OpenCV Calibration Audit

Executive summary

Quality gates: PASSED

3 accepted calibration views from 3 files. OpenCV RMS: 0.20340 px.

Pass/fail quality gates

• minimum_valid_images — 3 accepted views meet the minimum.

Dataset counts and metrics

{
  "discovered_count": 3,
  "readable_count": 3,
  "detected_count": 3,
  "accepted_count": 3,
  "warning_count": 2,
  "rejected_count": 0,
  "unreadable_count": 0,
  "detection_rate": 1.0,
  "coverage_grid": [
    [
      0,
      0,
      0,
      0
    ],
    [
      0,
      0,
      3,
      0
    ],
    [
      0,
      0,
      0,
      0
    ]
  ],
  "occupied_coverage_cells": 1,
  "coverage_ratio": 0.08333333333333333,
  "corner_density_grid": [
    [
      0,
      19,
      26,
      7
    ],
    [
      0,
      29,
      40,
      17
    ],
    [
      0,
      4,
      16,
      4
    ]
  ],
  "scale_min": 0.26277651193271157,
  "scale_max": 0.46862020197127985,
  "scale_median": 0.41266563998664263,
  "scale_iqr": 0.10292184501928414,
  "occupied_scale_bins": 3,
  "rotation_range_degrees": 89.3570877095579,
  "horizontal_perspective_range": 0.1370729353299674,
  "vertical_perspective_range": 0.2598575602435533,
  "duplicate_count": 0
}

Coverage heatmap

Sharpness distribution

Board-region variance of Laplacian for calibration views.

Scale and pose diversity

• Board area: min 0.26277651193271157, median 0.41266563998664263, max 0.46862020197127985
• Occupied scale bins: 3
• Rotation range: 89.3570877095579°
• Perspective ranges: horizontal 0.1370729353299674, vertical 0.2598575602435533

These assessments are heuristic, not a physical certificate.

Calibration parameters

{
  "camera_model": "pinhole",
  "image_size": [
    640,
    480
  ],
  "opencv_rms": 0.20340291970181212,
  "camera_matrix": [
    [
      535.4891370284612,
      0.0,
      341.6228649364716
    ],
    [
      0.0,
      536.7293891622663,
      234.43436418806436
    ],
    [
      0.0,
      0.0,
      1.0
    ]
  ],
  "distortion_coefficients": [
    -0.3167297374324814,
    0.21519540968566292,
    0.0016617683409911475,
    0.0009056149866629888,
    -0.16542528687003036
  ],
  "rotation_vectors": [
    [
      -0.4343827824153476,
      0.26711541650527393,
      -3.0868237724543386
    ],
    [
      0.6878954813521805,
      -0.4395432201112522,
      1.6552546571540114
    ],
    [
      -0.2644229356490152,
      -0.39376875227334673,
      -2.747693589710688
    ]
  ],
  "translation_vectors": [
    [
      0.14275680805780136,
      0.02681254482728464,
      0.4392453956190113
    ],
    [
      0.09962415135176081,
      -0.05094937457297831,
      0.25429897380302097
    ],
    [
      0.1186726034405404,
      0.09085357999699209,
      0.29068391619419687
    ]
  ],
  "calibration_flags": 0,
  "mean_per_view_rmse_px": 0.20260696602909348
}

Reprojection-error chart

Accepted, warning, and rejected images

Annotated previews show detected inner corners and the evaluated physical board boundary.

Preview	Image and decision	Metrics	Decision details
	left01.jpg ACCEPTED Accepted for calibration Detector: findChessboardCornersSB	Physical board area: 0.26277651193271157 Board sharpness: 579.8427672879492 In-sample RMSE: 0.1974014474935786	No warnings or rejection reasons.
	left02.jpg WARNING Accepted with warning Detector: findChessboardCornersSB	Physical board area: 0.41266563998664263 Board sharpness: 340.73278981353707 In-sample RMSE: 0.22675551335368663	BOARD_CLIPPED_OR_NEAR_BORDER WARNING Estimated physical board boundary is clipped or close to an image border. Measured: 0.007870239516099295; Threshold/rule: 0.01
	left03.jpg WARNING Accepted with warning Detector: findChessboardCornersSB	Physical board area: 0.46862020197127985 Board sharpness: 587.9059410669647 In-sample RMSE: 0.18366393724001526	BOARD_CLIPPED_OR_NEAR_BORDER WARNING Estimated physical board boundary is clipped or close to an image border. Measured: -0.038849925994873045; Threshold/rule: 0.01

Configuration and tool version

Version 0.2.1; generated 2026-07-31T12:00:00+03:00

{
  "pattern": {
    "cols": 9,
    "rows": 6,
    "square_size": 30.0,
    "unit": "mm"
  },
  "recursive": false,
  "min_valid_images": 3,
  "min_board_area": 0.01,
  "max_board_area": 0.9,
  "min_sharpness": null,
  "max_per_view_error": null,
  "disable_fallback_detector": false,
  "fail_on_warning": false,
  "overwrite_output": false,
  "log_level": "INFO",
  "coverage_cols": 4,
  "coverage_rows": 3,
  "max_file_size_mb": 100,
  "near_border_ratio": 0.01,
  "relative_sharpness_factor": 0.35,
  "duplicate_center_distance": 0.03,
  "duplicate_log_area_distance": 0.08,
  "duplicate_rotation_degrees": 5.0,
  "duplicate_perspective_distance": 0.08,
  "min_coverage_ratio": 0.35,
  "min_scale_bins": 3,
  "min_rotation_range_degrees": 15.0
}

Metric limitations

Sharpness depends on resolution, optics, and target scale. Coverage and diversity thresholds are heuristics. Per-view reprojection error is an in-sample residual: it is calculated after fitting the camera model on the same accepted dataset and is not independent validation. Calibration results apply only to the accepted dataset and pinhole model.