ACCURACY ASSESSMENT OF UNDERWATER PHOTOGRAMMETRIC THREE DIMENSIONAL MODELLING FOR CORAL REEFS

Tiande Guo, Alessandro Capra, Matthias Troyer, A. Gruen, Andrew J. Brooks, James L. Hench, Russell J. Schmitt, Sally J. Holbrook, Marco Dubbini2016 · DOI

TL;DR

This paper tackles the challenge of accurately 3D modeling coral reefs underwater, a crucial task for monitoring their health and growth. By comparing two low-cost camera systems (GoPro and a professional system camera) and employing an in-situ calibration method, they demonstrate that millimeter-level accuracy is achievable, paving the way for more accessible and detailed ecological monitoring.

What Problem Does This Paper Address?

Monitoring the health and growth of coral reefs is vital for understanding the impacts of climate change. Traditionally, this has been done using 2D assessments of coral cover, which fail to capture changes in architectural complexity. While 3D modeling offers a more comprehensive approach, achieving accurate 3D reconstructions underwater is significantly more challenging than in air. Factors like water turbidity, light absorption, and the optical properties of camera housings (especially flat ports) introduce complexities that affect image formation and subsequent 3D reconstruction. This paper addresses the need for a reliable and accurate method to quantify 3D changes in coral reefs over time, specifically aiming for millimeter-level accuracy to track coral growth, which can be as little as 11-14 mm per year.

Methods

The core of this study lies in comparing the accuracy of 3D point clouds generated from two different underwater camera systems: a popular, low-cost GoPro camera in a flat waterproof housing and a more professional Lumix GH4 camera in a housing with a hemispherical dome port. A key methodological innovation is the use of an in-situ calibration approach. Instead of calibrating cameras in a controlled environment like a pool, they placed a precisely measured 3D calibration frame directly into the target underwater scene. This frame, equipped with numbered targets, was scanned using a laser scanner to establish ground truth. This in-situ method aims to minimize environmental influences and improve the accuracy of the 3D models.

To further assess accuracy and investigate potential error sources, they also used cinder blocks with similar surface roughness to corals. These blocks, along with the calibration frame, were imaged in both air and underwater conditions. The automated photogrammetric software Agisoft PhotoScan was used for image orientation, camera calibration, and 3D point cloud generation. The accuracy of the generated point clouds was then evaluated by comparing them against the known measurements of the calibration frame and the reference 3D models of the cinder blocks created in air.

Datasets

The study did not utilize pre-existing datasets for training. Instead, they generated their own datasets by capturing images of:

  • A precisely measured 3D calibration frame (PVC bars with targets) in both air and underwater.
  • Several cinder blocks of various shapes, also in air and underwater.

These captured image sets, along with the ground truth measurements of the calibration frame, formed the basis for their accuracy assessments.

Key Insights and Ideas

  • In-situ calibration is crucial for underwater photogrammetry: The paper strongly advocates for performing calibration directly in the working environment to account for real-world optical conditions and minimize errors.
  • Dome ports offer better optical performance than flat ports: The study highlights that the hemispherical dome port of the Lumix camera resulted in smaller correction parameters during self-calibration compared to the flat port of the GoPro. This suggests that dome ports are superior for minimizing refraction-induced distortions.
  • Low-cost systems can achieve high accuracy: Despite the challenges of underwater environments, the research demonstrates that even a consumer-grade GoPro camera, when properly calibrated and used with appropriate software, can achieve millimeter-level accuracy, making 3D reef monitoring more accessible.
  • Understanding distortion is key: The paper visually and quantitatively shows the radial distortion differences between the cameras, particularly how it increases towards the image borders underwater, impacting 3D reconstruction quality.

Key Results

  • GoPro camera accuracy: Achieved a relative accuracy of 1.3 mm in air and 2.0 mm in water at a working distance of approximately 1.5 m.
  • System camera accuracy: Achieved an accuracy of 1.8 mm in water at the same working distance.
  • Overall accuracy: Both camera systems, when used underwater, met the millimeter-level accuracy requirements for monitoring coral growth.
  • Calibration frame accuracy: The self-calibration of the reference frame yielded RMSE values of 0.15 mm in air and 0.36 mm underwater for the Lumix system, and 0.40 mm underwater for the GoPro system.
  • Object comparison accuracy: Relative accuracy of 3D point clouds for individual objects (cinder blocks) compared to their air-based Lumix reference were: 1.34 mm (GoPro in air), 1.82 mm (Lumix underwater), and 1.97 mm (GoPro underwater).

Limitations and Future Work

  • Statistical analysis of calibration parameters: The authors note that the Agisoft PhotoScan software did not provide a statistical analysis of self-calibration parameters (e.g., covariance matrix), making it difficult to fully assess the stability and significance of individual parameters.
  • Image network limitations: In underwater scenarios, achieving an optimal and evenly distributed image network can be challenging for divers, leading to gaps in the 3D point clouds.
  • Background influence: While they attempted to mitigate background influence by comparing individual objects, variations in background coverage (blue plastic sheet vs. sand) did affect the overall scene comparison.
  • Future Work: The authors plan to conduct a complete accuracy assessment of 3D point clouds of actual coral patches. They also aim to detect changes in corals in 3D and link these changes to ecosystem studies, specifically focusing on coral reef responses to environmental disturbances and their regrowth rates. Further research will also explore methods to correct for optical shifts (like blue shift) that occur at image borders due to refraction.

Summary updated

Abstract

Recent advances in automation of photogrammetric 3D modelling software packages have stimulated interest in reconstructing highly accurate 3D object geometry in unconventional environments such as underwater utilizing simple and low-cost camera systems. The accuracy of underwater 3D modelling is affected by more parameters than in single media cases. This study is part of a larger project on 3D measurements of temporal change of coral cover in tropical waters. It compares the accuracies of 3D point clouds generated by using images acquired from a system camera mounted in an underwater housing and the popular GoPro cameras respectively. A precisely measured calibration frame was placed in the target scene in order to provide accurate control information and also quantify the errors of the modelling procedure. In addition, several objects (cinder blocks) with various shapes were arranged in the air and underwater and 3D point clouds were generated by automated image matching. These were further used to examine the relative accuracy of the point cloud generation by comparing the point clouds of the individual objects with the objects measured by the system camera in air (the best possible values). Given a working distance of about 1.5 m, the GoPro camera can achieve a relative accuracy of 1.3 mm in air and 2.0 mm in water. The system camera achieved an accuracy of 1.8 mm in water, which meets our requirements for coral measurement in this system.
BibTeX
@inproceedings{guo2016accuracy,
  author = {Tiande Guo and Alessandro Capra and Matthias Troyer and A. Gruen and Andrew J. Brooks and James L. Hench and Russell J. Schmitt and Sally J. Holbrook and Marco Dubbini},
  title = {{ACCURACY ASSESSMENT OF UNDERWATER PHOTOGRAMMETRIC THREE DIMENSIONAL MODELLING FOR CORAL REEFS}},
  booktitle = {˜The},
  year = {2016},
  doi = {10.5194/isprs-archives-xli-b5-821-2016},
}
Figure