Lenses and ports for a custom downward-facing ROV camera

Third Party Products Hardware
1

Calling all underwater photographers!

I’d like to request some feedback and recommendations re: a custom camera / lens solution.

We use BlueROV2s at the Seattle Aquarium to conduct benthic surveys within kelp forests throughout coastal Washington (see here on the forums for details). We are now in the process of upgrading our downward-facing camera from a GoPro 12 to the full-frame Sony ILX-LR1. Our objective: obtain high-resolution imagery of benthos, and via trained ML models, extract information across photos re: kelp (and other algae), invertebrates, fishes, and substrate (seafloor) type. We are also keen to use the ILX-LR1 for photogrammetry applications.

For downward-facing illumination, we have integrated x4 DeepSea LSL-2075 SeaStrobe lights into the ROV with power coming supplied via the Outland Topside Power Supply – these 75 degree beam angle lights each provide 10,000 lumens at 100% power output (100W each), enabling ample light to optimize camera shutter speed, ISO, etc. (light integration, BOM, etc., all contained here; lights depicted below at 50% power output).

Re: ROV parameters when conducting photo surveys: given the dark, turbid waters of coastal Washington (and of temperate rocky reef ecosystems in general), our ROV surveys take place at an altitude of 0.8m above the seafloor, at speeds of ~ 0.125m/s. We do not want to manually focus or make camera adjustments on the fly.

Re: fitting a custom housing / camera / lens / dome vs flat port all underneath the BlueROV2 such that it faces downwards, we clearly face a tradeoff between housing size and optics. We want a compact yet optically robust (minimal distortion) solution.

We’ve tested the following lenses:

  • Sony FE 16-35mm f/2.8 G
  • Sony FE 16mm f/1.8 G
  • Sony FE 20mm f/1.8 G
  • Sony FE 24mm f/2.8 G

We tested the 16-35mm zoom lens with a flat BR acrylic port (results here), and the three prime lenses with a 5" BR dome (results here).

Unsurprisingly, the dome does a fantastic job of correcting distortion, thus we’d like to use a dome (or some type of wet wide lens) if possible. As you can see from the photos below at an altitude of 0.8m and with the dome, the 16mm prime (left) and 24mm prime (right) capture a field of view of 1.8 * 1.2 = 2.16 m^2, and 1.4 * 0.95 = 1.33 m^2, respectively.

Between these two lenses, the Sony FE 24mm f/2.8 G is highly compact but with a relatively narrow field of view – this would clearly provide unparalleled pixel density beneath the ROV, enabling effective data extraction. On the other hand, the Sony FE 16mm f/1.8 G is larger, likely necessitating a larger housing, but it provides a fantastic field of view which would benefit photogrammetry applications.

As far as I’m aware, our options for pairing lenses with ports includes the following:

  • 16mm: dome or WACP-2 (the latter of which is too expensive).
  • 20mm: dome or WACP-2 (same as above)
  • 24mm: dome or WACP-2 (same as above)
  • 28mm: dome or WWL-1B, WACP-C, WACP-1B (the latter of which is too large)

At present, I like the idea of a 16mm within a dome, or a 24mm within a dome, especially as I believe the lens would be nested forward slightly within the dome such that the entrance pupil of the lens aligns with the dome center, thus reducing some of the total height of the downward-facing camera housing. That being said, from a size perspective, the WWL-1B may be tenable.

My question to the forum: does anyone have experience with dome ports or wet wide lenses they’d be willing to share? Given our 0.8m ROV survey altitude (w/ instances of 0.6 – 1.2m ROV altitudes), does anyone have any insight or recommendations?

Thanks everyone!

2

What is your final output, Orthomosaics?

You go into depth about distortion, if data capture is purely for photogrammetry leading to orthomosaics, why not go fisheye? I appreciate that native E-mount fisheye are few (if in-fact ‘a thing’) but a metabones mk5 EF-E adaptor mean you can mount the Canon EF 8-15mm f/4 L USM fisheye, and the Canon EF 15mm f/2.8 fisheye lenses.

I have used the former lens very successfully since 2016 for photogrammetry on full framed cameras, and the latter on maybe a dozen occasions for photogrammetry. Size wise, the latter is a tiny lens - though discontinued many years ago so will have to be bought second hand.

Oh, I also forgot, as Sony, you could also look for an adapted Nikonos RS 13mm fisheye - these lenses are thought to be some of the best UW glass available, as they were designed specifically for UW use, as opposed to topside lenses put behind a dome (they are like rocking horse sh1t though)

3

Thanks for your response, @3dMB!

Regarding data derived from our downward-facing photographs, we have three distinct types:

  1. Percent-cover metrics for aggregate taxa and categories (e.g., brown, red, green, and encrusting algae, sessile invertebrates, and substrate type . . .things we can’t individually “count” easily). We train classification models to classify ~ 28 categories of percent-cover within 224 x 224 pixel “patches,” and we calculate the overarching percent-cover of each category within a photograph (nine of our percent-cover categories are displayed below, with n=36 examples of our 224 x 224 pixel patches per category).

  1. Density metrics of mobile and/or individually conspicuous individuals (e.g., fish, crabs, sea stars, sea urchins, large anemones). We train object detection models to gather these data.

  2. Photogrammetry models with point clouds to calculate various geometric features, overlay textured mesh.

Of these, the percent-cover data are the most sensitive to distortion, thus if we went with a wide-angle lens, we’d likely only make percent-cover inference on a central, cropped part of each image (which should be fine assuming resolution / pixel density within that cropped portion is sufficient for our classification model to operate accurately).

The density data would likely benefit from an expanded field of view associated with a wide-angle lens, as would photogrammetry. So, at the moment I’m hoping to pursue as large a field of view as possible, while still packaging everything underneath the ROV within the BR Payload Skid.

The Canon EF 15mm f/2.8 is very interesting! Thank you for flagging – it could perhaps pair nicely with a dome port.

And I’ve heard of the legendary Nikonos lenses . . . it’s too bad no one is making a similar product anymore – I’d love to make use of one of those, though I’m a little wary given how rare they are and that we want whatever system we utilize to be readily replicable by others . . .

4

If rectilinear is more desirable, then the canon EF 14mm f2.8 L is very good, as is the 11-24, though the latter is a bloody big lens.

With the “while still packaging everything underneath the ROV within the BR Payload Skid”, payload skids are very easy to design and manufacture, so suggest you don’t allow that to be a limiting factor. I have had 2 bespoke skids made:

IMG_7254
IMG_72541920×1440 646 KB

Photogrammetry skid which is lower profile than the BR skid, with adjustable pitch camera mount

IMG_1717
IMG_17171920×1440 501 KB

Camera skid larger profile to the standard BR skid, shown with Canon R6 mk2 in Nauticam, also takes RED Komodo-X or RED Raptor Nauticam housings.