Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Performance Evaluation of 3D Keypoint Detectors and Descriptors on Coloured Point Clouds in Subsea Environments

Kyungmin Jung, Thomas Hitchcox, James Richard Forbes

TL;DR

This work addresses the challenge of robust 3D feature extraction for subsea navigation by evaluating 3D keypoint detectors and descriptors on coloured underwater point clouds. A novel colourisation pipeline fuses an underwater laser scan with RGB imagery to create coloured point clouds, enabling assessment of colour-enhanced descriptors within a bathymetric SLAM-like pipeline. An extensive experimental study (8 detectors × 6 descriptors, 48 pairs) analyzes repeatability under rotation and noise, matching success across environment types, and full pipeline alignment using coarse and fine registration steps. The findings identify ISS keypoints and USC descriptors as the most robust combination for underwater loop closures, while colour information does not consistently improve performance under low illumination; the colourisation approach, however, enables future colour-aware descriptor development and multimodal fusion for subsea navigation.

Abstract

The recent development of high-precision subsea optical scanners allows for 3D keypoint detectors and feature descriptors to be leveraged on point cloud scans from subsea environments. However, the literature lacks a comprehensive survey to identify the best combination of detectors and descriptors to be used in these challenging and novel environments. This paper aims to identify the best detector/descriptor pair using a challenging field dataset collected using a commercial underwater laser scanner. Furthermore, studies have shown that incorporating texture information to extend geometric features adds robustness to feature matching on synthetic datasets. This paper also proposes a novel method of fusing images with underwater laser scans to produce coloured point clouds, which are used to study the effectiveness of 6D point cloud descriptors.

Performance Evaluation of 3D Keypoint Detectors and Descriptors on Coloured Point Clouds in Subsea Environments

TL;DR

This work addresses the challenge of robust 3D feature extraction for subsea navigation by evaluating 3D keypoint detectors and descriptors on coloured underwater point clouds. A novel colourisation pipeline fuses an underwater laser scan with RGB imagery to create coloured point clouds, enabling assessment of colour-enhanced descriptors within a bathymetric SLAM-like pipeline. An extensive experimental study (8 detectors × 6 descriptors, 48 pairs) analyzes repeatability under rotation and noise, matching success across environment types, and full pipeline alignment using coarse and fine registration steps. The findings identify ISS keypoints and USC descriptors as the most robust combination for underwater loop closures, while colour information does not consistently improve performance under low illumination; the colourisation approach, however, enables future colour-aware descriptor development and multimodal fusion for subsea navigation.

Abstract

The recent development of high-precision subsea optical scanners allows for 3D keypoint detectors and feature descriptors to be leveraged on point cloud scans from subsea environments. However, the literature lacks a comprehensive survey to identify the best combination of detectors and descriptors to be used in these challenging and novel environments. This paper aims to identify the best detector/descriptor pair using a challenging field dataset collected using a commercial underwater laser scanner. Furthermore, studies have shown that incorporating texture information to extend geometric features adds robustness to feature matching on synthetic datasets. This paper also proposes a novel method of fusing images with underwater laser scans to produce coloured point clouds, which are used to study the effectiveness of 6D point cloud descriptors.
Paper Structure (16 sections, 10 equations, 10 figures, 6 tables)

This paper contains 16 sections, 10 equations, 10 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Coloured Point Cloud Generation
  4. Underwater Field Dataset
  5. Submap Generation
  6. Coloured Point Clouds
  7. Experimental Method
  8. Detector/Descriptor Pairs
  9. Keypoint Evaluation
  10. Feature Evaluation
  11. Full Pipeline Evaluation
  12. Results and Discussion
  13. Keypoint Evaluation
  14. Feature Evaluation
  15. Full Pipeline Evaluation
  16. ...and 1 more sections

Figures (10)

  • Figure 1: Point cloud alignment via feature matching is shown here. Fig \ref{['cover_image_a']} shows two point cloud submaps generated from laser scans of a subsea pipeline. Note the submaps are separated in the vertical direction for easier visualization. The extracted keypoints from the two submaps are shown as red dots. The aligned submaps are shown in Fig \ref{['cover_image_b']}.
  • Figure 2: A laser line scanner, Insight Pro (top), and an imaging system, Observer Micro (bottom), provided by Voyis Imaging Inc. are used to collect point cloud and image data, respectively.
  • Figure 3: The vehicle trajectory over a pipeline target is shown as a black line, while loop-closure crossings are shown as red circles. The colour of the underlying point cloud scan represents relative depth across the dataset.
  • Figure 4: A shipwreck is reconstructed from laser scans (left). Occluded points from the camera's viewpoint are removed using the method from Katz2007Direct (right), so as to only add colour to points visible to the camera.
  • Figure 5: A 3D point projected to an image frame ${\mathcal{I}}$, shown as a red dot, is visible in multiple images at different pixel locations. Each colour candidate is weighted with the corresponding value of the 2D Gaussian. A unique colour is assigned by taking the sum of all weighted colour candidates \ref{['eq:colour']}.
  • ...and 5 more figures