#CAMERA ACCURACY
19 messages in this thread
We are trying to resolve a discrepancy between scenes rendered from
within AutoCAD and 3D Studio. The goal is a precise alignment of a
rendered image generated from a Digital Terrain Model and a
Photograph. The issue has called into question the accuracy of the
Camera/FOV in 3D Studio. The problem manifests as an apparent shift
in the geometry, suggesting that the FOV relationship to the focal
length of the lens is not accurate. What I am looking for is some
way to prove the accuracy of the 3DS camera. It seems that there was
a discussion about this back in October. Anyone who has any
information on the subject of camera/fov accuracy please speak up.
At the present time we suspect our methodology in ACAD but would like
to subject 3D Studio to a proof first. Thanks for any suggestions or
help you may have.
As I recall, the field of view is correct, it is the lens size that is
off a bit to be compatible with AutoCAD.
>>3DS Camera/FOV…
You are correct in your observation of the focal length/FOV
un-relationship. We had some discussion on this earlier. A group I
do some work for was using some of my images to simulate a robotic
vision system. The optics guru ran into the same (small)
discrepency. However, with image processing calculations, "small" is
relative. <g>
The bottom line: don't use the Lens focal length value (50mm, etc) to
do scientific work. For historical reasons, the math behind the
focal length calculation was simplified to maintain consistency with
other Adesk software (as per Yost Group). (I guess at the time they
never thought a bunch of crazies would be trying to do science with
3DS!) The focal length value is fine for artistic merit and as a
good starting point for people familiar with 35mm cameras.
The FOV angle, however, proved to be mathematically right-on. You can calcuate
the appropriate focal length from the FOV.
JKJ (in focus this time)
PS: the robot vision system was an fantastic success – you CAN do great science
and engineering and (new military buzz-word:) Rapid Prototyping with 3DS.
John,
>>the math behind the focal length calculation was simplified to maintain
consistency with other Adesk software
My recollection of the conversation was that the calculations were
pretty accurate. All relative of course. I checked the R2 Camera
against data for various Nikon lenses and there was very little
inconsistency. Do you recall the formula they are using? The
difference in the FOV between ACAD and 3DS renderings amount to the
difference between a 60 mm and a 70 mm lens. That is more than a
little relative. It did occur to me that the Perspective function
might be affecting our results.
Thanks for mentioning the work you did on the robotic vision system. I
just did a robotics project over Christmas (no wrest for the wricked)
that was a concept presentation. It is amazing to see how much joy a
little 500 frame flic can bring an engineer. His presentation was a
major success so I expect that I will be doing more robotics. I am
interested to know more about how you used 3DS in a Rapid Prototyping
context.
Michael,
>>do you recall the exact formula for camera focal length and FOV…
Yes, but there's a little more to it. I took the liberty of zipping and
EMAILing the entire thread to you – it has Yost's explaination as well as a
description of what we found in the lab. There are also some comments as to
what I did with the 3DS model. Basically, I built a model of a part that
didn't exist and rendered a series of images which were used as the basis for
developing the image processing code used to control the robot. It saved a lot
of money by simulating the hardware instead of building it. The client (and
the project manager!) was well pleased, and a larger project resulted from the
initial efforts.
We are buying a recordable laser disk for the next phase of this project. I
plan to render a huge database of images representing every reasonable camera
position and store these images on laser disk. The vision computer will look
at an image and decide which way to move the robot arm. A second computer will
simulate the arm movement, pull a new image from the database, and send it to
the vision computer. This way, we can simulate the entire process, easily
adding environmental conditions and modifying the robot hardware configuration
and appearance as desired.
Check your email for the camera thread file…
JKJ
Thanks so much for saving and forwarding that thread. We will chew it
up and I will report the results here later.
JKJ,
Any possiblity of uploading the thread to library. I'd like a copy also.
Rick LAPIDUS
Rick,
>>upload camera thread to library…
I wondered about that, thinking however about two things: 1) that it would
primarily concern those doing scientific imaging with 3DS; and 2) the subject
could use some additional independent analysis.
To recap, (so no one will feel slighted <g>): for historical reasons, the
camera focal length reported in 3DS seems to be a very good approximation,
although not mathematically exact. It was constrained to match other previous
Adesk products. The FOV number, however, seems mathematically exact, and
should be used for scientific work where numerical precision is required. If
one is doing art and architecture, the focal length values are probably as good
as the 35mm cameras we carry around.
I've EMAILed the messages I have so far to you – let me know your thoughts.
I'll try to accumulate any additional comments people have and send them to
anyone who asks. I think it would be nice to get a few additional
confirmations on the mathematics, then write it up right (like the networking
"white paper") and upload it then. Whaddya think?
JKJ
PS: I wrote this a couple of days ago and forgot whether I sent is or not.
Must be the early brain damage. Sorry if I just sent it for the second time.
JKJ
I am teaching an intro class but see the potential in scientific
engineering apps. Thanks
Camera Accuracy?
I would appreciate any kind of information about 3D Studio's "camera".
I have been combining photography and 3D Studio/AutoCad generated
images for the past five years, and have noticed that camera
"properties" have changed between software versions, and that camera
"focal lengths" do not seem to correspond to real world formats.
I wish that 3DS's camera behaved more (and sometimes less) like a real
camera. I wish that we could switch between standard film formats
and that the lens properties would change accordingly. There is a
world of difference between the field of view of a 50mm lens on a
video camera and a 50 mm lens on a 35mm camera. But in 3D Studio,
changing the hieght and width of the output image changes only the
vertical field of view–the horizontal field of view always stays the
same.
I also wish that we could alter the properties of a camera–the way
that a person can with a view camera, to "correct" perspective, or
even force 2 point perspective. I have lost three prospective
(architectural) renderings because we could not frame the desired
view without destortions that a "real" illustrator or photographer
could easily remedy.
Donald,
<< Cameras, AutoCad and 3DS >>
John K Jordon has a wonderful thread for you. We plan on
uploading his saved thread for all (soon).
For now, John?
jonas[adesk]
I can't wait to see the uploaded thread. I know some film pre-visualization
people that would love to have information (and contacts) who can make the
camera do what they expect it to do.
n
Nancy,
<< see (polished version) of the camera accuracy thread >>
It's going to be a little bit before it gets uploaded. Currently
you can ask John K Jordon for a copy of everything that he's
collected (which is the bulk of the future polished version).
I plan on including any additional insights to the camera
focal length and field of view issue from an IPAS programming
perspective. There has been a recent thread, confirming JKJ's
(the infamous premier MMM) findings.
jonas[adesk]
Nancy:
Please take a look at the message in the branched thread same subject. We
concluded our camera accuracy tests and there is a report of the test and the
results.
Where can I order a copy of your book?
Michael E. Bartlett – AEtherium Visions.
Please take a look at the other branch of this thread. We completed
some tests on camera accuracy. I have posted the results, and a
brief description of the test. I hope it will be of some help. I
would also know where I can order a copy of your book?
In addition to the thread that John is uploading, you might want to
check out the Render Blowup command (page 7-189) before you lose any
more architectural clients.
Jack,
I haven't lost any clients yet, but time and projects slip away quickly when we
come up against hard limitations in our software.
Render blowup does offer some great new ways to frame views. However, it
really doesn't have much to do with optics or cameras. Render blowup is
basically a good way to "crop" a view.
The problem that we often run into with 3D Studio and architectural rendering
is that there is no satisfactory way (to my knowledge) to "correct" vertical
perspective in the manner of a perspective correcting lens or a constructed two
point perspective.
Using a longer "lens" from further away can help, but also introduces the kind
of wierd forshortening seen in telephoto shots of city streets… We get
verticals that are closer to being square with the frame, but we lose the sense
of space.
3D Studio is a virtual photography studio…but the "camera" is an upgraded
instamtic!
Yours,
Donald Newlands
Donald,
Render Blowup WILL do what you want. Try the following:
Place the camera in the required spot (your basic back against the wall
scenerio) and aim your target PERFECTLY LEVEL. Now increase your FOV until
the required portion of the scene is in view. This could take you to a
12-18mm lens but don't worry. Now, use Render Blowup to crop the desired area
of the scene – perfectly vertical lines and the fish-eye lens is causing very
little distortion at the extremeties. This is essentially how a 4×5 camera
"correction" works.
Since it is available in the Keyrframe, you are also able to produce a
perspectively corrected walkthrough – something impossible with a PC lens.
Now if we could only save the settings/crop-points of a Render Blowup for
subsequent re-use <g>. Hope that helped a little,
– Phil
Take a look at Phil's message.
– J