Thanks for the info. I'm not familiar enough with real cameras to have
known the distance information, even if it is a guess on your part, it's
better than I could have done.
I'll try the rendering with perspective and then matching against the
original instead of trying to line up my wheels in parallel view. The
problem with drawing a true circle over the image is that the wheel do
look round and the circles seem to line up... The anti-aliased edges
make it hard to discern where the true edge is as well.
I think I'm only off by a little, so I may just reduce my wheel-base a
little and call it good. Now if Realsoft would add a torus analytic I
could be 100% true to the MAGI model. :-)
Matthew
Neil Cooke wrote:
The correction system I first suggested does not really apply here ... it's
more used in architecture.
Fact is no correction of an image will create pixels missing due to
fore-shortening ... it simply elongates the area.
The Distortion we're seeing on this image is due to the "camera" being used
at wide angle ... probably about 28mm focal length 35mm film equivalent ...
or even less, maybe 17mm or so. However there would be very little
foreshortening since the camera is perpendicular to the viewing plane of the
cycle ... there is curvature from the extra wide angle used.
To determine what is going on perhaps model both wheels with guessed rim
configurations and render that, in perspective, with an effective 28mm lens
setting and see whether the distortion matches. You would then have data for
real dimensioning from this draft model. Personally I would ignore the
distortion and call the image flat absolute and re-model from there ... the
difference would not upset me. If I still felt doubtful I would guess a
small addition at each edge of the overall length.
Another clue in your image is the knowledge that the wheels are perfect
circles ... draw a perfect circle over the image at wheel size (probably do
this in a vector programme with the bitmap imported) ... if the
foreshortening is significant it will show and it would be simple enought to
stretch the image until the wheels were perfect circles as they must be in
ortho. At adjusted stretch the true length of the bike will be almost but
not entirely correct. Use that distortion factor of top views as well but
only in the same axis of course. And if this was done in a vector programme
then all the data you need would most likely be very readily available. Or
something.
To determine absolute dimensions some known dimension has to come into play.
Perhaps there is an image with a human figure of average height to work
from, perhaps assume Harley wheel and frame sizes and go from there.
That's all I can immediately think of. Hope it adds to the fun
Neil Cooke
----- Original Message -----
From: "Matthew Hagerty" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Wednesday, July 19, 2006 1:04 PM
Subject: Re: Compensating for perspective
Here is a link to the reference image I'm using:
http://digitalstratum.com/images/tron/lightcycle_all_sides.jpg
Note the front wheel. On the right you see some blue between the black
inner part of the wheel and the edge (I call this the rim), which is
missing by the time you get to the left side of the wheel due to the
perspective. Also, the center point of the rear wheel is actually in
line with the rear of the body, but it looks slightly set-in. The
bottom view also shows what looks like the rear wheel intersects the
body and does not fit in the wheel-well, but that's not really the case,
it does fit.
It is these kinds of errors that makes it particularly difficult to
determine several things:
1. Overall length of the lightcycle.
2. The true curvature of the canopy.
3. The size and placement of objects.
I was hoping there would be some way I can adjust for the perspective as
I move away from the center (where ever that happens to be...) I think
the perspective is worse from side-to-side than top-to-bottom, but it's
hard to tell.
Matthew
Neil Cooke wrote:
2D images can be reworked in Photopaint/shop to have no perspective
showing
in the verticals and with some images it is possible to go some way to
also
remove this in one horizontal direction. Maybe post us one of these
problem
images?
To remove perspective in all aspects of a comprehensive 2D picture ...
it's
probably best to start again and remodel in RS ... or live with it.
Maybe
minimise it with (non) lighting, blur it, etc.
Neil Cooke
----- Original Message -----
From: "Matthew Hagerty" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Wednesday, July 19, 2006 11:54 AM
Subject: Compensating for perspective
I have an image I'm using as a reference to accurately reproduce a
model
from (yes, still the lightcycle, but I'm very much an absolute
perfectionist.) The problem is that every known image was rendered
with
perspective, which may have been the only option in 1982. When trying
to size objects and place them properly, the perspective causes an
error
which makes it difficult to get things like the curves and angles
correct. Is there any way to compensate for the perspective in the
original 2D image?
Thanks,
Matthew
