Experimental test for correct lens position within a dome-port

[Klaus]

There is apparently an experimental solution to solve the lens positioning problem within a dome-port. I found a publication (see below) from a team that calibrates photogrammetry cameras for deep-sea rovers at an oceanographic research institute. Perhaps it is already well-known in this community – but I figured I’d share it anyways. Note: This is not about choosing the right dome size, curvature etc. – just the lens position relative to the dome.

The idea is quite straightforward: The goal of the dome-port is to avoid refraction by permitting the light to traverse the water-glass-air interfaces at a 90° angle – there should be no difference between air and water then. This is only possible when the lens is in the perfect position, hence our dilemma. If the position is not optimal, there will be refraction (grossly oversimplyfied: the dome will act a little bit like a flat port). Now the trick is to TAKE A SPLIT SHOT of a straight object that enters the water. If the lens is well positioned, there will be no refraction and the lines continue perfectly. Otherwise, the lines will be displaced and slightly enlarged or reduced in size.

In the publication they had the camera on a screw-mount and could adjust its position (and hence the lens) „live“. For a housed „normal“ camera one would vary the port extensions until the best possible image is obtained.

I tested this with my newly acquired (second-hand) Athena Fisheye dome on an Olympus Pen housing with the Rokinon/Samyang 7,5 mm fisheye. This is probably a sub-optimal setup because the dome was designed for the Panasonic 8mm lens which is longer than the Samyang. The image clearly shows that the lines are displaced a little bit. It’s best if you zoom in to the air-water-interface on the right, which is nice and „flat“. According to the publication, I should thus vary the lens position = spacer length. If I understood the publication right, then a displacement towards the outside (as in my test-shot) translates to shortening the spacer (which I can’t – there’s none) whereas a displacement to the inside indicates that a longer spacer is required. I don’t own a spacer and thus cannot try to aggravate the issue either for further demonstration.

This seems like a simple test to perform once you have the lens, the dome and a collection of spacers at your disposal. It is unfortunately not helpful if you are trying to figure out what to order. But perhaps with the help of friends or at a workshop this can be of interest.  

Certainly my representation of the physics here is oversimplified, certainly there are many more things to consider - I am not claiming to be a specialist. But experiments don't need to be absolutely perfect as long as they are informative and useful.

Publication:

Mengkun She, Yifan Song, Jochen Mohrmann and Kevin Köser: Adjustment and Calibration of Dome Port Camera Systems for Underwater Vision (2019), part of Lecture Notes in Computer Science, https://doi.org/10.1007/978-3-030-33676-9_6

(you need a subscription for this, but try Google with Authors / Title before you pay…)

[dentrock]

Thanks for a fascinating post. I will experiment when I get the chance, as I have a collection of 'spacers'!

While we can get to a certain point by measuring EPs and aligning those as best we can with calculated optical centres for domes, a method to test the chosen dome position is helpful, and will allow us to see 'how close is good enough' (at different apertures).

[Guest]

The test is useful as last step when you have already determined that the dome will contain the lens field of view and only if you have a lens that will focus as close as your test rig aka works better in a pool that in a sink

it is possible in most cases to predict everything without even buying the parts

obviously if you have extension rings in 5mm step and you want to fine tune this may as well work with limited knowledge of the lens but almost nobody is

in this situation 

[Chris Ross]

Interesting post, it would be an ideal method for housing manufacturers I would think - all they would have to do is half fill their test tanks to try it out.

[Guest]

1 hour ago, Chris Ross said:

Interesting post, it would be an ideal method for housing manufacturers I would think - all they would have to do is half fill their test tanks to try it out.

A difference of 2 cm only results in 2-3 degrees difference which is difficult to work out at home you need a very stable rig ideally a dry part with macro slide (nauticam has it) 

this doesn’t ensure the optical quality is maximised (that depends on dome size) and only fully work with perfect hemispheres that contain the lens field of view 

i do a test like this at the end of my process not at the beginning 

[Klaus]

Yes, as I pointed out this will not help for assembling an order. I also think (but haven’t tested) that the effect kind of scales with the camera-object distance, hence when comparing configurations you probably need to get the camera into the same spot, using a  tripod or some other stationary holding thing. Pool or plastic box does not influence the physics, but you need a calm water surface for the split.
I think that as I progressively succumb to GAS, I may come back to this once in a while and play around with the available combinations. I can then maybe find the best solution available for me, not necessarily the ideal one (as mentioned in the first response). And while this test can keep me amused on a rainy winter weekend, it certainly won’t improve composition or diving skills - this is what’s currently limiting most of my shots.
I just like the concept; it is charmingly simple and direct, test the port exactly to what it is intended for. In the publication this is only the first step, they then continue on with simulations and calculations to deal with the errors that REMAIN after the alignment. This is obviously of no relevance for me but apparently improves the photogrammetry. 
 

[Klaus]

On 3/18/2024 at 12:39 AM, dentrock said:

Thanks for a fascinating post. I will experiment when I get the chance, as I have a collection of 'spacers'!

 

I think the easiest may be if you fix a test-object to the cold-shoe of your housing via a strobe-arm - I made a sketch in case this sounds cryptic. That way it may be fairly straightforward to get a small series of comparable shots with different length extensions, perhaps quicker that mounting it on a tripod again every time.

In the article I found they say "misaligned front" when the uw part is enlarged (and vice-versa); I interpret this as the rays from the dome converge in front of the lens EP, but perhaps it is the lens that's in front of the convergence point? If you can do the test, I'd appreciate to know about your conclusion.   I just don't have those toys (yet) to do it myself.. 😞

If you look at Phil Rudine's recent post on the 10 mm Laowa and dome-position test, there's a split with a floating pool thermometer and you can see that the uw-part is slightly enlarged. Perhaps that's exactly why he took this shot, after all it's quite intuitive - we want those lines form the thermometer (aka dive-boat) to continue perfectly straight into the water.

[Guest]

You don't need any strobes a white sink works

The issue is to keep the dome exactly in the middle which needs a base

The two side by side are extensions that differ in 5mm. The longest has a small vignette in air the shortest is worse

For reference the worst is Nauticam suggested extension for the lens, the best is my calculated extension for the lens...

 

[Phil Rudin]

On 3/20/2024 at 4:15 PM, Klaus said:

😞

If you look at Phil Rudin's recent post on the 10 mm Laowa and dome-position test, there's a split with a floating pool thermometer and you can see that the uw-part is slightly enlarged. Perhaps that's exactly why he took this shot, after all it's quite intuitive - we want those lines form the thermometer (aka dive-boat) to continue perfectly straight into the water.

 

 

What is most interesting is that the referenced photo was taken with the 130 AOV Laowa 10mm in a 230mm dome with a 20mm extension at F/22. In the Marelux 140mm fisheye port with a 40mm extension and  both shades for the lens and port removed the results at F/22 are quite similar even with the subject a bit further away from the dome, I am talking an inch or two. Notice the slight vignette in the lower corners. This ultra wide rectilinear lens is what keeps the lines straight from air to water. Also I don't imagine anyone here would have thought a rectilinear lens this wide would be working this well in a port largely used only for fisheye lenses. Because the 10mm focuses at 12cm unlike the Sony FE 16-35mm F/2.8 II which close focuses at 22cm and so needs a larger port even with a much narrower AOV the 10mm has exceeded my expectations. 

 

I would of course use the larger 230 or 12 inch ports for splits but it is nice to know it can be done with such a small port.

 

 

[Klaus]

1 hour ago, Interceptor121 said:

 

 

Nice picture, thanks a lot! Which one has the longer extension- the one on the left or the one on the right?

Sorry if I don’t get it when I should…

[Klaus]

Dome port physics aside, I have to say it is a nice thermometer… I like it.

[Phil Rudin]

2 minutes ago, Klaus said:

Dome port physics aside, I have to say it is a nice thermometer… I like it.

Thanks I got tired of shooting the well worn yellow duck. 

 

[Guest]

1 hour ago, Klaus said:

Nice picture, thanks a lot! Which one has the longer extension- the one on the left or the one on the right?

Sorry if I don’t get it when I should…

The one that’s almost straight 

we are talking 5mm it isn’t that apparent which is why I always do this at the end

i already knew the answer to the question this is only useful for lenses with narrower field of view

[Guest]

This image should make it more apparent

You can see that net of the water line that was moving the one on the left is straight and the one of the right is not

 

In this specific case though you did not need anything particularly fancy the lens is a 16-35 that is wider than 16mm it has a field of view of around 109 degrees which is at the limit of the 180 wide angle port so you push until it vignettes and go back

From here you can see that those 5mm loose a few degrees of fied of view

 

[Klaus]

Thanks for the clarification. So let’s sum this up:

If the lines in the test shot continue with an offset to the outside of the image (i.e. the uw part is slightly enlarged), you need to increase the distance between lens and port. That‘s the opposite of my initial interpretation of the publication text. I suppose that if you go beyond the optimum, then the lines will be off-set to the inside. And even a 5mm difference from the optimal dome position is visible in this test.

 

Whether 5 mm matter in practice is probably a different question and practical considerations, such as vignetting or plain availability of parts are also limiting the options. For example going back to my own shot I wold thus need to add an extension to avoid the offset, but there is already some vignetting so I won‘t do this. 

It‘s reassuring that theory and test come to the same conclusion (as they should). With the calculations you can get an answer BEFORE you have the parts in hand, provided all the required information is available. 

Cool!

[Guest]

1 hour ago, Klaus said:

Thanks for the clarification. So let’s sum this up:

If the lines in the test shot continue with an offset to the outside of the image (i.e. the uw part is slightly enlarged), you need to increase the distance between lens and port. That‘s the opposite of my initial interpretation of the publication text. I suppose that if you go beyond the optimum, then the lines will be off-set to the inside. And even a 5mm difference from the optimal dome position is visible in this test.

 

Whether 5 mm matter in practice is probably a different question and practical considerations, such as vignetting or plain availability of parts are also limiting the options. For example going back to my own shot I wold thus need to add an extension to avoid the offset, but there is already some vignetting so I won‘t do this. 

It‘s reassuring that theory and test come to the same conclusion (as they should). With the calculations you can get an answer BEFORE you have the parts in hand, provided all the required information is available. 

Cool!

No you got it wrong

in both cases extension too long or too short the field of view in water drops

if the extension is too long you get pincushion distortion 

if it is too short you get barrel distortion 

so you will not get any visual indication of what the issue is unless you have a perfectly rectilinear lens and you shoot a grid

which is why this method at practical level is not useful until you are very close to they correct extension and is totally invalid for fisheye optics

[Klaus]

34 minutes ago, Interceptor121 said:

if the extension is too long you get pincushion distortion 

if it is too short you get barrel distortion 

 

That is exactly what I meant, and if the information is correct that the image on the LEFT is with a longer spacer then that fits the interpretation. The shorter extension leads to barrel distortion and that causes a displacement to the outside (because that distortion does not happen in air).

 

I‘m not trying to give a lecture on optics but to point out a simple test that will indicate in which direction one should make a change (displacement to the outside > longer, displacement to the inside > shorter).

If that is true, then it is quite likely that already the second configuration tested will be the best one available to a given person. 
 

Let‘s just leave it with this and see whether anyone finds it useful. That is the main point here. For me it already is, because I couldn‘t find the information for either the Samyang lens or the Athena port (lack of perseverance). But I now know that this will have some limitations that I can only solve at the expense of more vignetting, so I won‘t bother.

GAS cured, thanks to your help!
 

 

 

[Guest]

6 hours ago, Klaus said:

That is exactly what I meant, and if the information is correct that the image on the LEFT is with a longer spacer then that fits the interpretation. The shorter extension leads to barrel distortion and that causes a displacement to the outside (because that distortion does not happen in air).

 

I‘m not trying to give a lecture on optics but to point out a simple test that will indicate in which direction one should make a change (displacement to the outside > longer, displacement to the inside > shorter).

If that is true, then it is quite likely that already the second configuration tested will be the best one available to a given person. 
 

Let‘s just leave it with this and see whether anyone finds it useful. That is the main point here. For me it already is, because I couldn‘t find the information for either the Samyang lens or the Athena port (lack of perseverance). But I now know that this will have some limitations that I can only solve at the expense of more vignetting, so I won‘t bother.

GAS cured, thanks to your help!
 

 

 

No a simple test will not indicate the direction where you should go.

in both cases extension too long or too short you will loose

more field of view however when you have pincushion distortion you have less loss of field of view 

as we are talking about few degrees you will need a series is tests to find out where you are

in addition if you have a distorted optic it will simply be impossible to determine as the lens is already distorted 

What you could do assuming you know your dome field of view is go as long as it goes before vignette snd go backwards however for a dome hemisphere this is not practical and at the end still painful because you need all the possible extension rings

in practical terms this method is good to fine tune your set up when you are already 90% there otherwise you need a macro slide and a test tank and this is not simple at all

[Klaus]

It happens right at the air-water-interface. The goal is to find the extension length that will lead to a minimal offset in how the line continues under water vs. in air. If you have no offset at all, then your lens renders exactly the same image above and below the water line. This happens only when the lens and the port are optimally arranged with respect to each other. 

I have probably sent people on the wrong track when I wrote about „stright lines“. Please excuse my lack of eloquence, I am not a native English speaker. Correctly stated, the observed effect here is the lateral offset of straight lines entering the warter. I have cropped the critical regions from the nice images provided by interceptor121 (thanks again) and annotated it accordingly - the test is essentially like aligning an inkjet print head (and this comes from the publication I cited, not my imagination).

null

The optical laws have been derived through simple experiments and vice-versa we can verify optical geometry with such tests. Of course everyone is welcome to suggest additional approaches to complement and/or replace it, I am not at all saying that this is a mandatory exercise. I also think that having the port in the ideal position by no means implies that it’s the right port for the lens in question. It may be possible to derive this from the rest of the images, but I have no idea how and that is explicitely not the topic of this post. 

I have said what I wanted (and by now more than once), so I will stop posting in this thread.

Cheers,

Klaus

 

 

 

[Guest]

1 hour ago, Klaus said:

It happens right at the air-water-interface. The goal is to find the extension length that will lead to a minimal offset in how the line continues under water vs. in air. If you have no offset at all, then your lens renders exactly the same image above and below the water line. This happens only when the lens and the port are optimally arranged with respect to each other. 

I have probably sent people on the wrong track when I wrote about „stright lines“. Please excuse my lack of eloquence, I am not a native English speaker. Correctly stated, the observed effect here is the lateral offset of straight lines entering the warter. I have cropped the critical regions from the nice images provided by interceptor121 (thanks again) and annotated it accordingly - the test is essentially like aligning an inkjet print head (and this comes from the publication I cited, not my imagination).

null

The optical laws have been derived through simple experiments and vice-versa we can verify optical geometry with such tests. Of course everyone is welcome to suggest additional approaches to complement and/or replace it, I am not at all saying that this is a mandatory exercise. I also think that having the port in the ideal position by no means implies that it’s the right port for the lens in question. It may be possible to derive this from the rest of the images, but I have no idea how and that is explicitely not the topic of this post. 

I have said what I wanted (and by now more than once), so I will stop posting in this thread.

Cheers,

Klaus

 

 

 

Sorry you simply do not get it

 

Any error will result into the part in water go wider in both cases longer or shorter. The line of the water is because it was not still there is nothing to be derived.

 

If you bought the article you linked you would see that they use a checkered chart so that they can observe the distortion

 

An incorrect position of the dome of 2 cm results in an error of 1-2 degrees this is too small to be measured without a proper test rig

 

If you want to know if the extension was too long or short you would need a pattern similar to those used for lens correction and this would only work with a perfectly corrected lens and a totally aligned rig

 

To recap

Fisheye lenses are distorted impossible to determine if you are long or short using this method, however normally pulling out until it vignettes is good enough assuming you have an hemishphere

 

Rectilinear wide angle lenses you need to run a calculation first or find the entrance pupil with that you are 90% on target use this test just to check you are ok and fine tune

 

If you wanted to use this method you would need a test thank with the dome only in water and a macro slider. Take photos in steps of 5mm and check alignment. Or use a test rig again super stable with a remote control and use a grid pattern to detect distortion assuming you have a perfect lens profile.

 

The problem is that as the water line is always larger and the part in water goes off on both sides you will not know from the first shot if you are too long or short unless you have horizontal and vertical pattern to check for distortion and your rig is perfectly aligned.

 

I used this method 5 years ago and it was frustating I have since developed a set of formulas when I don't have the lens parameters and  have a 90% hit rate normally I am off not more than 5mm

 

 

 

[Chris Ross]

8 hours ago, Interceptor121 said:

Sorry you simply do not get it

 

Any error will result into the part in water go wider in both cases longer or shorter. The line of the water is because it was not still there is nothing to be derived.

 

 

I found a preprint of the article here: 

https://www.geomar.de/fileadmin/personal/fb2/mg/kkoeser/domecalibration_preprint.pdf

 

And in figure 3 on page 5 they have a diagram showing the UW part getting wider/narrower as the position changes as was stated earlier in the article.  I haven't read through it all or looked at the optics involved, but it looks like you are saying the authors are wrong.  Can you say what their error is?

 

[Guest]

1 hour ago, Chris Ross said:

I found a preprint of the article here: 

https://www.geomar.de/fileadmin/personal/fb2/mg/kkoeser/domecalibration_preprint.pdf

 

And in figure 3 on page 5 they have a diagram showing the UW part getting wider/narrower as the position changes as was stated earlier in the article.  I haven't read through it all or looked at the optics involved, but it looks like you are saying the authors are wrong.  Can you say what their error is?

 

i did not imply anything other than the op hasn’t read the article carefully otherwise it would have understood that you need to put the ruler horizontally 
The rig in the experiment has the checkered pattern grid I mentioned and checks for distortion if you have read the article it doesn’t use rulers vertically the figure 3 is a simplified schematic the article doesn’t live up to

it in practical because the distortion coefficients are under a few degrees which is smaller than the degree of precision a rough method can handle

which is what they say at the end 

distortion will always result in a loss of field of view otherwise you would not need the dome correctly positioned 

to fine tune your rig you need a test rig that works at the level of millimetres once you hit a distance of one foot which is a typical ruler size the underwater part always looks bigger

if I have some time I will take a picture with my 14mm that has a too short extension later

 

[Chris Ross]

1 hour ago, Interceptor121 said:

i did not imply anything other than the op hasn’t read the article carefully otherwise it would have understood that you need to put the ruler horizontally 
The rig in the experiment has the checkered pattern grid I mentioned and checks for distortion if you have read the article it doesn’t use rulers vertically the figure 3 is a simplified schematic the article doesn’t live up to

it in practical because the distortion coefficients are under a few degrees which is smaller than the degree of precision a rough method can handle

which is what they say at the end 

distortion will always result in a loss of field of view otherwise you would not need the dome correctly positioned 

to fine tune your rig you need a test rig that works at the level of millimetres once you hit a distance of one foot which is a typical ruler size the underwater part always looks bigger

if I have some time I will take a picture with my 14mm that has a too short extension later

 

I'm not saying that you imply anything I'm simply seeking an explanation of the apparent contradiction.  They claim that distortion one way makes an object look bigger and the other way smaller and it seems you disagree with that premise.

[Guest]

1 hour ago, Chris Ross said:

I'm not saying that you imply anything I'm simply seeking an explanation of the apparent contradiction.  They claim that distortion one way makes an object look bigger and the other way smaller and it seems you disagree with that premise.

I totally agree with the fact that distortion shows where to go however I disagree that the difference is as apparent as figure 3 makes it looks it is not

When you have distortion if you look at the absolute centre of the image you do not see it the centre is always straight so looking at the split line is not going to tell you much that is why I drew the lines

What you need to do is to use a measuring chart and check the part underwater in fact you don't even need to shoot a split you can just shoot a grid and measure it, obviously to do that you need to entirely submerge the grid but this is actually easier than taking a split

The challenge of this method is that you need a perfect lens to start with, so one where a lens profile is available and project totally straight lines (Like adobe one) 

I therefore disagree that this is an easy method, it is not and I know because I have tried it years ago there are several challenges

1. You need to have a bunch of extention ring

2. You need a stable test rig

3. You need to be able to measure distortion

 

There is also another practical issue distorted optics are as the word says distorted so there is no lens profile and is a bit difficult to judge what is going on when you no longer have straight lines

In practical terms it is more effective and efficient to model the lens and then do this test at the very very end as a proof of concept, it is not effective not simple, not quick and not easy to start without any idea of what is going on because it is much more complicated and difficult to measure than figure 3 implies

[Klaus]

One more thing nonetheless…

Why bother?

That’s arguably the most important question. If the port position is off the ideal location, then the port will cause diffraction as seen by the lens (because the light that is captured does not traverse the port at a 90° angle). Unless you are doing photogrammetry, you might not even notice the resulting distortion - and if you do, it may be straightforward to correct in Lightroom. However, the diffraction does not affect all colors to the same extent, hence it leads to color fringes (=chromatic aberration). Seeking the best possible port position i.e. extension ring size will thus reduce fringing.

[Of note: Light that hits the port „straight on“ will always traverse at a right angle, hence the center will always look good (and not be distorted). Therefore fringing increases off-center and is worst towards the corners of the image.]

Personally, I don’t care about the distortion (except maybe for splits in very calm water) but fringing can be distracting when it’s really bad. The good thing is that we anyways use a small f-stop to deal with the curved virtual image and that also helps a bit with the fringes. But a well-placed port may get you the same result perhaps one stop wider? I don’t think that it is necessary to achieve „ideal“ positioning and the limited set of extensions provided by the manufacturers only allows for a coarse-grained adjustment. For example, I am using an MFT Pen port and AOI only makes two extension rings for that (14 and 24 mm), so only about 3 reasonable combinations (14, 24 and 38 mm) are available. In addition, the position of the lens‘ entrance pupil may move when focusing. Thus, don’t let „perfect“ be the enemy of „good enough“! Whenever possible, the port charts or your own calculations will tell you most of what you need to know given the limited options you have.

 

Then why did I bother?

Because I could not find any information on the Samyang/Rokinon 7.5 mm fisheye lens with the Athena port under water. I like to buy second hand gear and could snap up the Athena port for a very reasonable price, plus the (manual focus) Samyang 7.5 mm fisheye sells for really cheap – less than 1/3 of the Panny 8 mm even second hand. AFAIK, the Athena is not as „mainstream“ as the Zen or Precision ports, and the Samyang lens is cherished by astro-photographers but nobody seems to use it below surface. Thus, no possibility to do any calculations beforehand. Once everything had arrived, I wanted to know whether this is totally off or worth a try under water.

[I will post some images elsewhere here once I get a chance to go diving with it.]

If you also want to try some funky lens that is off the beaten path, or adapt a well-known lens via Metabones to a different system, you may also be looking for a relatively easy way to check which combination you‘ll take below the surface for your first trial. The method I found in the paper is an easy way to compare the 2-3 combinations you have available and select the best one.

 

A couple of misconceptions need to be corrected:

1)      Just like the port itself, any distortion caused by it is radially symmetric. It therfore absolutely does not matter whether you use vertical or horizontal objects for a test. But to reveal this distortion you need to cross the air-water line. Due to gravity on earth, I find this significantly easier with a vertical object. Just don’t place it dead-center, and try to shoot as much as possible the same perspective with the combinations you want to compare. If you don’t believe this, just turn the camera by 90° and take another shot – it will look exactly the same. (Except if you rigged it up to the cold shoe as I sugested, because then your object will also be rotated and it no longer crosses the water-line.)

 

2) It is a no-brainer that a fisheye will remain a fisheye when housed behind a dome port. I can't imagine anyone here would try to "de-fish" by deliberately mis-positioning the port. This would not work.

 

3) I picked this up from a publication that dealt with photogrammetry for deep-sea rovers. They indeed used checkered test charts, but that is because they also inserted them slanted. The goal was to develop an algorithm that corrects for distortion even after the best possible port positioning (e.g. due to the port not being of ideal spherical shape) because that allows for even higher precision. Calibrating distances, especially for computer vision, is of course much easier with checkered patterns than simple lines. But unless you care for highest positional accuracy in your image processing, you’ll be perfectly OK with a simple vertical thing (straw, ruler, thermometer, strobe arm…) that’s placed at about 2/3rd from the center of your image. Interceptor121 provided a really good proof-of-concept image above.

 

Of course, checkered test charts also work, but they are not easy to find when a workshop member invites you to try a new lens while you're on a liveaboard. If you’re lucky, however, Interceptor121 is also there. After you narrowed down the port position for your system experimentally, he can use his math to back-calculate the entrance pupil position (-range) of that particular lens and afterwards forward-calculate the best available scenario for the systems of all other workshop participants. That’s where the combination of theory and experiment really shines!