Camera Calibration and Image Quality Diagnostics in Agisoft Metashape

Calibration should be evaluated in relation to the decision the project must support.

• Identify whether the project requires mapping, measurement, inspection, modelling, heritage documentation or visualisation.
• Define absolute accuracy, relative geometry, surface detail or visual-quality requirements.
• Record the final products whose quality may be affected by camera modelling.
• Identify independent checkpoints, scale bars, known dimensions or other validation evidence.
• Define acceptance criteria before comparing calibration variants.
• Record whether the camera was pre-calibrated, self-calibrated or used with fixed parameters.
• Identify the reviewer responsible for accepting the camera solution.

Diagnostic record: intended use, acceptance criteria, calibration strategy and validation evidence.

Determine whether images assumed to share a camera model were captured under comparable conditions.

• Record camera model, lens, image dimensions, focal length, pixel size and file format.
• Check whether images were resized, cropped, rotated, converted or corrected before import.
• Identify zoom-lens use, autofocus changes, stabilisation, digital zoom or in-camera correction.
• Separate different cameras even when they share the same commercial model name.
• Record capture sessions, flights, temperature changes, lens removal or physical impacts.
• Review missing, inconsistent or modified EXIF information.
• Document rolling-shutter or global-shutter sensor type where known.
• Preserve original files and metadata for comparison.

Diagnostic record: sensor inventory and a list of events that may justify separate calibration behaviour.

Image defects can weaken matching and bias interpretation of the calibration model.

• Inspect focus consistency across the frame and across the full dataset.
• Identify motion blur, vibration, long exposure and directional smearing.
• Review clipped highlights, blocked shadows, low contrast and strong exposure variation.
• Identify high compression, repeated resaving or artefacts from preprocessing.
• Compare suspect images at full resolution rather than relying only on thumbnails.
• Distinguish isolated defective images from a systematic acquisition problem.
• Record exclusion criteria and retain a list of disabled or removed images.
• Avoid choosing an arbitrary quality threshold without reviewing project geometry and coverage.

Diagnostic record: image-quality audit with examples, affected groups and documented exclusion decisions.

Self-calibration requires a camera network capable of separating geometric effects.

• Review forward, side and cross-strip overlap.
• Assess diversity of camera positions, viewing directions, roll angles and distance to subject.
• Identify predominantly parallel imagery that may provide weak parameter separation.
• Review whether oblique imagery or convergent views strengthen the network where appropriate.
• Check spatial distribution of tie points across the image frame.
• Identify low-texture, reflective, repetitive, moving or masked areas.
• Review elevation range and control distribution for aerial projects.
• Document areas where geometry is insufficient for a strong conclusion.

Diagnostic record: camera-network assessment describing geometric strengths and unresolved weaknesses.

One physical camera does not always imply one stable calibration state.

• Confirm the groups created automatically from camera model, resolution and focal information.
• Create separate groups where capture session, route, focus, zoom, resolution or camera state changed.
• Avoid excessive fragmentation when groups do not have sufficient observations to support independent calibration.
• Do not combine different physical cameras merely because their EXIF model names match.
• Record which parameters are shared, adjusted, fixed or pre-calibrated.
• Review multi-camera systems and band relationships separately from ordinary single-camera grouping.
• Preserve a table connecting every image set to its calibration group.
• Document the reason for each manual group decision.

Diagnostic record: calibration-group register with membership, assumptions and parameter status.

A baseline makes later optimization changes interpretable.

• Record alignment settings, preselection, key-point and tie-point limits.
• Preserve the initial aligned state before cleaning, reference optimization or parameter changes.
• Review unaligned cameras and weakly connected image groups.
• Inspect tie-point coverage and spatial distribution.
• Record initial camera and marker residual information.
• Review calibration values and distortion plots for each group.
• Identify systematic patterns by flight, session, camera or image region.
• Do not evaluate only the average reprojection error.

Diagnostic record: saved baseline state with alignment settings, statistics and initial observations.

Focal length, principal point and distortion coefficients should not be judged independently of one another.

• Compare initial and adjusted values for focal length and principal-point coordinates.
• Review radial and tangential distortion behaviour using the available plots.
• Inspect parameter uncertainty and correlation where available.
• Look for implausible differences between nominally comparable groups.
• Identify parameter drift associated with weak geometry or unsuitable grouping.
• Check whether fixed parameters are supported by reliable pre-calibration evidence.
• Avoid copying calibration values between cameras without confirming compatibility.
• Record the interpretation, not only the numerical parameter table.

Diagnostic record: parameter comparison with plots, uncertainties, correlations and technical interpretation.

Change only parameters justified by the camera model and network evidence.

• Duplicate or save the project before each material optimization variant.
• Confirm reference accuracies and control/checkpoint roles before optimization.
• Record the parameters enabled at every stage.
• Compare camera, marker, checkpoint and reprojection residuals after each change.
• Inspect systematic residual direction and spatial pattern.
• Review parameter stability rather than selecting only the lowest aggregate error.
• Avoid enabling unsupported higher-order terms merely to improve fit.
• Document tie-point filtering or observation removal performed before optimization.
• Stop when added complexity is not supported by improved independent evidence.

Diagnostic record: versioned optimization comparison with settings, results and acceptance rationale.

A camera solution should be evaluated through its effect on independent project evidence.

• Confirm checkpoints remained excluded from the final adjustment.
• Compare checkpoint errors before and after calibration variants.
• Review horizontal, vertical and spatial bias where applicable.
• Use independent scale bars or known dimensions for close-range work.
• Inspect deformation patterns in products, profiles or sections.
• Compare repeated captures or external survey data where available.
• Evaluate whether improvement is consistent across the site rather than local to control points.
• State when independent evidence is insufficient to validate the solution.

Diagnostic record: independent comparison linked to the accepted calibration variant.

The conclusion should state what was demonstrated and under which conditions.

• Identify every calibration group and its image membership.
• Record initial and adjusted parameter status.
• Include relevant plots, residual statistics and validation results.
• Document excluded images, projections, points and optimization variants.
• Record software version, project state and processing report.
• State whether the solution applies only to the current dataset or can support a defined future workflow.
• Disclose weak geometry, limited validation, rolling-shutter risk or unstable parameters.
• Identify who reviewed and approved the conclusion.

Diagnostic record: calibration assessment report with evidence, scope, approval and limitations.