pith. sign in

arxiv: 2605.29452 · v1 · pith:R6NOACGMnew · submitted 2026-05-28 · 💻 cs.CV

Comparative evaluation of photogrammetric reconstruction methods and 3D Gaussian Splatting for road surface roughness analysis

Marouane Elmegdar , Teng Xiao This is my paper

Pith reviewed 2026-06-29 08:12 UTC · model grok-4.3

classification 💻 cs.CV
keywords photogrammetryroad surface roughness3D reconstructionGaussian Splattingpoint cloud processingpavement monitoringCOLMAPMeshroom
0
0 comments X

The pith

Open-source photogrammetry pipelines yield usable relative road roughness values from smartphone images.

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

The paper tests four image-based 3D reconstruction methods on road scenes captured by phone cameras and runs every resulting point cloud through one fixed sequence of alignment, segmentation, normal estimation, and roughness calculation. It reports that COLMAP detects the most micro-texture detail, Meshroom gives moderate variation, Metashape smooths surfaces internally, and 3D Gaussian Splatting adds noise while still showing irregularities. The central finding is that the two open-source pipelines remain viable for relative roughness comparison once the downstream steps are held constant. A reader would care because this route removes the need for specialized sensors and therefore lowers the cost of repeated pavement checks.

Core claim

After identical CloudCompare processing, COLMAP point clouds exhibit the highest sensitivity to micro-texture, Meshroom reconstructions show balanced roughness variation, Metashape produces the smoothest geometry because of its internal filtering, and 3D Gaussian Splatting captures visible surface irregularities but at the cost of higher noise and lower density; the comparison therefore shows that open-source pipelines support relative roughness evaluation for low-cost pavement monitoring.

What carries the argument

The single CloudCompare workflow of orientation alignment, segmentation, normal estimation, and roughness computation at fixed neighborhood radii of 0.2, 0.4, and 0.6 model units applied uniformly to all four point clouds.

If this is right

  • COLMAP reconstructions provide the strongest response to fine-scale surface texture.
  • Meshroom outputs allow moderate roughness variation without extreme smoothing or noise.
  • Metashape geometry is the most filtered and therefore least sensitive to micro-texture.
  • 3D Gaussian Splatting preserves visible irregularities but introduces more point-cloud noise.
  • Open-source pipelines become practical substitutes for commercial tools in relative roughness monitoring.

Where Pith is reading between the lines

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

  • Agencies could collect phone images during routine drives and process them on standard hardware for periodic network-level screening.
  • The same uniform workflow could be tested on images of cracked or rutted pavement to see whether roughness metrics predict distress progression.
  • If the relative rankings hold across varied lighting and road types, the approach could support crowd-sourced or vehicle-mounted data collection at city scale.

Load-bearing premise

That the same roughness calculation steps produce comparable numerical values from point clouds that differ in density, noise level, and internal filtering.

What would settle it

A side-by-side comparison of the computed roughness values against independent physical measurements taken with a contact profilometer or laser scanner on the same road sections, checking whether the relative ordering across pipelines matches the physical ordering.

read the original abstract

Image-based 3D reconstruction offers a low-cost alternative to traditional sensor-based techniques for road surface assessment. This study compares four reconstruction pipelines--COLMAP, Meshroom, Metashape, and 3D Gaussian Splatting (3DGS)--to evaluate their ability to estimate road surface roughness from smartphone imagery. All point clouds were processed in CloudCompare using a consistent workflow involving orientation alignment, segmentation, normal estimation, and roughness computation at neighborhood radiuses of 0.2, 0.4, and 0.6 model units. The results show that COLMAP provides the highest sensitivity to micro-texture, while Meshroom yields balanced reconstructions with moderate roughness variation. Metashape produces the smoothest geometry due to its internal filtering, and 3DGS captures visible irregularities but exhibits higher noise and lower density. The comparison demonstrates that open-source pipelines are viable for relative roughness evaluation, offering a practical approach for low-cost pavement monitoring.

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 compares four 3D reconstruction pipelines (COLMAP, Meshroom, Metashape, and 3D Gaussian Splatting) applied to smartphone imagery of road surfaces. All resulting point clouds are processed through an identical CloudCompare workflow (alignment, segmentation, normal estimation, roughness at fixed radii 0.2/0.4/0.6), yielding qualitative rankings: COLMAP highest micro-texture sensitivity, Meshroom balanced, Metashape smoothest due to internal filtering, and 3DGS noisier with lower density. The central claim is that open-source pipelines are viable for relative roughness evaluation in low-cost pavement monitoring.

Significance. If the comparability of roughness metrics holds, the work offers a practical, low-cost alternative to sensor-based methods for infrastructure assessment. The explicit multi-pipeline comparison including 3DGS is timely; however, the absence of dataset scale, ground-truth validation, or quantitative error metrics limits immediate applicability.

major comments (2)
  1. [Abstract] Abstract and Methods (described workflow): the viability conclusion rests on the assumption that fixed-radius roughness (std. dev. of point-to-plane distances at 0.2/0.4/0.6) remains comparable across pipelines despite documented differences in point density, noise, and filtering (Metashape internal smoothing, 3DGS higher noise/lower density, COLMAP micro-texture sensitivity). No density normalization, scale-adaptive radius, or ground-truth roughness reference is described, so observed differences could be reconstruction artifacts rather than surface properties. This assumption is load-bearing for the central claim.
  2. [Abstract] Abstract: no dataset size (number of images, road segments, or total points), ground-truth comparison, quantitative error statistics, or statistical tests are supplied, leaving the qualitative rankings only moderately supported.
minor comments (2)
  1. Clarify the coordinate units of the 'model units' used for the roughness radii.
  2. The abstract would benefit from a brief statement of the number of test scenes or total imagery used.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of our comparative evaluation. We address each major comment below and will revise the manuscript to improve clarity on methodological assumptions and dataset details.

read point-by-point responses

  1. Referee: [Abstract] Abstract and Methods (described workflow): the viability conclusion rests on the assumption that fixed-radius roughness (std. dev. of point-to-plane distances at 0.2/0.4/0.6) remains comparable across pipelines despite documented differences in point density, noise, and filtering (Metashape internal smoothing, 3DGS higher noise/lower density, COLMAP micro-texture sensitivity). No density normalization, scale-adaptive radius, or ground-truth roughness reference is described, so observed differences could be reconstruction artifacts rather than surface properties. This assumption is load-bearing for the central claim.

    Authors: We acknowledge that fixed-radius roughness computation assumes a degree of comparability despite pipeline-specific variations in density and noise. The radii (0.2/0.4/0.6) were chosen to support a uniform CloudCompare workflow for relative ranking rather than absolute measurement. We agree this is a load-bearing assumption and will revise the Methods and Discussion sections to explicitly note the lack of density normalization or ground-truth references, clarify that conclusions are limited to relative viability for low-cost monitoring, and discuss potential reconstruction artifacts as a limitation. revision: yes

  2. Referee: [Abstract] Abstract: no dataset size (number of images, road segments, or total points), ground-truth comparison, quantitative error statistics, or statistical tests are supplied, leaving the qualitative rankings only moderately supported.

    Authors: We will add explicit dataset details (number of images, road segments, and point counts) to the Methods section in the revision. The study intentionally focuses on comparative pipeline performance for relative roughness rather than absolute validation; therefore ground-truth comparisons and quantitative error statistics were outside scope. We will update the Abstract and add a limitations paragraph to state this scope clearly and note the qualitative nature of the rankings. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical comparison of pipelines via fixed workflow

full rationale

The paper conducts a direct empirical comparison of four reconstruction pipelines (COLMAP, Meshroom, Metashape, 3DGS) by applying an identical CloudCompare sequence (alignment, segmentation, normal estimation, roughness at fixed radii 0.2/0.4/0.6) to each output point cloud and reporting observed differences. No equations, fitted parameters, predictions, or derivations appear in the abstract or described workflow. The central viability claim rests on these side-by-side measurements rather than any self-referential definition, self-citation chain, or renaming of known results. Methodological concerns about cross-pipeline comparability (density/noise sensitivity) are validity issues, not circularity; the result is not forced by construction from its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Empirical comparison study; no mathematical model, free parameters, axioms, or new entities are introduced.

pith-pipeline@v0.9.1-grok · 5700 in / 997 out tokens · 39122 ms · 2026-06-29T08:12:27.586032+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

13 extracted references · 10 canonical work pages

  1. [1]

    Industry - and Academic -Based Trends in Pavement Roughness Inspection Technologies over the Past Five Decades: A Critical Review,

    A. Fares and T. Zayed, “Industry - and Academic -Based Trends in Pavement Roughness Inspection Technologies over the Past Five Decades: A Critical Review,” Remote Sens., vol. 15, no. 11, p. 2941, June 2023, doi: 10.3390/rs15112941

    work page doi:10.3390/rs15112941 2023

  2. [2]

    Automatic identification of dense damage -sensitive features in civil infrastructure using sparse sensor networks,

    S. Quqa, L. Landi, and P. P. Diotallevi, “Automatic identification of dense damage -sensitive features in civil infrastructure using sparse sensor networks,” Autom. Constr. , vol. 128, p. 103740, Aug. 2021, doi: 10.1016/j.autcon.2021.103740

    work page doi:10.1016/j.autcon.2021.103740 2021

  3. [3]

    Structure-from-Motion Revisited,

    J. L. Schonberger and J.-M. Frahm, “Structure-from-Motion Revisited,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , Las Vegas, NV, USA: IEEE, June 2016, pp. 4104– 4113. doi: 10.1109/CVPR.2016.445

    work page doi:10.1109/cvpr.2016.445 2016

  4. [4]

    Assessing road roughness using UAV-derived dense point clouds,

    N. Polat and Ş. Akça, “Assessing road roughness using UAV-derived dense point clouds,” Mersin Photogramm. J., vol. 5, no. 2, pp. 75–81, Dec. 2023, doi: 10.53093/mephoj.1358902

    work page doi:10.53093/mephoj.1358902 2023

  5. [5]

    A Review on 3D Reconstruction Techniques from 2D Images,

    M. Aharchi and M. Ait Kbir, “A Review on 3D Reconstruction Techniques from 2D Images,” in Innovations in Smart Cities Applications Edition 3, M. Ben Ahmed, A. A. Boudhir, D. Santos, M. El Aroussi, and İ. R. Karas, Eds., in Lecture Notes in Intelligent Transportation and Infrastructure., Cham: Springer International Publishing, 2020, pp. 510–522. doi: 10.1...

    work page doi:10.1007/978-3-030-37629-1_37 2020

  6. [6]

    Recent advances in 3D Gaussian splatting,

    T. Wu et al., “Recent advances in 3D Gaussian splatting,” Comput. Vis. Media, vol. 10, no. 4, pp. 613–642, Aug. 2024, doi: 10.1007/s41095-024-0436-y

    work page doi:10.1007/s41095-024-0436-y 2024

  7. [7]

    Performance Assessment of Neural Radiance Fields (NeRF) and Photogrammetry for 3D Reconstruction of Man -Made and Natural Features:,

    A. Polimera, M. Mohan, and K. Rajitha, “Performance Assessment of Neural Radiance Fields (NeRF) and Photogrammetry for 3D Reconstruction of Man -Made and Natural Features:,” in Proceedings of the 16th International Conference on Agents and Artificial Intelligence , Rome, Italy: SCITEPRESS - Science and Technology Publications, 2024, pp. 840–847. doi: 10.5...

    work page doi:10.5220/0012396700003636 2024

  8. [8]

    [Online]

    COLMAP. [Online]. Available: https://colmap.github.io

  9. [9]

    [Online]

    AliceVision. [Online]. Available: https://alicevision.org/

  10. [10]

    AliceVision Meshroom: An open-source 3D reconstruction pipeline,

    C. Griwodz et al., “AliceVision Meshroom: An open-source 3D reconstruction pipeline,” in Proceedings of the 12th ACM Multimedia Systems Conference , Istanbul Turkey: ACM, June 2021, pp. 241 –247. doi: 10.1145/3458305.3478443

    work page doi:10.1145/3458305.3478443 2021

  11. [11]

    [Online]

    Agisoft Metashape. [Online]. Available: https://www.agisoft.com

  12. [12]

    3D Gaussian splatting for real-time radiance field rendering,

    B. Kerbl, G. Kopanas, T. Leimkuehler, and G. Drettakis, “3D Gaussian Splatting for Real -Time Radiance Field Rendering,” ACM Trans. Graph., vol. 42, no. 4, pp. 1–14, Aug. 2023, doi: 10.1145/3592433

    work page doi:10.1145/3592433 2023

  13. [13]

    Generating LOD3 building models from structure-from-motion and semantic segmentation,

    B. G. Pantoja -Rosero, R. Achanta, M. Kozinski, P. Fua, F. Perez -Cruz, and K. Beyer, “Generating LOD3 building models from structure-from-motion and semantic segmentation,” Autom. Constr., vol. 141, p. 104430, Sept. 2022, doi: 10.1016/j.autcon.2022.104430

    work page doi:10.1016/j.autcon.2022.104430 2022