Underwater videography is critical to testing the models we’re developing during this project. We use a calibrated stereo pair of video cameras to measure the precise 3-D coordinates of the foraging activities of the fish. The methods for 3-D calibration and measurement using a pair of side-by-side videos are described on the website for the analysis software, VidSync. This page discusses only the hardware we use to obtain our footage.
This camera system is too elaborately customized for succinct description in the methods section of a scientific journal article. Therefore, the detailed description on this webpage serves as a supplement to a much briefer description in our papers.
Because new products are always entering the market, it is unlikely that anyone will want to replicate this exact hardware system in the future. But it may provide a useful template, and several unexpected lessons we learned may prove useful to others deploying similar systems on other hardware in the future.
Basic hardware
We use a pair of Nikon D5300 cameras with corresponding Ikelite #6081.53 polycarbonate underwater housings and Zen Underwater DP-230 9″ glass dome ports. We use two different pairs of fisheye lenses depending on the desired field-of-view: Samyang 8 mm, or Tokina 10-17 mm.
Manual video settings on the Nikon D5300
A major reason we chose a DSLR for video, apart from the availability of good fisheye lenses, was the ability to manually control focus and exposure. We hoped to control depth-of-field with a manual aperture setting, motion blurring with a manual shutter speed setting, and allow the camera to automatically change the ISO sensitivity to maintain a good exposure. Unfortunately, the D5300 allows more limited options. If we fix the shutter speed (S mode), it varies the aperture and ISO together automatically; if we fix the aperture (A mode), it varies the shutter speed and ISO together automatically.
We can fix shutter speed, aperture, and ISO all together in full manual (M) mode, but this only works well in predictably very uniform lighting; otherwise, variation due to passing clouds or tree shade results in poorly exposed video. When consistent direct sunlight was available, we obtained the best video quality using manual mode at f/11, ISO 640, and a 1/125 second shutter speed, which greatly reduces motion blur compared to default settings.
In most cases, we anticipated potential lighting changes and shot in aperture priority (A) mode instead to assure adequate depth-of-field. This often results in slower shutter speeds than we would prefer and moderate motion blurring in still frames, but this is preferable to the overly light or dark footage we obtain in full manual mode under highly variable lighting.
Focus settings
All of our lenses only focus manually, and manual focus is essential anyway when filming this kind of data. There is no opportunity to change the focus once the cameras are placed in the water to observe fish, so we discovered good general-purpose settings specific to our dome port and working distance by trial and error.
Lenses inside a dome port do not focus on the target object itself, but a “virtual image” of the object extremely close to the dome. We obtained the sharpest images when the Tokina 10-17 mm lenses were focused on the “1” in the “1.25” foot mark, and the Samyang 8 mm lenses were focused on the “.” in the “1.2” foot mark.
We use electrical tape to secure the focus settings (and the zoom ring, when we use the Tokina 10-17 mm lens) before placing the cameras in the housings. Otherwise, these rings sometimes accidentally slip during handling, which can ruin a data set by invalidating its stereo calibration or knocking everything out of focus.
External power
Our camera and recorders were powered together by a Goal Zero Yeti 400 solar generator, which was daisy-chained to two 39-AH motorcycle batteries. This seemed to provide enough capacity for all the recording we could fit into 3- to 4-day trips to the field (anywhere from 2 to 5 hours recording per site and 2-3 sites per day). We supplemented this system with three Goal Zero Nomad 20 solar panels, which rarely provided the full 20 watts each that we hoped, but nevertheless were sufficient to power the system almost entirely by themselves when exposed to direct sunlight.
Overriding the 30-minute recording limit
After beginning this project we discovered an almost show-stopping limitation to DSLR videography in general: despite being known for recording high-quality video, most of them cannot record for longer than 30 minutes, because otherwise they would be classified as “video recording devices” and subject to additional taxes in Europe. We would have avoided this problem by using external recorders, which capture the “Live View” feed transmitted by the cameras through HDMI cables, but the “Live View” itself is subject to the same timeout.
However, we found that the Nikon D5300 resets the 30-minute timer when it receives a variety of user inputs, indicating that someone is still “using” the camera. One such input is a half-press of the shutter button, which normally instructs the camera to autofocus. Because we are using manual focus, this instruction is ignored–but it still resets the 30-minute timer. Of course, we cannot press the button inside the underwater housing while recording fish, but we can trick the camera into thinking we’re doing that.
The Nikon MC-DC2 remote release cord sends the same “shutter button half-pressed” signal to the camera when the remote release button is half-pressed, which simply involves a black wire contacting a green wire inside the remote to close a circuit. We found that cutting off one of these cables (leaving just a 2-inch stub that plugs into the cameras) and short-circuiting the black wire to the green wire sends this same signal continuously and prevents the camera from shutting off. Furthermore, it still worked if we shorted the circuit via a kilo-ohm resistor, making the power drain of this hack seemingly negligible. (However, a mega-ohm resistor did not work.)
The procedure for employing this hack was to start the live view transmission while the cameras were still on land, then plug in the remote release stub as the last step before sealing the housings.
Avoiding bubbles and debris on the dome ports
On early trips we found that a disconcerting amount of fine detritus, or sometimes just fine air bubbles, accumulated on the dome ports, especially the bottom half of each port. Some advice online suggested treating the outsides of the ports with Rain-X, and this did seem to help. The bubble problems were mainly on shallow, riffly Panguingue Creek where there are many tiny bubbles in the water column. We found there that it also helped here to place the cameras as deeply in the water, and as far from the riffles, as possible. Even a few inches of depth makes a big difference in the number of bubbles brushing past the dome port and possibly sticking to the surface.
Preventing the cameras from overheating
We initially ran into problems when shooting some lengthy videos: one of the Ninja recorders would “skip,” restarting the recording, resetting its timecode, and showing a yellow kangaroo symbol on the viewer screen. By switching out various equipment between the left and right cameras we eventually narrowed this problem down to one of the Nikons overheating inside its housing when running for a long time in direct sunlight. Apparently the housings insulated the cameras from the cold water sufficiently that the combination of the heat they generated internally, heat they radiated from sunlight falling on the black camera bodies, and perhaps a greenhouse effect inside the clear housings caused them to overheat.
Our permanent solution was to cover the knobs and viewing screen on the housings with masking tape, rough up their exteriors with 220 grit wet/dry sandpaper, wash them with soap to get rid of any finger oils, and spraypaint them with several layers of Krylon Fusion Black (specially made for plastic) and Krylon Matte Finish (a probably unnecessary step to make things less shiny and maybe less alarming to the fish). The black paint on the outside of the housings blocked sunlight from heating the internals, and did so without heating up itself, because it was directly exposed to the cold water.
An unanticipated benefit of blacking out the housings was to prevent some glare we were seeing from the sun at certain angles, which was apparently a result of reflections in the dome port of light passing through the inside of the housing.
Preventing the recorders from overheating
Although painting the housings prevented most of the overheating glitches, some remained. We traced these to the recorders themselves, which generate considerable heat even when sitting in the shade. The sides of each recorder are made of flat aluminum meant to radiate heat, but apparently it did not radiate enough. Therefore, we bolted the recorders flush between two larger sheets of 1/4″ aluminum tall enough to rest in a plastic tub in a pool of cold water without the recorders getting wet. This system very effectively transferred heat into the water and kept the recorders from overheating. If the water itself begins to warm up, which was not usually a problem, the entire tub can be placed in the shallows of the river to cool it down.