Modeling from photographs (Photogrammetry)

[tvoght]

In Maury's thread on the Forest & Gallice, he mentions how he derived dimensions
for his model by scaling differences between points on photographs of the
prototype.

Rather than crowd that thread, I thought I might insert one here (for want of
a better place) on the subject of 'Photogrammetry'. It's a subject that may be
interesting to many of us here, and I'm hoping to spur a discussion.

I surely don't want to give the impression that I have any knowledge or understanding of the subject,

From Wikipedia:
"Photogrammetry is the practice of determining the geometric properties of
objects from photographic images."

Direct application of the technique requires understanding of some fairly
advanced physical and mathematical concepts, but there is commercial and free software available
that makes it more accessible.

Perhaps there are members who know more, or who have used available software
packages who may want to flesh this out, or comment.

Here are some links:

Wikipedia:
  https://en.wikipedia.org/wiki/Photogrammetry#Stereophotogrammetry

A site devoted to the subject:
  http://www.photogrammetry.com/

Direct Linear Transformation (another important search term):
  http://www.kwon3d.com/theory/dlt/dlt.html

  Free iPad software from Autodesk!
  http://www.123dapp.com/

Other commercial Software:
  http://www.photomodeler.com/index.html

--Tim

[Maryak]

Thanks for the heads up Tim,

I have just finished reading a couple of PDF's and watched some you tube video on this fascinating subject.

Suffice to say that as a point and shoot camera operator with a very low skill and knowledge level of photography and all those magical terms like focal length, depth of field, etc this is way above my head.

I made a CAd drawing of this



from this photograph



I did know the basic propeller geometry such as Diameter and Pitch and I traced the blade shape into CAD which by scaling gave me a very close projected outline of one blade and from there after a lot of help from forum members such as Ausdier and Jo a propeller was re-born.

I reckon that if you have a reasonable understanding of photography this would be a very useful addition to your modelling know how.


Best Regards
Bob

[IanR]

It's possible to do this with pencil, rulers, dividers, etc. and was done long before computers were invented. I've got a magazine article on it somewhere.

[tvoght]

Ian,
I'd like to see an article like the one you describe. Knowing the basic procedure using rule, dividers, etc. one could take a technological step-up by importing photos into a 2D CAD program and taking the measurements digitally in a drawing layer above the photos.

--Tim

[Zephyrin]

" I was trying to describe a method for getting scaled geometry without having absolute dimensions."

I read the sentence above in the Maury's thread on the Forest & Gallice, me too.
So, I don't know where to send my post now...

This problem reminds me a subject on which I worked hard (as a hobby!) few years ago.
Namely how to get the actual proportions of objects (the geometrical or 3D) in a single photograph or a perspective image, with projective geometry. 
It is necessary to start from a natural perspective view of objects having geometric shapes, with right angles and well defined elements situated in the depth dimension, to determine accurately the vanishing lines and points of the three directions of space.
With these three vanishing points, the principal point, the position of the point of view and the distance of the object to the picture plane and its orientation are determined. Then, all the necessary data to rebuild the object by simple geometry are now available (limited to its visible parts of course). Starting from here, the method follows the exact reverse of drawing in perspective.
In the example below, the exact geometric forms of the small house are deduced from the perspective drawing. I agree that this picture is rather confused, but my method works with building from old paintings and engravings. 
And here we are, it may works also with an old steam or IC engine as long as the object can be reduced to simple geometric shapes, and if the determination of the vanishing lines can be accurate (i.e. a very good picture).
In the common case where only one or two vanishing points are present in the picture, solutions can be found frequently to determine the depth and distance, and to proceed further with the reconstruction of the object geometry.

[AussieJimG]

Working backwards! That's a very clever idea Zephrin.

Jim

[Anzaniste]

It might be a bit simplistic but I have done it by printing a photograph on to squared paper and then working out a sale from a known dimension. My interest vintage cars. It's not difficult to find the tyre size which can be the base measurement to work from. Whilst not super precise it gives a pretty good place to start from.

[Tin Falcon]

Interesting something else to study in my "spare time"

when I was taking engineering classes years ago we had to create 3 D drawings from 2-D drawings . the opposite of what you are describing and bit mute with the advent of parametric drawing programs.
I also have seen some studies and techniques on 3-D trig calculations.
Tin

[AussieJimG]

MuellerNick started with photogrammetry of an engine and then cast the components. An amazing story: http://www.homemodelenginemachinist.com/f31/man-diesel-engine-1907-dm-2-100-a-19073/

Jim

[steamer]

Actually I hadn't seen this since my senior year in High School...I believe it was called "Descriptive Geometry" then...though I didn't get a chance then to work on an engine...

Thanks for posting this!

Dave

[Zephyrin]

Of course, I saw the topic about photogrammetry on HMEM and the awesome engine that resulted.
By photogrammetry is a different approach, it uses dedicated pictures made for this purpose, and the Reconstruction is more akin with stereo view, and the softwares are $$$$.
 but here, the aim is to use available images, photographs or engraving of old engines for instance, sometimes vanished, and to use just a drawing board, well with Turbocad I confess.
I have to revise a little bit, and I can develop the method.
true mathematicians in computer science have developed amazing results with old paintings.
look for
http://www.robots.ox.ac.uk/~vgg/projects/SingleView/models/sjerome/movies/sjerome.mpg
and google for "single view reconstruction".
several labs publish regularly on this topic, but I'm really out of touch...

[MuellerNick]

Here I am!
I have used Photomodeller. It costs about 1200$
It takes some learning to get reliable results. But after that, you can get resolutions of 1:5000. Means, if the largest dimension is 5 m, you are accurate to 1 mm.
With special cameras, you can get much better (and get really poor).


Every  lens projects the image a bit like a fishh's eye. So they need calibrating. That  does the software with a special pattern being photographed from different angles.
Also, the software can interpolate the center of more than 3 views of the same point. And it does sub-pixel resolution. Sometimes, a  pixel on the object is smaller thaan a pixel on the image sensor. And they interpolate that too. Impressive!


Still, it is a ***LOT** of work to reference all points (the version I had acces to made too much nonsense wwith auto-referencing, but that was years ago!). And then, shape the body in  CAD,  with a zillion of points. Not really fun!