A few examples of photogrammetry work in SketchUp using the PhotoScan plugin from tgi3D.
This is a 3D model made from a set of photos, all shot from
ground level, of a sculpture made of stainless steel, Volare by
Arnold Haukeland at Kristiansand airport.
It is an example of a type of object that could not easily be
created using any of the automatic tools around, like PhotoFly.
First problem is that the shiny steel surface would cause lots
of problems for auto-creation (AC). Same problem would
occur with transparent objects.
Second problem is the shape, with lots of concave surfaces,
which would also cause lots of problems for AC. It would
require at least 100 photos, including many shots from birds
eye view, to show enough details to generate a decent model.
Next problem is that there are lots of overlapping parts, which
also is a big problem for ACs.
Next problem is that ACs will have a hard time defining sharp
edges, as everything gets a bit blobby.
If you still managed to generate a point-cloud mesh it would
be huge, and need a lot of cleanup to be usable.
The model was made from 14 photos (could probably have used less)
in SketchUp, using the PhotoScan plugin from tgi3D.
I did not use any of the automatic functions of PhotoScan, but
modeled manually on top of the photos, using the brilliant view locking
tool in PhotoScan. You just draw a curve on one photo, viewlock it,
move to the next photo and drag the curve points so that it fits the
same curve, and the curve is located correctly in 3D space :) With
several curves/lines ready you just fill in the hole with a mesh and
smooth it. You could also use surface fitting, but then you may face
some of the problems related to shiny surfaces.
You could also grab the materials from the photos, but in this case I
just applied a generic steel material.
3D model on top of undistorted/calibrated photo
3D model overlaid on photo - slightly different viewing angles
3D model showing only the contours - sharp edges
3D model showing that it is a low-poly model with sharp edges
3D model on top of photo showing how well defined and lowpoly 3D objects
can be when modelled “by hand”. What would happen with the hand rails,
windows etc with AC? A complete mess?
Click images to view larger versions
You can probably tell that I’m not a fan of automatic “scanner-
type” 3D model generators that creates huge point clouds? ;)
As you can see on the head to the right PhotoScan can also
generate huge meshes that follows the surface closely. The
difference is that you can combine it with well defined edges,
and you can add lots of 3D details where needed and keep
larger flat areas very simple. Unless you’re going to make a
mold for casting replicas of the head to the right you’re usually
just as well served by using the lowpoly photo textured model
on the right side, rather than the fully detailed version as
shown in the red patch around the eye.
Although I love PhotoScan, a lot of modelling from photos can
be done in SketchUp using only the built-in PhotoMatch tool.
Another advantage of manual modelling is that it doesn’t
matter that much if there’s a tree in front of a house, some
clutter inside a room, or other unwanted objects in the photos
- you simply skip them when modelling, and edit them out
from the photo textures.
Bjørn K Nilssen
A few examples of photogrammetry work in SketchUp using the PhotoScan plugin from tgi3D.
This is a 3D model made from a set of photos, all shot from
ground level, of a sculpture made of stainless steel, Volare by
Arnold Haukeland at Kristiansand airport.
It is an example of a type of object that could not easily be
created using any of the automatic tools around, like PhotoFly.
First problem is that the shiny steel surface would cause lots
of problems for auto-creation (AC). Same problem would
occur with transparent objects.
Second problem is the shape, with lots of concave surfaces,
which would also cause lots of problems for AC. It would
require at least 100 photos, including many shots from birds
eye view, to show enough details to generate a decent model.
Next problem is that there are lots of overlapping parts, which
also is a big problem for ACs.
Next problem is that ACs will have a hard time defining sharp
edges, as everything gets a bit blobby.
If you still managed to generate a point-cloud mesh it would
be huge, and need a lot of cleanup to be usable.