Low-cost small action cameras in stereo generates accurate underwater measurements of fish

https://doi.org/10.1016/j.jembe.2015.02.013Get rights and content

Highlights

  • Small action cameras arranged in stereo generated accurate measurements at less than 5 m distance and 25° angle.

  • Small action cameras were slightly less accurate than handheld cameras in the pool but no difference was observed in situ.

  • Small action cameras provide cheap and sufficiently accurate measurement of fish when compared to handheld cameras.

Abstract

Small action cameras have received interest for use in underwater videography because of their low-cost, standardised housing, widespread availability and small size. Here, we assess the capacity of GoPro action cameras to provide accurate stereo-measurements of fish in comparison to the Sony handheld cameras that have traditionally been used for this purpose. Standardised stereo-GoPro and Sony systems were employed to capture measurements of known-length targets in a pool to explore the influence of the type of camera, distance to camera rig, angle to the optical axis and target speed on measurement accuracy. The capacity to estimate fish length in situ was also compared by measuring the same fish on a coral reef with two baited remote underwater video systems, each fitted with both a GoPro and a Sony camera system. Pool trials indicated that the GoPros were generally less accurate than the Sonys. Accuracy decreased with increased angles and distance for both systems but remained reasonably low (< 7.5%) at 5 m distance and 25° angle for GoPros. Speed of target movement did not result in any consistent decrease in accuracy. In situ measurements revealed a strong correlation (R2 = 0.94) between Sony and GoPro length measurements of the same individual fish, with a slope not different from 1 and an intercept not different from 0, suggesting that GoPro measurement errors do not result in a consistent bias at the level of individual fish. Moreover, the investigation of kernel density functions of the length distribution of the entire fish assemblage indicated that difference in measurement accuracy becomes negligible for purposes of comparing population size structure. We suggest a measurement protocol for the use of GoPro stereo-camera systems that improves accuracy, where distance to target is limited to 5 m and angle to optical axis is restricted to 25°. For distances up to 7 m, angles should be restricted to 15°. This protocol supports the use of small action cameras such as the GoPro system, providing reductions in cost and increases in effective sampling efforts, compared with traditional rigs based on relatively expensive handheld cameras.

Introduction

Accurately measuring change in diversity, relative abundance and population characteristics is a key component of successful marine ecology and conservation. Body size is a particularly important phenotypic metric with a fundamental role in population ecology, providing insights to condition, physiology and behaviour (Peters 1986) and reflecting variability in age structure (Olsen et al., 2004), genetic variability (Olsen et al., 2009) and environmental conditions (Sarma et al., 2008). In marine fishes, size is also closely related to reproductive output and recruitment (Beldade et al., 2012). Moreover, size is a sensitive response to anthropogenic influences such as fishing (Olsen and Moland, 2010), climate change (Genner et al., 2010) and pollution (Farkas et al., 2003).

Low-cost, non-destructive methodologies that enable accurate size estimates of the individual animals that comprise marine communities are increasingly sought. Lethal sampling can be inappropriate if sampling: (1) inside no-take marine protected areas (MPAs); (2) rare or red listed species; (3) species that are restricted by quotas; or (4) in circumstances where the killing of animals to study them raises ethical questions. Traditional underwater visual census (UVC) has provided a wealth of coverage on shallow reef communities but is limited due to depth restrictions imposed by safe SCUBA practices and the need for highly specialised skills for in situ species identification and length estimation. In temperate and tropical reefs, stereo baited remote underwater video systems (BRUVS) are increasingly utilised to assess the diversity and abundance of fish over time and space (Harvey et al., 2007) and to assess impact of anthropogenic activity and management and conservation strategies (Mclean et al., 2011). Stereo-camera systems also provide relatively accurate estimates of individual body lengths (Harvey and Shortis, 1998) and thus provide information related to fish population structure. Customised software (www.seagis.com) applies trigonometric principles and generates estimates of horizontal and vertical orientation (Letessier et al., 2013a), lengths of target (mm), 3-dimensional positioning (x, y, z coordinates) and angle to the optical axis. Estimates of lengths using stereo-BRUVS are demonstrably more accurate and precise than those derived using single cameras (Harvey et al., 2002b) and those estimated during UVC (Harvey et al., 2004).

A great strength of stereo-BRUVS is their ability to generate large data sets with extensive spatial coverage in relatively short time periods across a wide range of depths. For instance, up to 45 one-hour samples can be generated per day with the deployment of 15 individual rigs (Meeuwig, unpublished data), with typical surveys collecting on the order of > 200 samples over a week of vessel time. Stereo-BRUVS do have a number of limitations including (1) the length of time potentially required to process video (Holmes et al., 2013); (2) undersampling of small (e.g., Pomacentridae), relatively cryptic (e.g., Apogonidae, Holocentridae) or similar (e.g., Ctenochaetus striatus vs Ctenochaetus binotatus) species (Mallet, 2014); and (3) general uncertainty surrounding the area of the bait odour plume, such that fish abundance is recorded as a proxy (MaxN) rather than as densities or biomass per square meter (but see Priede and Merrett, 1996). Nevertheless, stereo-BRUVS can cover large areas at a high temporal frequency and are very efficient at surveying highly mobile species such as large predators that are attracted by the bait. Such properties are particularly critical where animals are rare and patchily distributed as is the case for pelagic ecosystems (Letessier et al., 2013b) and/or where top-order predators are of interest. Sampling in these cases can require a high number of replicates for elucidation of spatial and/or temporal patterns. However, potential for broader adoption of stereo-BRUVS has been limited, in part, by costs associated with the acquisition of stereo-camera rigs (approx US$ 4500 per rig; www.seagis.com; accessed August 2014). This cost is primarily linked to the cost of purchasing customised housing and mid- to high range handheld video cameras.

Small action cameras such as GoPros are a relatively novel range of cameras that are receiving interest in their capacity to be used in underwater videography studies because they are of relatively low cost and because they are supplied with a standard underwater housing rated to 60 m. Housings rated to greater depth are also available commercially (see for example http://www.cam-do.com/GoProUnderwaterSolutions.html) such that the total cost of a stereo-camera rig based on GoPros is approximately $1300. Moreover, their shape and size make them well suited for use in stereo (Schmidt, 2012). As they are small and lightweight, they can also be easily and cheaply mobilised to remote field sites in large numbers, as well as fit to lightweight rigs (Letessier et al., 2013a). Moreover, GoPros in particular are widely available, which makes their adoption attractive. However, concerns have been raised with respect to their capacity to generate robust lengths estimate (Letessier et al., 2013a, Letessier et al., 2013b), in part because they utilise rolling shutters and because they have curved ‘fish-eye’ lenses. Rolling shutters can distort the image during moments of rapid motion (Liang et al., 2005), and fish-eye lenses result in greater field of view than flat lenses, but with a barrel distortion, where magnification decreases with distance to the optical axis (Shah and Aggarwal, 1994).

Here we assess the capacity of stereo-GoPros to accurately measure lengths compared with a traditional Sony-based stereo rig using two experimental approaches:

  • 1.

    Standardised stereo-GoPro and Sony systems were employed to capture measurements on targets of known lengths in a pool to explore the influence of camera system, distance to camera system, angle to the optical axis and target movement speed on accuracy of length measurements. This experiment was aimed at determining the technical limits of both camera systems to measure size underwater.

  • 2.

    Two BRUVS rigs, each fitted with a GoPro and Sony camera system, were deployed on a coral reef, in order to estimate the lengths of fish appearing simultaneously in the sampling fields of both camera systems. This experiment aimed at determining the capacity of both systems to measure fish size in real conditions of use at sea based on a standard BRUVS protocol.

Section snippets

Camera systems

Our assessment of stereo-measurements accuracy were conducted using Sony HDR-CX12 and GoPro Hero 2 cameras, both in pool and on a coral reef. Cameras on each stereo pair system were situated 800 mm apart, with an inward convergent angle of 8°. GoPros were set to record at medium field of view (127°), 1080p and 30 FPS. The vertical field of view is set at 93°. Sonys were recording at 1080p, set at 30 FPS, with a horizontal field of view of 68.8° and a vertical field of view of 46°. Both systems

Pool trials

Pool trials resulted in 924 and 1213 measurements with the Sony and GoPro systems, respectively. The overall greater number of measurements obtained with the GoPro system was due to the larger field of view, which enabled measurements at greater angles to the optical axis than the Sony systems (> 34°, Table 1). In general, errors in length estimates were smaller for Sony than GoPro systems (Fig. 3, Fig. 4) but were, for both camera systems, low (< 5%) at angles to the optical axis less than 10°

Discussion

Stereo-camera-based methodologies are becoming increasingly common as a standardised method of surveying fish communities and are showing promise for studies in the mid-water (Letessier et al., 2013b, Santana-Garcon et al., 2014) and for experimental monitoring of captive populations (Letessier et al., 2013a). The growing application of stereo technology to assess body length has been hampered by relatively high costs, partly associated with acquisition of underwater camera technology. Here we

Acknowledgement

We are grateful for guidance and support provided by David Mouillot and Laurent Wantiez. We thank the Government of New Caledonia and the crew of the Amborella for excellent seamanship and support during our camera deployment activities. For technical support during pool experiments, we thank Mr Lloyd Groves and Mr Robert Sanzogni. TBL was supported by the Marine Biodiversity Hub through the Australian Government's National Environmental Research Program (NERP). NERP Marine Biodiversity Hub

References (37)

  • E.S. Harvey et al.

    Calibration stability of an underwater stereo-video system: implications for measurement accuracy and precision

    Mar. Technol. Soc. J.

    (1998)
  • E. Harvey et al.

    Improving the statistical power of length estimates of reef fish: a comparison of estimates determined visually by divers with estimates produced by a stereo-video system

    Fish. Bull.

    (2001)
  • E.S. Harvey et al.

    A comparison of the accuracy and precision of measurements from single and stereo-video systems

    Mar. Technol. Soc. J.

    (2002)
  • E. Harvey et al.

    A comparison of underwater visual distance estimates made by scuba divers and a stereo-video system: implications for underwater visual census of reef fish abundance

    Mar. Freshwat. Res.

    (2004)
  • E.S. Harvey et al.

    Bait attraction affects the performance of remote underwater video stations in assessment of demersal fish community structure

    Mar. Ecol. Prog. Ser.

    (2007)
  • T.H. Holmes et al.

    A comparison of visual- and stereo-video-based fish community assessment methods in tropical and temperate marine waters of Western Australia

    Limnol. Oceanogr. Methods

    (2013)
  • J.B. Juhel et al.

    Occurrence of sub-adult tiger sharks (Galeocerdo cuvier) in a small and very remote atoll lagoon

    Mar. Biodivers.

    (2014)
  • T.J. Langlois et al.

    Similarities between line fishing and baited stereo-video estimations of length-frequency: novel application of Kernel density estimates

    PLoS One

    (2012)
  • Cited by (70)

    • Fussy eaters, bait loss and escapees: How reliable are baited-hook assessments of shark abundance in shallow, coastal waters?

      2022, Journal of Experimental Marine Biology and Ecology
      Citation Excerpt :

      Nevertheless, although hook-timers facilitate a more accurate estimation of fishing time, they cannot provide taxonomic information in the case of escapees or bait loss, nor can they provide information on animals that approached the gear but did not physically engage with it. The use of video equipment to assess species occurrence and abundance has become popular among fish researchers, due in large part to the widespread availability of affordable so-called sport action cameras in recent years (Letessier et al., 2015). This has resulted in a marked increase in the use of baited remote underwater video stations (BRUVS) in fish population studies (Whitmarsh et al., 2017), with the added benefit that the technique is non-invasive and does not require the species under consideration to physically engage with the equipment in order to be ‘captured’ (Cappo et al., 2003).

    • Estimating length composition of fish observed with stereo-video cameras: A simulation study with application to red snapper (Lutjanus campechanus)

      2022, Fisheries Research
      Citation Excerpt :

      Knowing the length and age of sampled fish is vital to many modern stock assessment and population demographic methods (Ono et al., 2015). Consequently, video collection programs are increasingly using stereo-video, which allows the estimation of fish length (Cappo et al., 2003; Cappo et al., 2004; Watson et al., 2005; Watson et al., 2010; Shortis et al., 2013; Letessier et al., 2015; Langlois et al., 2015; Langlois et al., 2020; Schramm et al., 2020). This is an important advance in video data collection, but may pose issues with video enumeration methods that were established primarily to estimate abundance rather than length composition.

    View all citing articles on Scopus
    View full text