# Real-Time Dense 3D Mapping of Underwater Environments

**Authors:** Weihan Wang, Bharat Joshi, Nathaniel Burgdorfer, Konstantinos Batsos, Alberto Quattrini Li, Philippos Mordohai, Ioannis Rekleitis  
**arXiv:** [2304.02704](https://arxiv.org/abs/2304.02704)  
**Date:** 2023-04-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.02704v1/

## Key points

- The paper presents a novel approach to real-time dense 3D reconstruction for Autonomous Underwater Vehicles (AUVs) that are resource-constrained.
- Underwater vision-guided operations are extremely challenging due to the combination of 3D motion in the presence of external forces, limited visibility, and absence of global positioning.
- The proposed approach uses SVIn2, a robust VIO method, together with a real-time 3D reconstruction pipeline for obstacle avoidance and effective path planning.
- The authors provide extensive evaluation on four challenging underwater datasets and demonstrate that their pipeline produces comparable reconstruction with that of COLMAP, which is considered the state-of-the-art offline 3D reconstruction method.
- Their proposed approach achieves high frame rates on a single CPU without relying on GPUs.
- Related work section focuses on approaches tailored for underwater deployment such as visual SLAM algorithms leveraging acoustic, inertial and altimeter/depth sensors in addition to cameras.
- The proposed approach leverages multi-threaded CPU implementations (OpenMP) without relying on GPUs by carefully designing every step of the stereo matching process.
- This paper's proposed approach is essential for autonomous operation which is central to environmental monitoring marine archaeology resource utilization and underwater cave exploration.

## Summary

This paper presents a novel approach to real-time dense 3D reconstruction for Autonomous Underwater Vehicles (AUVs) that are resource-constrained. Underwater vision-guided operations are extremely challenging due to the combination of 3D motion in the presence of external forces, limited visibility, and absence of global positioning. To address these challenges, the authors propose using SVIn2, a robust VIO method, together with a real-time 3D reconstruction pipeline. The proposed pipeline is essential for obstacle avoidance and effective path planning that require online dense reconstructions of the environment. The authors provide extensive evaluation on four challenging underwater datasets and demonstrate that their pipeline produces comparable reconstruction with that of COLMAP, which is considered the state-of-the-art offline 3D reconstruction method. Moreover, their proposed approach achieves high frame rates on a single CPU without relying on GPUs. The related work section focuses on approaches tailored for underwater deployment such as visual SLAM algorithms leveraging acoustic, inertial and altimeter/depth sensors in addition to cameras. The authors use the approach of Rahman et al. [9] to obtain robust camera pose estimates by fusing visual and inertial information in real time. Beall et al. [14] demonstrate accurate sparse 3D reconstruction of underwater structures from stereo videos while Joshi et al. [15] augment a visual SLAM algorithm so that after loop closures, the map is deformed to preserve the relative pose between each point and its attached keyframes. Amongst few authors who tackle dense underwater stereo mapping, Queiroz-Neto et al. [16] model light propagation to overcome poor contrast and illumination while general-purpose dense 3D reconstruction algorithms operating in online mode on one or more video streams have been evaluated qualitatively since appropriate benchmarks with ground truth data are not available. The proposed approach leverages multi-threaded CPU implementations (OpenMP) without relying on GPUs by carefully designing every step of the stereo matching process. The authors provide a detailed description of their implementation of matching cost computation which estimates the likelihood for each possible disparity that can be assigned to a given pixel of the reference image. Overall, this paper's proposed approach is essential for autonomous operation which is central to environmental monitoring marine archaeology resource utilization and underwater cave exploration. The authors' pipeline provides real-time dense 3D reconstruction that is crucial for obstacle avoidance and effective path planning in underwater environments.