Dry‐lab benchmarking of a structure from motion workflow designed to monitor marine benthos in three dimensions

Abstract

Structure from motion (SfM) has emerged as a popular method for characterising marine benthos (seabed organisms), particularly in clear, tropical waters. However, there are many environmentally sensitive benthic organisms inhabiting temperate waters, including the reef‐forming glass sponges of the north‐east Pacific Ocean. Broader questions are raised, not just about whether SfM is a capable spatial data acquisition and ecological characterisation method in temperate waters; but whether a systematic assessment of capture methods in dry and wet laboratory conditions reveals critical relationships between SfM parameters, data products and their implications for underwater surveys. This paper, the first of two empirical assessments, reports on a series of dry‐lab tests quantifying the impact that lighting, camera type, camera settings and capture strategy have on data accuracy. These tests provide a crucial accuracy baseline for subsequent wet‐lab and field‐based surveys, revealing that photographs captured from a controlled and stable platform produce superior data products. While the measurable differences were small, they may be critical for accurate change detection in temperate environments.

Evaluating the 3D Integrity of Underwater Structure from Motion Workflows

“…As dive time, and thus data collection, is constrained by depth, identifying the quantity and configuration of photographs required to maximise SfM performance was critical. While underwater SfM workflows have proven to be a relatively low-cost, simple, and informative method of documenting the 3D structure of benthic organisms in tropical regions, this wet-lab study, as well as prior dry-lab research (Lochhead and Hedley, 2021), were necessary steps preceding subsequent field surveys (Lochhead and Hedley, 2020).…”

“…This choice, however, makes it difficult to draw direct comparisons between the outcomes from this study and those quantifying other morphological characteristics (e.g., surface area and volume) (Olinger et al., 2019; Prado et al., 2021). Furthermore, the 3D accuracy of SfM modelling is influenced by variables such as the chosen equipment, data‐capture strategy, data format, and the variable physical properties of the subject (e.g., sponge) and the context (e.g., location, water properties, and light) in which it is surveyed (Kersten and Lindstaedt, 2012; Thoeni et al., 2014; Capra et al., 2015; Lavy et al., 2015; Lochhead and Hedley 2021), making the performance or suitability of a technique (i.e., SfM) hard to overtly express without careful consideration of context.…”

“…The decision to start small was intentional; selected to identify what is ultimately possible with isolated colonies prior to developing a reef‐scale SfM workflow. The lessons learned in the dry‐lab (Lochhead and Hedley, 2021) and wet‐lab tests highlight the challenges associated with underwater surveying (e.g., time constraints, lighting and camera stability) and their implications for 3D model accuracy. This research represents an important and necessary step towards monitoring glass sponge morphology in 3D.…”

“…This paper reports on the second phase of a three‐phase research programme (1 = dry‐lab, 2 = wet‐lab and 3 = fieldwork) designed to establish foundational temperate marine SfM data quality benchmarks. The dry‐lab research (Lochhead and Hedley, 2021) evaluated the performance of HEXYZ‐1 and the SfM workflow in a controlled dry environment – establishing an optimal lighting configuration and a set of performance benchmarks for different cameras, camera settings and photograph/video capture strategies. By contrast, the wet‐lab research presented here evaluates the effect that the (simulated) low light levels and cold, turbid conditions typical of temperate marine environments and glass sponge habitats have on the performance of the SfM workflow.…”

“…There were three main objectives for this research: (1) to assess the performance of a custom‐made SfM data‐capture platform (named HEXYZ‐1) in a controlled underwater environment; (2) to compare the results of wet‐lab SfM surveys against their dryland equivalents and use manual measurements to quantify the performance of the underwater SfM workflow; and (3) to determine the optimal number and configuration of photographs so as to maximise model accuracy while accounting for the temporal constraints (approximately 20 min per dive) imposed by subsequent field‐based data collection at depths of 20–25 m. The wet‐lab research discussed here was an essential step in this research programme because it was informed by the initial dry‐lab system and SfM workflow performance tests (Lochhead and Hedley, 2021) and informed subsequent field‐based surveys of glass sponge morphology (Lochhead and Hedley, 2020). This approach not only delivers an operable field survey architecture but also establishes the data science and data quality benchmarks necessary to understand SfM performance in temperate marine environments.…”