identify the actual timestamp source and meaning.