[Paper Review] Semihierarchical Reconstruction and Weak-area Revisiting for Robotic Visual Seafloor Mapping

Mengkun She, Yifan Song+2 more|arXiv (Cornell University)|Aug 11, 2023

Underwater Vehicles and Communication Systems54 references4 citations

TL;DR

This paper presents a semihierarchical reconstruction and weak-area revisiting framework for robust, large-scale 3D seafloor mapping using AUV-based visual data. By combining SLAM and Structure-from-Motion with color normalization and iterative revisiting of poorly constrained regions, the system achieves geometrically consistent reconstructions even under challenging underwater lighting and imaging conditions, significantly improving pose graph connectivity and reconstruction quality.

ABSTRACT

Despite impressive results achieved by many on-land visual mapping algorithms in the recent decades, transferring these methods from land to the deep sea remains a challenge due to harsh environmental conditions. Images captured by autonomous underwater vehicles (AUVs), equipped with high-resolution cameras and artificial illumination systems, often suffer from heterogeneous illumination and quality degradation caused by attenuation and scattering, on top of refraction of light rays. These challenges often result in the failure of on-land SLAM approaches when applied underwater or cause SfM approaches to exhibit drifting or omit challenging images. Consequently, this leads to gaps, jumps, or weakly reconstructed areas. In this work, we present a navigation-aided hierarchical reconstruction approach to facilitate the automated robotic 3D reconstruction of hectares of seafloor. Our hierarchical approach combines the advantages of SLAM and global SfM that is much more efficient than incremental SfM, while ensuring the completeness and consistency of the global map. This is achieved through identifying and revisiting problematic or weakly reconstructed areas, avoiding to omit images and making better use of limited dive time. The proposed system has been extensively tested and evaluated during several research cruises, demonstrating its robustness and practicality in real-world conditions.

Motivation & Objective

• To address the challenge of inconsistent and incomplete 3D reconstructions in deep-sea visual mapping due to non-uniform illumination, light attenuation, and scattering.
• To improve the robustness of visual mapping systems in underwater environments where standard on-land SLAM and SfM methods fail or diverge.
• To develop an efficient, automated workflow that maximizes image utilization during limited AUV dive time by identifying and revisiting weakly registered areas.
• To enhance feature matching and pose graph consistency through color normalization and hybrid SLAM-SfM processing.
• To produce high-resolution, geometrically accurate 3D reconstructions of large seafloor areas suitable for scientific and ecological analysis.

Proposed method

• The system employs a hybrid SLAM and Structure-from-Motion (SfM) pipeline to balance speed and accuracy, leveraging keyframe-based tracking and incremental optimization.
• It applies color normalization to reduce illumination inconsistencies caused by co-located artificial lighting, improving feature correspondence reliability.
• A semihierarchical reconstruction strategy organizes the mapping process into levels of detail, enabling efficient handling of large-scale datasets.
• Weak-area detection identifies images with insufficient relative pose constraints, triggering targeted revisiting to improve connectivity and reduce reconstruction gaps.
• The pose graph is optimized iteratively after revisiting, ensuring improved consistency and reduced drift across the entire reconstruction.
• Dense Multi-View Stereo (MVS) and meshing are applied using OpenMVS on both original and normalized images to generate high-resolution textured 3D models.

Experimental results

Research questions

• RQ1 How can visual mapping systems be made robust to non-uniform illumination and radiometric degradation in deep-sea environments?
• RQ2 To what extent does color normalization improve feature matching and pose graph connectivity in underwater visual reconstruction?
• RQ3 Can a hybrid SLAM-SfM approach achieve faster processing than incremental SfM while maintaining comparable reconstruction accuracy?
• RQ4 How effective is weak-area revisiting in recovering images that would otherwise be omitted or poorly registered?
• RQ5 What impact does image preprocessing and reconstruction pipeline design have on the final quality of large-scale seafloor 3D models?

Key findings

• Color normalization significantly improved view graph connectivity, especially between side tracks, by reducing illumination inconsistencies.
• The use of color-normalized images led to more reliable SIFT feature matching, as evidenced by increased match density in the view graph.
• Weak-area revisiting successfully eliminated all weakly constrained regions, achieving 100% improvement in pose graph robustness.
• The final 3D reconstructions showed clear compensation for lighting effects, resulting in visually consistent and detailed meshes.
• The system achieved high-resolution, geometrically accurate seafloor models across diverse datasets, including challenging hard-scene conditions.
• The hybrid SLAM-SfM approach ran significantly faster than incremental SfM while maintaining comparable reconstruction quality.