Here is an example of a theatrical set piece I will be modeling in Image Modeler.  I had talked earlier about projecting dots onto the surface, but haven't tried that yet.
 
  I am a little vague about some details inPoint Cloud method.  I suspect this method will be the predominate method I use since nothing we build actually is based on a primitve shape. Even a house that looks square will have a false perspective built in, making it not a square.  I think the only other method I can use is to draw each face individually.
 
  For the point-cloud method, Should I locate each point I want to make part of the mesh on each view?  I am not certain what is automatic about it if I must do this.  Can someone summarize the process of using the point-cloud feature?
 
  Note about the photographs.  They were taken using a high resolution digital camera.  Since the files sizes were so large, I increased the JPEG compression to make them more reasonable.

On the first picture, I couldn't figgure out how to put more than one photo on. Now, I see that it should be possible, but when I do so the "Posted Attachements" button doesn't seem to do anything.  Anyway, it is possible that the 4 additional pictures were added.  If not, I will just omit them for now.  Tell me if you want to see the other photos.

Hi,
 
  A proposition for additionnal helpers on your scene for calibration (see attached image).  
Place calibration helpers on the stone wall, but also around your object (flat helpers(paper) on the ground, higher helpers(sticks) ...). Try to keep them visible in as many shots as possible.
The best position for a helper, is a position seen from every shots or almost all shots. This is not always possible... If allowed or possible, photographs taken from above are nice to get all helpers visible.
 
 
For modelisation, If you do not need an "exact" mesh of your subject, I think there is enough "features" in the texture of the stones to get many modeling points.
 
Stef

I continued to make addtional efforts towards digitizing this model, however gave up on the existing photos.  While there seem to be a number of points that I can choose in multiple view, as I zoom in, it gets much harder to discern them.  In some situations, what seem to be points, are actually rounded edges which don't have a distinct corner.
 
  The other problem I have, is that I can't get more than a few of my calibration targets to turn green. I suspect this is a result of the high-resolution camera.  The targets I first tried were at least 10 pixels diameter in the photo. Manually finding the center of the dot was not a precise process.  I suspect that the small retroreflective dots will both show up very well in all views and provide a smaller object to click into the center of.
 
  The dots on the ground are actually numbers painted onto the floor.  I though these would serve as calibration targets, but from the angle I took the picture they are not readable, and I cannot discern which dot is which.  We will re-take the photos, and add calibration targets (retroreflective dots which will show up brightly from the flash).  I will also add them to important places on the object.
 
   For the contour of irregular surfaces, I am considering making a retro-reflective rope. I will get a length of rope and a few different colors of retro-reflective tape. Then I will wrap a piece of the tape around the rope every 6".  Bylaying this rope over the surface, I will create a line of dots every 6" over the surface.
 
Joe Dunfee

I tried to make a model from your photos, but it wasn't easy because they had been compressed so much that most detail was lost when zooming in. I did manage to calibrate the 3 frontal photos however, and made a very crude model starting with a cube and subdividing and splitting it while snapping to locators set on the stones etc. But the photos were not really good enough.
 
In the background of your photos you have the ultimate tool for taking such photos, that yellow "scissor lift", platform or whatever it's called. My experience with photogrammetry programs (Photomodeler, Canoma and ImageModeler) is that if at all possible you should take your photos from above, which should be very easy with that platform. It is much easier to get it right if the objects don't cross the horizon/eye level, iow that you take the photos from above (or below). Then you also get a good view of 3 sides instead of only 2.
Another thing is that it is not a good idea to clean up the area before taking the photos. It is usually a very good help having markers or other items spread around on the floor, and also up in the air if possible. For algining the world, which IMO is very important to be able to model efficiently, it is usually best to use locators on the floor. You should also put a long broom stick or something with a known length on the floor to be able to get it correctly sized. Placing that stick so that it aligns with the x or y axis (with z up) is also a good idea IMO.
 
I don't know what you're going to use those models for, but you mentioned something about passing them on to the CAD department?
Modelling such a model with lots of precise details will take quite a long time. But to make simple textured models shouldn't take more than a couple of hours I think.

Thank you all for the replies.  There are actually several reasons to create a CAD model of this set piece.  Our lighting designer is using a simulation package which can be used to light a virtual model of the stage.  This means he doesn't need to take as much time in the theater itself. He actually needs a low-polygon version of the model, because a lot of detail in the design slows down the rendering.
 
  But, I use CAD myself to design the structual aspects of the sets.  We are now building for a new show which will use some existing sets.  The internal rigid structure of older sets my be known from old drawings, but the rock surfaces are created in the shop with only a very rough idea of the overall dimensions.  For me to build a new set piece that must attach to an existing one, it is VERY helpful to have the surface in the CAD program.  For my use, the more detail the better.  
 
  I really was attracted to the automatic dot recognision that Image Modeler had and could use with a projected grid of dots.  I also liked its ability to do nurbs. But, since our media dept was familiar with RealViz product, their preference prevaled.   Even if I had these features, it wouldn't necessarily make the process easier.  The projected grid wouldn't help find the corners of surfaces, because the projected dots are unlikely to fall on a corner.
 
  I suspect the ideal solution would be one of the laser scanners coupled with a good surface creation program.  But, these products are still extremely expensive (they shouldn't anymore - The Disto laser distance meter costs under $800 nowdays) .  But the biggest issue with these scanners is the large point cloud they generate.  It software to process them into reasonable meshes has a long way to go, and the better versions are also quite expensive.
 
  I recall a few years ago seeing a product which projected a fine grid of lines from a normal slide projector.  Then, in a process similar to Image modeler, the software would use pictures to extract the 3d-shape.  The REALLY big benefit was that this process was automatically done.  They now sell the technology at a much  higher price by bundling it wth a large mecanical system.
 
  I even came across a university professor that wrote a program that would view the image of an object on a cheap-USB camera.  The video showed the shadow of a stick passing over the object.... and that was all the hardware there was; a $20 camera, a desk lamp and a stick.  They sold the technology to a company which is not releasing a cheap digitizing solution, but seems to be incorporating it into very expensive systems.
 
  However, I suspect that the pricing of all this stuff is going to drop very significantly at some point.  There is just no technical justification for $40,OOO machines which can be replicated using $20 of hardware along with some clever software.  
 
  I can see ImageModler incorporating the totally automatic mesh creation from projected lines as part of a future release, and blowing the systems costing 10's of thousands out of the water.
 
Joe Dunfee

It would be very interesting to see some photos of that rock wall with colored vertical light stripes projected on them, and to try making a 3D mesh from them. It looks like you already have all the necessary equipment right there, with powerful projectors where you could probably easily insert a transparency with "white" lines.
I think it would be pretty easy to manually make a model from in IM. It would probably also be possible to use the point cloud tool in IM on them.
PhotoModeler has several options when making point cloud meshes, like 2 1/2D and 3D IIRC. In essence I think such projected lines or points is most useful when working with "relief" type of objects, which are basically one-sided, or where there is a front and a back side with little happening on the sides.
 
BTW, here's my 1-hour mesh from your photos if you would like to have a look at it.

Bjorkn, thank you for your efforts on the model, but I can't view them right now becuase I have uninstalled the program so I can give it to the person in the media department... hmmm just realized that I can re-install but not register so that it will run in demo mode.
 
As for projected lines, I don't know how useful they will be in Image modeler.  The reason is the lack of nurbs.  In Photomodeler, you can lay down a nurb line in two views, and it will extract the 3d curve, without the user having to choose the intermediary points in a consistant manner.  These 3d nurbs can then be used for lofting.  Note that dispite the regular comments about Photomodeler having additional features, it should not necessarily be construed as saying that PM is better than IM.  IM certainly has some nice things about it which are geared towards the video industry. Whereas PM is definitely geared towards forensics.
 
Still, both products require a great deal of manual effort for all projects, and I suspect that is where the next level of improvements will be.
 
Joe Dunfee

I agree that NRUBS are nice, ut they are not really very suitable for such irregularly shaped objects as your rock wall. The NURBS in PM5 allows you to add curves following the shape of a line projected onto a surface, but it has to be drawn on at least 2 photos, and they have to share one referenced end point. The other end can be lose, which means there's a lot of problems if the curves doesn't stop at the same place. It's also often difficult to find a common starting point. The NURBS tutorials in PM5 shows curves drawn on a paper, and lines drawn on a lamp with a bulb (also with lines drawn on it). The bulb contour line is revolved around a center cylinder, and a lamp shade is later added outside that. But having to have those lines there does limit its usefulness IMO. Doing any kind of detailed intricate work on it seems to be out of the question. My own experiments with NURB in PM5 has not been very successful at all.
The main difference between IM and PM is IMO that PM is a pure reconstructing program with a lot of focus on precision, whereas in IM you can not only reconstruct what is seen on the photos, but you can also model things into the scene which are not there yet. This can't be done in PM, because everything has to be referenced in at least 2 photos.  
This may not be a big deal because you can do the same thing in an external programs like Lightwave and Cinema4D, using cameras linked to their actual photo backgrounds and do the modelling on top of the exported model. But it sure is convenient :-)  
IM doesn't have the best GUI around (neither does PM BTW), and its modeller isn't the most flexible (why can't we flip the normal of a single face for instance?), but all in all I think I've chosen IM for my current projects which is (re)constructing canvas hoods on small boats.  Some of them already have a hood which needs to be replaced, while some have only the bent tubular frames, and some doesn't have anything yet.
 
If only I could find the ideal program to unfold those shapes afterwards, so that templates for plotting/cutting could be made...

I guess my efforts are focused on certain tasks, which I suspect the nurbs would help a lot on.  Specifically, the idea of a continous line down the center of the walking surface on my rock set would illustrate those surfaces very nicely.  I would then use an assumptio that my surface is flat to each side of that center line, and then extrude a surface to represent that.
 
  For your boat awning work, a projected line wouldn't help  much.  I imagine the tubular structure is one of those things that are pretty straightforward to digitize.
 
  As for unfolding, I've had an interest in that for other reasons.  Here are links to two programs that will unfold a polygon mesh surface.  Though, creating that mesh for unfolding can be quite a challenge for some objects.
 
Touch-3d - probably the most elaborate CAD program for this purpose.
http://www.algonet.se/~ludesign/
 
Pepakura is a $39 Japanese program geared towards paper folding, but it also seems to be pretty full features.  Creating the model must be done in another program, and again is where the real work is.
http://www.e-cardmodel.com/pepakura-en/
 
Joe

I'm still not convinced that NURBS will serve such a purpose better than polygons. Have you tried with the PM5 demo? Apparently it only works with the tutorial images, but I guess it should be possible to test NURBS.
I think that it would be just as easy to mark points on such a projected profile line. Using the helper lines will make it very easy and fast to mark the entire line, even will small steps between the markers. The automatching would possibly work as well, wlthough it seems to work only now and then, and only on high contrast areas like edges.
Then it's pretty fast to make a polygon and extrude it to both sides, or use other profile lines to connect them to. Then you can move the points around afterwards, subdividing it and splitting faces as needed, snapping to points or just modeling "freehand".
 
I agree that projected lines wouldn't be of much help on my boat awnings (is that what it is called? I thought it was hood). As they have a skeleton inside it's usually pretty easy to find places to mark. But the photos have to be taken from above.  
 
It's interesting that you mention those unfolding programs, because they are two of those I've been evaluating.  
Actually I bought Tenkai a couple of years ago, which is the predecessor of Pepakura. It worked very well, but unfortunately it does not maintain the size of the object, which means that I wouldhave to use still another app to try to get it back into "shape".
To try out Touch-3D I had to install a Mac emulator, but it still refused to load my dxf files, so I never got to really try it out.
However there is also TouchCAD, which comes for both Windows and Mac. It works very well, although you have to manually trace the polygons into curves and then loft them into surfaces which can be connected and split, and finally unfolded following various rules.  
So far this seems to be the best choice.
Amapi can also do the job BTW, but there's no seam overlaps etc.
And then there are some expensive industry programs that can do the job.  
The problem is that this is supposed to be a system that should be operated by people with as little training as possible.
The solution may be to write our own software. It's not exactly rocket science to unfold a simple polygon model...

Actually, I wasn't necessarily wanting to keep the nurbs in my final model. What attracted me was the PM method of tracing out the line in two views, and not having to choose exact same points for the intermediary vertices.  The workflow would just be faster. However, as long as I photograph with a method of marking everything in detail before I photograph, then it doesn't matter which method I use.
 
  You have mentioned the other unfolding programs that I have also looked at.  If you are considering writing your own, you might consider working with a gentleman who has written a simple unfolding program for AutoCAD (I think it also works with IntelliCAD).  He might be able to customize his program for you.  
 
His name is Jochen Schulz. http://www.black-cad.de
 
Joe Dunfee

In theory it sounds good not having to choose exactly the same points, and probably in some cases it is very useful.  
But in practice using the helper lines in IM (and PM) makes it very fast to mark the same point in two or more photos, or maybe even using the automatching feature.  
In the beginning I made the error trying to mark on the object before I had calibrated the cameras. Just forgetting the object in the begginning made the work much easier.  
We first started out putting a lot of small red stickers on the object - a lot of work. Now that we work with calibrated photos we don't need any stickers at all.  
We also made an "axis system" with 3 1m long arms with wooden balls at the ends and center. This didn't work too well as it was difficult to center on the balls when they were shaded in the light. The balls also were too big, because they were made for 1600x1200 images. The idea was to have a fixed system that could be used for "Control points" in PM.
Now we try to establish the x axis by using a telescopic stick at 1.5m or more that we put across the boat somewhere perpendicular to the keel. This works better, as we don't have to worry about the angles being off 90 degrees, and it's also much easier to carry around.  
 
Thanks for the tip on that German guy writing unfolding programs :-)

Thanks about the tip about the helper lines. I hadn't considerd that as an aid.  In the 2nd view, the helper line would cross my projected stripe at only one point.
 
>We also made an "axis system" with 3 1m long arms with wooden  
>balls ... This didn't work too well as it was difficult to center on the  
>balls when they were shaded in the light. The balls also were too big
 
   I was considering that exact same concept.  Again, I suspect that the retro-reflective markers can be very useful.  You can make them pretty small, because they will reflect the flash very brightly.
 
   I was also thinking that a 3-dimensional axis system would be cumbersome, so I was thinking that a 2-axis system would be a better choice.  This way, you could add right-angle constraint.  It should also be pretty easy to make it foldable.
 
  Regarding the unfolding.  I know some of the routines, particularly the one in Rhino, uses an "unrolling" method.  They will unroll a surface between two splines.  I think the splines can be 3d.  The method used by the Black-CAD program is simlar, but it first segements the line into a series of straight line segments.
 
  Keep me updated on your unfolding efforts!