Accurate Bathymetric Maps From Underwater Digital Imagery Without Ground Control

David G Zawada,Christine Kranenburg,Gerald A Hatcher,Andrew C Ritchie,Jonathan A Warrick,Kimberly K Yates,Evan T Dailey

doi:10.3389/fmars.2020.00525

Abstract

Structure-from-Motion (SfM) photogrammetry can be used with digital underwater photographs to generate high-resolution bathymetry and orthomosaics with millimeter-to-centimeter scale resolution at relatively low cost. Although these products are useful for assessing species diversity and health, they have additional utility for quantifying benthic community structure, such as coral growth and fine-scale elevation change over time, if accurate length scales and georeferencing are included. This georeferencing is commonly provided with “ground control,” such as pre-installed seafloor benchmarks or identifiable “static” features, which can be difficult and time consuming to install, survey, and maintain. To address these challenges, we developed the SfM Quantitative Underwater Imaging Device with Five Cameras (SQUID-5), a towed surface vehicle with an onboard survey-grade Global Navigation Satellite System (GNSS) and five rigidly mounted downward-looking cameras with overlapping views of the seafloor. The cameras are tightly synchronized with both the GNSS and each other to collect quintet photo sets and record the precise location of every collection event. The system was field tested in July 2019 in the U.S. Florida Keys, in water depths ranging from 3 to 9 m over a variety of bottom types. Surveying accuracy was assessed using pre-installed stations with known coordinates, machined scale bars, and two independent surveys of a site to evaluate repeatability. Under a range of sea conditions, ambient lighting, and water clarity, we were able to map living and senile coral reef habitats and sand waves at mm-scale resolution. Data were processed using best practice SfM techniques without ground control and local measurement errors of horizontal and vertical scales were consistently sub-millimeter, equivalent to 0.013% RMSE relative to water depth. Survey-to-survey repeatability RMSE was on the order of 3 cm without georeferencing but could be improved to several millimeters with the incorporation of one or more non-surveyed marker points. We demonstrate that the SQUID-5 platform can map complex coral reef and other seafloor habitats and measure mm-to-cm scale changes in the morphology and location of seafloor features over time without pre-existing ground control.

Highlights

Benthic ecosystems throughout the world have been stressed by human impacts, including land-based pollution, overharvesting, coastal engineering, ocean acidification, and climate change (Hughes, 1994; Edinger et al, 1998; Kleypas and Yates, 2009; Storlazzi et al, 2015; Toth et al, 2015; Prouty et al, 2017; Takesue and Storlazzi, 2019)
The reef and rubble were assumed to be stationary between the two surveys, such that they were used in direct point cloud-to-point
With SQUID-5 we collected photo quintets and associated Global Navigation Satellite System (GNSS) positions that were used with SfM processing to generate point clouds, surface models, and orthomosaics (Warrick et al, 2020) of complex seafloor structure that included biotic and abiotic components of benthic habitat such as coral heads, rubble, and sandy ripples (Figures 5, 6, 8)

Summary

Introduction

Benthic ecosystems throughout the world have been stressed by human impacts, including land-based pollution, overharvesting, coastal engineering, ocean acidification, and climate change (Hughes, 1994; Edinger et al, 1998; Kleypas and Yates, 2009; Storlazzi et al, 2015; Toth et al, 2015; Prouty et al, 2017; Takesue and Storlazzi, 2019). New opportunities to map accurately and at high resolution have been made possible by advancements in Structurefrom-Motion (SfM) photogrammetry, which include feature identification and matching in photos; bundle adjustment computations of camera positions, orientations, and lens distortions; multi-view stereo techniques; GPU and networkbased computing. When combined, these techniques can rapidly produce high-density, three-dimensional point clouds with associated color information from source photos (Matthews, 2008; Westoby et al, 2012; Fonstad et al, 2013). With the addition of high-quality digital cameras to novel photographing platforms including drones, and digital sources of historical photos, a resurgence of photogrammetric mapping for natural resource management has occurred during the past decade (Chirayath and Earle, 2016; Storlazzi et al, 2016; Casella et al, 2017; Pizarro et al, 2017; Warrick et al, 2017)

Methods

Results

Discussion

Conclusion