Here is an image of a school of reef fish taken with the robot I mentioned in part 4:
And here is one taken very close to a school small tropical fish
A 3 Volt Hankscraft “Display” motor rotates the camera rig at 1 rpm. I put the motor in a PCV housing with the shaft going through a 1/8″ o-ring sealed camera control fitting. I made a hook mount on the shaft and a bracket to attach to the camera bracket on the other end. There is also a switch and a battery holder (2 AA cells) inside the case (photo below).
The upper end is attached to the lower end of a section of a pool noodle float – just enough so there remains about 1kg of negative buoyancy for the whole rig. I use the robot as shown in photo 4 for shots close to the bottom or add a length of aluminium strap to raise the camera off the bottom if required.
A GentLED Auto time lapse module triggers the camera. The camera fits snugly in the Ikelite camera housing and there is no room to place even such a tiny control module as the GentLED Auto. James Gentled was so kind as to make a special GentLED Auto unit with the LED on a short wire extension so I could fit it into the Ikelite housing. I attach the GentLED Auto using a sticky clay-like goo purchased in a stationary store (photo below). I turn on the timer, but not the camera, just before getting into the water and then seal the camera case. The control is set to trigger the camera once every 3 seconds, giving some 20 photos per 360 degrees.
When I have the camera in place I turn it on and check to be sure it is taking photos. Then Freddy and I swim off and spend 15 to 20 minutes taking photos with our other cameras. 20 minutes gives the fish time to get back to their normal behaviour and yields some 400 images. When I return I unhook the float and the motor housing and attach these to each other. Freddy swims off with them and I take the down and up images with the liberated camera.
Post processing becomes a challenge when you have to select 12 to 14 images out of 400 – especially when they contain constantly shifting schools of fish. Unlike land photos, you can’t control the rotational angle of each image because the camera is always moving from wave surge and currents. Sometimes a wave surge will move the camera rapidly – sometimes the focus isn’t correct – and you need an overabundance of images to select from.
Once I have my prime series selected I do all of the color, lighting and contrast corrections to each image in Photoshop before stitching. I save the processed RAW images in TIFF format.
I have tried both PTGui and AutoPano for stitching the underwater panos. I found AutoPano was the easiest and best for stitching these complex images. It works wonders – usually stitching the pano on the first try with practically no adjustments needed. It’s anti-ghosting feature is excellent and I rarely have any half-fish – a big issue when there are hundreds of fish in the image. It also adjusts for color and lighting variations so the end result looks great. After stitching I save the image as a TIFF file and correct any problems in Photoshop. Next I convert the image to cube faces using Pano2VR. I de-fisheye the down shot, open the bottom cube face in Photoshop and cover the nadir with the down shot, carefully matching it with the rest of the bottom view. (I rarely can get the bottom to stitch well in the original image because it is difficult to reposition the camera at exactly the same location). Then I fill in zenith hole in the top cube face and reassemble the cube faces back into a rectilinear image in Pano2VR.