pith. sign in

arxiv: 2605.15860 · v1 · pith:OG4JVUMAnew · submitted 2026-05-15 · 💻 cs.CV

On RGB-TIR Stereo Calibration under Extreme Resolution Asymmetry

Micha{\l} Kr\'ol , Micha{\l} Salamonowicz , W{\l}adys{\l}aw Skarbek , Micha{\l} Tomaszewski This is my paper

Pith reviewed 2026-05-20 19:37 UTC · model grok-4.3

classification 💻 cs.CV
keywords RGB-TIR calibrationstereo calibrationlow-resolution thermalbundle adjustmentcorner detectionactive calibration targetbuilding energy analysis
0
0 comments X

The pith

Baseline-constrained bundle adjustment recovers accurate geometry for RGB paired with 625-times lower resolution TIR camera.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a calibration approach for stereo systems that pair a high-resolution RGB camera with a thermal infrared camera limited to just 80 by 62 pixels. An active OLED screen projects checkerboard patterns for the thermal sensor and ChArUco patterns for the RGB sensor on the same physical surface to generate repeatable contrast. A custom corner finder that rectifies perspective, locates saddle points, and refines positions with mean shift works reliably on the tiny thermal frames without manual adjustments per image. A bundle adjustment step that fixes the physical baseline length then corrects for the geometric instability introduced by a flat calibration target. The resulting alignment supports mapping thermal measurements onto RGB images for applications such as evaluating heat loss in building structures.

Core claim

The framework achieves a recovered stereo baseline of 32.7 mm against a nominal 30 mm value together with an overall reprojection error of 0.382 pixels by combining modality-specific patterns on an active OLED screen, a dedicated low-resolution corner detector, and a baseline-constrained bundle adjustment that enforces physically consistent rig geometry despite degeneracy from the planar calibration object.

What carries the argument

Baseline-constrained bundle adjustment that enforces physically consistent rig geometry under the planar-calibration-object degeneracy.

If this is right

  • The calibrated system produces consistent TIR-to-RGB projections that support both constant-depth and per-pixel depth estimation.
  • Validation on a building mock-up demonstrates suitability for multimodal energy performance assessment.
  • The overall reprojection error stays at 0.382 pixels across the extreme resolution pair.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The active single-surface pattern approach could reduce setup time for calibration in field conditions where separate targets are impractical.
  • The same constrained adjustment principle may stabilize calibration for other sensor pairs that suffer from large resolution differences.

Load-bearing premise

The dedicated corner detection algorithm reliably locates checkerboard corners in 80 by 62 pixel thermal images through perspective rectification, Hessian analysis, and mean-shift refinement without requiring per-frame parameter tuning.

What would settle it

A validation run on the thermally active building mock-up that produces either a measured baseline differing by more than a few millimeters from 30 mm or a reprojection error substantially above 0.382 pixels would show the constrained adjustment fails to maintain consistent geometry.

Figures

Figures reproduced from arXiv: 2605.15860 by Micha{\l} Kr\'ol, Micha{\l} Salamonowicz, Micha{\l} Tomaszewski, W{\l}adys{\l}aw Skarbek.

Figure 1
Figure 1. Figure 1: Overview of the proposed RGB–TIR stereo calibration pipeline. Each stage corresponds to a dedicated section of the paper, as indicated on the right. 2. Related Work Research directly relevant to this work spans three themes: (1) geometric calibration of RGB– thermal infrared (TIR) camera systems, with particular attention to configurations exhibiting resolution [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: RGB–TIR measurement setup and active OLED screen: (a) complete RGB–TIR acquisition rig with preview display; (b) front view of the rigid camera stereo configuration; (c) schematic of the measurement setup; (d) thermally active building mock-up, 1:10 scale; (e) checkerboard pattern displayed on the OLED screen for the TIR modality; (f) ChArUco pattern displayed on the OLED screen for the RGB modality. 3.2. … view at source ↗
Figure 3
Figure 3. Figure 3: Preliminary passive and semi-active calibration objects evaluated during initial experiments: (a– f) examples of silicone and plywood ChArUco boards and further passive calibration object variants. Based on these findings, an OLED display was adopted as the active OLED screen. Light emission from the display induces localised temperature variations across the pattern elements, producing thermal contrast be… view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the proposed thermal corner detection pipeline. Stages 1–4 perform image preprocessing and geometric normalisation; Stages 5–6 detect candidate corners using differential analysis; Stage 7 evaluates geometric consistency and may reject the entire frame; Stage 8 maps accepted corners to the original image coordinates via perspective projection. 4.3.1. Percentile Normalisation Raw thermal images … view at source ↗
Figure 5
Figure 5. Figure 5: ROI estimation and perspective rectification: (a) normalised thermal image with percentile scaling; (b) binary ROI mask obtained via Otsu thresholding and morphological refinement; (c) quadrilateral enclosing the checkerboard derived from the ROI mask; (d) image after rectification and histogram equalisation (CLAHE) [29] with the expected grid overlay. 4.3.3. Hessian Saddle-Point Response Checkerboard corn… view at source ↗
Figure 6
Figure 6. Figure 6: Saddle-point detection and corner localisation: (a) Hessian saddle response map S with the expected grid overlay; (b) Mean Shift convergence from nominal grid positions (blue crosses) to detected corner locations (green markers), with search windows indicated; (c) detected corners overlaid in the rectified thermal image with grid indices. 4.3.5. Quality Gate After corner localisation, a frame-level quality… view at source ↗
Figure 7
Figure 7. Figure 7: Quality gate diagnostics for a representative frame: (a) triangle areas of adjacent grid cells colour-coded by deviation from the expected value; (b) adjacent missing node test — detected corners shown as green markers, with two synthetically removed adjacent nodes (red circles) connected by a red segment to illustrate a gate-failing configuration (|M| = 2). 4.3.6. Perspective Projection to Original Coordi… view at source ↗
Figure 8
Figure 8. Figure 8: Corner positions after perspective projection to the original thermal image coordinate system: (a) corners in the rectified space; (b) corresponding positions after inverse homography, overlaid on the normalised thermal image with grid labels. Representative detection results obtained by the proposed algorithm in thermal images from a calibration session are presented in [PITH_FULL_IMAGE:figures/full_fig_… view at source ↗
Figure 9
Figure 9. Figure 9: Examples of checkerboard corner detection results (a–h) obtained by the proposed thermal corner detection algorithm on images of 80 × 62 px resolution. Detected corners are connected by grid lines; frames rejected by the quality gate are indicated by a red border. Detection statistics and gate diagnostics are annotated for each frame. 4.4. Frame Pairing Across Modalities During calibration sessions, the nu… view at source ↗
Figure 10
Figure 10. Figure 10: Examples of calibration point sets detected in the RGB and TIR modalities for selected frames [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Illustration of RGB point subsampling for harmonisation with the thermal point grid. The titles above each subplot indicate the frame ID and the bounding box of the corner region in the format ID: (xmin, ymin) − (xmax, ymax), where (xmin, ymin) represents the top-left corner and (xmax, ymax) represents the bottom-right corner. The black circle indicates the centre point of the image; only the calibration … view at source ↗
Figure 12
Figure 12. Figure 12: Comparison of RGB–TIR fusion using two approaches based on the estimated stereo calibration parameters (R, T): (a–c) variant using per-pixel plane-intersection depth derived from the ChArUco board pose: (a) detection and pose estimation, (b) TIR overlay on RGB, (c) TIR isolines; (d–f) variant using per-pixel depth estimation with DepthPro: (d) depth map illustrating the character of an outdoor scene with … view at source ↗
read the original abstract

Accurate geometric calibration of RGB-thermal infrared (TIR) stereo camera systems is essential for multimodal building envelope analysis, yet remains challenging when low-cost thermal sensors with very low spatial resolution are employed. This paper presents a practical stereo calibration framework for an RGB camera (2028 x 1520 px) paired with a TIR camera operating at only 80 x 62 px - a pixel-count ratio of approximately 1:625. An active OLED screen dynamically switches modality-specific patterns (checkerboard for TIR, ChArUco for RGB) on a single physical surface, providing controlled and repeatable thermal contrast. A dedicated corner detection algorithm combining perspective rectification, Hessian saddle-point analysis, and Mean Shift localisation achieves reliable checkerboard detection at 80 x 62 px without per-frame parameter tuning. A baseline-constrained bundle adjustment enforces physically consistent rig geometry under the planar-calibration-object degeneracy, yielding a stereo baseline of 32.7 mm (nominal 30 mm) with an overall reprojection error of 0.382 px. The system is validated on a thermally active building mock-up using constant-depth and per-pixel depth estimation, demonstrating consistent TIR-to-RGB projection suitable for building energy performance assessment.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper presents a stereo calibration framework for RGB-TIR camera pairs under extreme resolution asymmetry (RGB 2028×1520 px paired with TIR 80×62 px). It uses an active OLED screen to display modality-specific patterns (checkerboard for TIR, ChArUco for RGB), a dedicated corner detector combining perspective rectification, Hessian saddle-point analysis and Mean Shift localisation for reliable TIR corner finding without per-frame tuning, and a baseline-constrained bundle adjustment to enforce physically consistent rig geometry despite planar-calibration-object degeneracy. The method reports a recovered stereo baseline of 32.7 mm (nominal 30 mm) and overall reprojection error of 0.382 px, with validation via constant-depth and per-pixel depth estimation on a thermally active building mock-up.

Significance. If the central results hold, the work offers a practical solution for multimodal calibration in applications such as building energy performance assessment, where low-cost low-resolution TIR sensors are paired with high-resolution RGB cameras. The active display approach for controlled thermal contrast and the explicit baseline constraint in bundle adjustment directly address the degeneracy that arises with planar targets at such extreme pixel-count ratios (~1:625). The reported baseline accuracy and sub-pixel reprojection error indicate potential for consistent TIR-to-RGB projection, which is a load-bearing requirement for downstream depth and thermal analysis tasks.

major comments (2)
  1. [Abstract / corner detection section] Abstract and methods description of corner detection: the claim that the dedicated detector (perspective rectification + Hessian saddle-point + Mean Shift) achieves reliable checkerboard detection at 80×62 px without per-frame parameter tuning is load-bearing for the entire pipeline, yet the manuscript supplies only the aggregate reprojection error of 0.382 px. No independent localisation error metrics, per-modality residual statistics, detection success rates, or comparison against manual annotations are reported; at this resolution even a 0.3–0.5 px systematic bias corresponds to several degrees of angular error that would directly affect the bundle-adjustment solution yielding the 32.7 mm baseline.
  2. [Validation / experimental results] Validation section: the demonstration on the thermally active building mock-up reports consistent TIR-to-RGB projection via constant-depth and per-pixel depth estimation, but lacks quantitative controls such as comparison to independent ground-truth depth measurements, per-pixel error maps, or ablation of the baseline constraint's contribution to the final accuracy.
minor comments (2)
  1. [Abstract] The stated pixel-count ratio of approximately 1:625 is slightly rounded; an exact calculation (2028×1520)/(80×62) yields ~621.5, which should be reported precisely for reproducibility.
  2. [Results] Notation for the recovered baseline (32.7 mm) and nominal value (30 mm) would benefit from explicit uncertainty intervals or covariance from the bundle adjustment to allow readers to assess how close the result truly is to the physical rig.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their thorough review and constructive comments on our manuscript. We address each of the major comments point by point below, indicating the changes we will make in the revised version.

read point-by-point responses

  1. Referee: [Abstract / corner detection section] Abstract and methods description of corner detection: the claim that the dedicated detector (perspective rectification + Hessian saddle-point + Mean Shift) achieves reliable checkerboard detection at 80×62 px without per-frame parameter tuning is load-bearing for the entire pipeline, yet the manuscript supplies only the aggregate reprojection error of 0.382 px. No independent localisation error metrics, per-modality residual statistics, detection success rates, or comparison against manual annotations are reported; at this resolution even a 0.3–0.5 px systematic bias corresponds to several degrees of angular error that would directly affect the bundle-adjustment solution yielding the 32.7 mm baseline.

    Authors: We agree that additional metrics specific to the corner detection would provide stronger evidence for the reliability of the detector at such low resolution. The reported 0.382 px reprojection error is an end-to-end metric after bundle adjustment, and the close match of the estimated baseline (32.7 mm) to the nominal value (30 mm) serves as supporting evidence for the overall calibration quality. Nevertheless, to address the referee's concern directly, we will revise the manuscript to include the detection success rate over the sequence of frames, average per-modality residual statistics from the bundle adjustment, and a limited comparison of detected corners against manual annotations for a representative subset of images. These additions will allow for an independent assessment of the localisation accuracy. revision: yes

  2. Referee: [Validation / experimental results] Validation section: the demonstration on the thermally active building mock-up reports consistent TIR-to-RGB projection via constant-depth and per-pixel depth estimation, but lacks quantitative controls such as comparison to independent ground-truth depth measurements, per-pixel error maps, or ablation of the baseline constraint's contribution to the final accuracy.

    Authors: We appreciate the suggestion for more rigorous quantitative validation. In the revised manuscript, we will add per-pixel error maps for the depth estimation results and an ablation study demonstrating the effect of including the baseline constraint in the bundle adjustment on both the reprojection error and the recovered baseline length. However, our experimental validation on the thermally active building mock-up was designed around the consistency of projections and depth estimates derived from the calibrated system itself; we did not collect independent ground-truth depth data using additional sensors. We will explicitly discuss this aspect as a limitation of the current validation approach. revision: partial

standing simulated objections not resolved
  • Provision of independent ground-truth depth measurements for the building mock-up, since no such external depth reference data was acquired in the experiments.

Circularity Check

0 steps flagged

No circularity: derivation relies on standard models plus independent physical constraint

full rationale

The paper applies established camera models, a custom but explicitly algorithmic corner detector (perspective rectification + Hessian + Mean Shift), and bundle adjustment augmented by a physical baseline constraint drawn from rig geometry. The reported 32.7 mm baseline is an optimization output that differs from the nominal 30 mm input, showing the result is not forced by construction. No equations or claims reduce the final calibration parameters to fitted inputs or self-citations; the method is presented as a practical pipeline whose correctness is checked against external physical expectations rather than internal redefinitions. This is a self-contained engineering contribution against standard benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on established pinhole camera models and planar calibration assumptions, with the novel elements being the detection algorithm and physical constraint rather than new entities or fitted constants.

axioms (2)
  • standard math Standard pinhole camera model with radial and tangential distortion applies to both RGB and TIR sensors.
    Implicit foundation for all stereo calibration computations.
  • domain assumption Planar calibration target introduces degeneracy that can be resolved by enforcing known baseline distance.
    Directly invoked to justify the baseline-constrained bundle adjustment.

pith-pipeline@v0.9.0 · 5766 in / 1303 out tokens · 38561 ms · 2026-05-20T19:37:08.126062+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages

  1. [1]

    Global Buildings and Construction Report 2023

    International Energy Agency. Global Buildings and Construction Report 2023. Technical report, International Energy Agency, Paris, France, 2023. Accessed: 4 November 2025

    work page 2023

  2. [2]

    Semantic Building Energy Modeling: Analysis across Geospatial Scales.Building and Environment2025,276, 112883

    Wolk, S.; Reinhart, C. Semantic Building Energy Modeling: Analysis across Geospatial Scales.Building and Environment2025,276, 112883. https://doi.org/10.1016/j.buildenv.2025.112883

    work page doi:10.1016/j.buildenv.2025.112883 2025

  3. [3]

    Point Cloud Generation of a Building from Close Range Thermal Images.ISPRS Archives 2019,XLII-5/W2, 29–33

    Dlesk, A.; Vach, K. Point Cloud Generation of a Building from Close Range Thermal Images.ISPRS Archives 2019,XLII-5/W2, 29–33. https://doi.org/10.5194/isprs-archives-XLII-5-W2-29-2019

    work page doi:10.5194/isprs-archives-xlii-5-w2-29-2019 2019

  4. [4]

    A Thermal Performance Detection Method for Building Envelope Based on 3D Model Generated by UAV Thermal Imagery.Energies2020,13, 6677

    Zheng, H.; Zhong, X.; Yan, J.; Zhao, L.; Wang, X. A Thermal Performance Detection Method for Building Envelope Based on 3D Model Generated by UAV Thermal Imagery.Energies2020,13, 6677. https: //doi.org/10.3390/en13246677

    work page doi:10.3390/en13246677

  5. [5]

    Hoegner, L.; Abmayr, T.; Tosic, D.; Turzer, S.; Stilla, U. Fusion of 3D Point Clouds with TIR Images for Indoor Scene Reconstruction.The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences2018,XLII-1, 189–194. https://doi.org/10.5194/isprs-archives-XLII-1-189-2018

    work page doi:10.5194/isprs-archives-xlii-1-189-2018 2018

  6. [6]

    ThermoNeRF: A Multimodal Neural Radiance Field for Joint RGB–Thermal Novel View Synthesis of Building Facades.Advanced Engineering Informatics2025,65, 103345

    Hassan, M.; Forest, F.; Fink, O.; Mielle, M. ThermoNeRF: A Multimodal Neural Radiance Field for Joint RGB–Thermal Novel View Synthesis of Building Facades.Advanced Engineering Informatics2025,65, 103345. https://doi.org/10.1016/j.aei.2025.103345

    work page doi:10.1016/j.aei.2025.103345 2025

  7. [7]

    Iwaszczuk, D.; Stilla, U. Camera pose refinement by matching uncertain 3D building models with thermal infrared image sequences for high quality texture extraction.Photogrammetry and Remote Sensing2017, 132, 33–47. https://doi.org/10.1016/j.isprsjprs.2017.08.006

    work page doi:10.1016/j.isprsjprs.2017.08.006 2017

  8. [8]

    RGB-D and Thermal Sensor Fusion: A Systematic Literature Review.IEEE Access2023,11, 93347–93379

    Brenner, M.; Reyes, N.H.; Susnjak, T.; Barczak, A.L.C. RGB-D and Thermal Sensor Fusion: A Systematic Literature Review.IEEE Access2023,11, 93347–93379. https://doi.org/10.1109/ACCESS.2023.3301119

    work page doi:10.1109/access.2023.3301119 2023

  9. [9]

    Infrared Camera Geometric Calibration: A Review and a Precise Thermal Radiation Checkerboard Target.Sensors2023,23, 3479

    ElSheikh, A.; Abu-Nabah, B.A.; Hamdan, M.O.; Tian, G.Y. Infrared Camera Geometric Calibration: A Review and a Precise Thermal Radiation Checkerboard Target.Sensors2023,23, 3479. https://doi.org/10.3390/s230 73479

    work page doi:10.3390/s230

  10. [10]

    Robust Low Resolution Thermal Stereo Camera Calibration

    Zoetgnandé, Y.W.K.; Fougères, A.J.; Cormier, G.; Dillenseger, J.L. Robust Low Resolution Thermal Stereo Camera Calibration. In Proceedings of the Eleventh International Conference on Machine Vision (ICMV 2018), Munich, Germany, 2019; Vol. 11041,Proceedings of SPIE, p. 110411D. https://doi.org/10.1117/12.2523440

    work page doi:10.1117/12.2523440 2018

  11. [11]

    ROCHADE: Robust Checkerboard Advanced Detection for Camera Calibration

    Placht, S.; Fürsattel, P .; Assoumou Mengue, E.; Hofmann, H.; Schaller, C.; Balda, M.; Angelopoulou, E. ROCHADE: Robust Checkerboard Advanced Detection for Camera Calibration. In Proceedings of the Computer Vision – ECCV 2014, Cham, 2014; Vol. 8692,Lecture Notes in Computer Science, pp. 766–779. https://doi.org/10.1007/978-3-319-10593-2_50

    work page doi:10.1007/978-3-319-10593-2_50 2014

  12. [12]

    Modeling and Calibration of Active Thermal-Infrared Visual System for Industrial HMI.Electronics2022,11, 1230

    Chen, M.; Tian, S.; He, F.; Fu, Q.; Gu, Q.; Wu, B. Modeling and Calibration of Active Thermal-Infrared Visual System for Industrial HMI.Electronics2022,11, 1230. https://doi.org/10.3390/electronics11081230. 26 of 27

    work page doi:10.3390/electronics11081230

  13. [13]

    ThermalGS: Dynamic 3D Thermal Reconstruc- tion with Gaussian Splatting.Remote Sensing2025,17, 335

    Liu, Y.; Chen, X.; Yan, S.; Cui, Z.; Xiao, H.; Liu, Y.; Zhang, M. ThermalGS: Dynamic 3D Thermal Reconstruc- tion with Gaussian Splatting.Remote Sensing2025,17, 335. https://doi.org/10.3390/rs17020335

    work page doi:10.3390/rs17020335

  14. [14]

    A Flexible New Technique for Camera Calibration.IEEE Transactions on Pattern Analysis and Machine Intelligence2000,22, 1330–1334

    Zhang, Z. A Flexible New Technique for Camera Calibration.IEEE Transactions on Pattern Analysis and Machine Intelligence2000,22, 1330–1334. https://doi.org/10.1109/34.888718

    work page doi:10.1109/34.888718

  15. [15]

    OpenCV Documentation: Camera Calibration and 3D Reconstruction (calib3d Module), 2024

    OpenCV Developers. OpenCV Documentation: Camera Calibration and 3D Reconstruction (calib3d Module), 2024. Accessed: 01 December 2025

    work page 2024

  16. [16]

    Mapping Infrared Data on Terrestrial Laser Scanning 3D Models of Buildings.Remote Sensing2011,3, 1847–1870

    Alba, M.I.; Barazzetti, L.; Scaioni, M.; Rosina, E.; Previtali, M. Mapping Infrared Data on Terrestrial Laser Scanning 3D Models of Buildings.Remote Sensing2011,3, 1847–1870. https://doi.org/10.3390/rs3091847

    work page doi:10.3390/rs3091847

  17. [17]

    Improving Calibration of Thermal Stereo Cameras Using Heated Calibration Board

    Saponaro, P .; Sorensen, S.; Rhein, S.; Kambhamettu, C. Improving Calibration of Thermal Stereo Cameras Using Heated Calibration Board. In Proceedings of the Proceedings of the 2015 IEEE International Conference on Image Processing (ICIP), Quebec City, QC, Canada, 2015; pp. 4718–4722. https://doi.org/10.1109/ICIP . 2015.7351702

    work page doi:10.1109/icip 2015

  18. [18]

    A Geometric Calibration Method for Thermal Cameras Using a ChArUco Board.Infrared Physics & Technology2024,138, 105219

    Roshan, M.C.; Isaksson, M.; Pranata, A. A Geometric Calibration Method for Thermal Cameras Using a ChArUco Board.Infrared Physics & Technology2024,138, 105219. https://doi.org/10.1016/j.infrared.2024.1 05219

    work page doi:10.1016/j.infrared.2024.1 2024

  19. [19]

    A Mask-Based Approach for the Geometric Calibration of Thermal-Infrared Cameras.IEEE Transactions on Instrumentation and Measurement 2012,61, 1625–1635

    Vidas, S.; Lakemond, R.; Denman, S.; Fookes, C.; Sridharan, S.; Wark, T. A Mask-Based Approach for the Geometric Calibration of Thermal-Infrared Cameras.IEEE Transactions on Instrumentation and Measurement 2012,61, 1625–1635. https://doi.org/10.1109/TIM.2012.2182851

    work page doi:10.1109/tim.2012.2182851 2012

  20. [20]

    Marker-based Extrinsic Calibration for Thermal–RGB Camera Pair with Different Calibration Board Materials

    Sher, B.A.; Xu, X.; Chen, G.; Feng, C. Marker-based Extrinsic Calibration for Thermal–RGB Camera Pair with Different Calibration Board Materials. In Proceedings of the Proceedings of the 40th International Symposium on Automation and Robotics in Construction (ISARC), Chennai, India, 2023; pp. 490–497. https://doi.org/10.22260/isarc2023/0066

    work page doi:10.22260/isarc2023/0066 2023

  21. [21]

    A Passive Stereo Calibration Technique for Visible– Thermal, Low-Resolution Imaging in Remote Sensing Applications.Measurement2024,231, 114647

    Piccinelli, N.; De Rossi, G.; Daffara, C.; Muradore, R. A Passive Stereo Calibration Technique for Visible– Thermal, Low-Resolution Imaging in Remote Sensing Applications.Measurement2024,231, 114647. https: //doi.org/10.1016/j.measurement.2024.114647

    work page doi:10.1016/j.measurement.2024.114647 2024

  22. [22]

    3D Thermal Mapping of Building Interiors Using an RGB-D and Thermal Camera.Proceedings of the IEEE International Conference on Robotics and Automation (ICRA)2013, pp

    Vidas, S.; Moghadam, P .; Bosse, M. 3D Thermal Mapping of Building Interiors Using an RGB-D and Thermal Camera.Proceedings of the IEEE International Conference on Robotics and Automation (ICRA)2013, pp. 2311–2318. https://doi.org/10.1109/ICRA.2013.6630890

    work page doi:10.1109/icra.2013.6630890 2013

  23. [23]

    Multi-modal Image Matching to Colorize a SLAM Based Point Cloud with Arbitrary Data from a Thermal Camera.ISPRS Open Journal of Photogrammetry and Remote Sensing2023, 9, 100041

    Elias, M.; Weitkamp, A.; Eltner, A. Multi-modal Image Matching to Colorize a SLAM Based Point Cloud with Arbitrary Data from a Thermal Camera.ISPRS Open Journal of Photogrammetry and Remote Sensing2023, 9, 100041. https://doi.org/10.1016/j.ophoto.2023.100041

    work page doi:10.1016/j.ophoto.2023.100041 2023

  24. [24]

    Camera Pose Revisited.Applied Sciences2026,16, 2690

    Skarbek, W.; Salamonowicz, M.; Król, M. Camera Pose Revisited.Applied Sciences2026,16, 2690. https: //doi.org/10.3390/app16062690

    work page doi:10.3390/app16062690

  25. [25]

    ThermalGaussian: Thermal 3D Gaus- sian Splatting

    Lu, R.; Chen, H.; Zhu, Z.; Qin, Y.; Lu, M.; zhang, L.; Yan, C.; anke xue. ThermalGaussian: Thermal 3D Gaus- sian Splatting. In Proceedings of the The Thirteenth International Conference on Learning Representations, 2025

    work page 2025

  26. [26]

    Rotationally Invariant Image Operators

    Beaudet, P .R. Rotationally Invariant Image Operators. In Proceedings of the Proceedings of the International Joint Conference on Pattern Recognition, Kyoto, Japan, 1978; pp. 579–583

    work page 1978

  27. [27]

    Automatic Generation and Detection of Highly Reliable Fiducial Markers Under Occlusion.Pattern Recognition2014,47, 2280–2292

    Garrido-Jurado, S.; Muñoz-Salinas, R.; Madrid-Cuevas, F.J.; Marín-Jiménez, M.J. Automatic Generation and Detection of Highly Reliable Fiducial Markers Under Occlusion.Pattern Recognition2014,47, 2280–2292. https://doi.org/10.1016/j.patcog.2014.01.005

    work page doi:10.1016/j.patcog.2014.01.005 2014

  28. [28]

    Camera Calibration with ChArUco Boards, 2024

    OpenCV Developers. Camera Calibration with ChArUco Boards, 2024. Accessed: 01 December 2025

    work page 2024

  29. [29]

    Contrast Limited Adaptive Histogram Equalization

    Zuiderveld, K. Contrast Limited Adaptive Histogram Equalization. InGraphics Gems IV; Heckbert, P .S., Ed.; Academic Press: San Diego, CA, 1994; pp. 474–485

    work page 1994

  30. [30]

    A Threshold Selection Method from Gray-Level Histograms.IEEE Transactions on Systems, Man, and Cybernetics1979,9, 62–66

    Otsu, N. A Threshold Selection Method from Gray-Level Histograms.IEEE Transactions on Systems, Man, and Cybernetics1979,9, 62–66. https://doi.org/10.1109/TSMC.1979.4310076

    work page doi:10.1109/tsmc.1979.4310076 1979

  31. [31]

    Mean Shift: A Robust Approach Toward Feature Space Analysis.IEEE Transactions on Pattern Analysis and Machine Intelligence2002,24, 603–619

    Comaniciu, D.; Meer, P . Mean Shift: A Robust Approach Toward Feature Space Analysis.IEEE Transactions on Pattern Analysis and Machine Intelligence2002,24, 603–619. https://doi.org/10.1109/34.1000236

    work page doi:10.1109/34.1000236

  32. [32]

    Faugeras, O.; Luong, Q.T.; Papadopoulo, T.The Geometry of Multiple Images: The Laws That Govern the Formation of Multiple Images of a Scene and Some of Their Applications; MIT Press: Cambridge, MA, 2001

    work page 2001

  33. [33]

    Hartley, R.; Zisserman, A.Multiple View Geometry in Computer Vision, 2 ed.; Cambridge University Press: Cambridge, UK, 2003

    work page 2003

  34. [34]

    27 of 27

    Ma, Y.; Soatto, S.; Kosecka, J.; Sastry, S.S.An Invitation to 3-D Vision: From Images to Geometric Models; Springer: Berlin/Heidelberg, Germany, 2004. 27 of 27

    work page 2004

  35. [35]

    Unified Temporal and Spatial Calibration for Multi-Sensor Systems

    Furgale, P .; Rehder, J.; Siegwart, R. Unified Temporal and Spatial Calibration for Multi-Sensor Systems. In Proceedings of the Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Tokyo, Japan, 2013; pp. 1280–1286. https://doi.org/10.1109/IROS.2013.6696514

    work page doi:10.1109/iros.2013.6696514 2013

  36. [36]

    Bundle Adjustment — A Modern Synthesis

    Triggs, B.; McLauchlan, P .F.; Hartley, R.I.; Fitzgibbon, A.W. Bundle Adjustment — A Modern Synthesis. In Vision Algorithms: Theory and Practice; Triggs, B.; Zisserman, A.; Szeliski, R., Eds.; Springer: Berlin, Heidelberg, 2000; Vol. 1883,Lecture Notes in Computer Science, pp. 298–372. https://doi.org/10.1007/3-540-44480-7_21

    work page doi:10.1007/3-540-44480-7_21 2000

  37. [37]

    Depth Pro: Sharp Monocular Metric Depth in Less Than a Second

    Bochkovskii, A.; Delaunoy, A.; Germain, H.; Santos, M.; Zhou, Y.; Richter, S.R.; Koltun, V . Depth Pro: Sharp Monocular Metric Depth in Less Than a Second. In Proceedings of the Proceedings of the 13th International Conference on Learning Representations (ICLR), 2025. Available online: https://openreview.net/forum? id=aueXfY0Clv

    work page 2025