1. 03/10/2013
1
3D from images, an overview
9th October 2013
0
Perspective and Stereo
 We do perceive the environment three-
dimensionality
 thanks to our stereo vision AND the
perspective
 projection that occurs in our eyes..
 Using the geometric laws of this two
phenomena
 we can build the 3D geometry of the scene

2. 03/10/2013
2
Perspective Camera
 All starts here... a camera is defined by
parameters:
 - Position
 - Orientation
 - Focal Length
 - Viewport
 Distortion
 Given a point in 3D space, it is know the
projected point on
 photo plane
 Given a point in the photo, it is possible
to trace back the
 ray which originated the pixel
2 paths...
 Two possible directions
 - Assisted Modeling
 - Automatic Stereo Matching
 The underlying principles are the same... but
they sprouted
 different kind of tools...
 I'll cover the first, Matteo will take over later on
 As usual, this distinction is becoming blurry...
 ...things does converge, after all

3. 03/10/2013
3
ASSISTED MODELING
Assisted Modeling

4. 03/10/2013
4
Sketchup
 A very strange modeling tool.
 Follows more the way a technical drawing is done on paper
 (reporting/referencing) than the usual 3D modeling metaphore.
 Easier for people with a technical drawing/sketching background.
 Easier for people with no experience in 3D modeling.
 Focused towards modeling of buildings
 and mechanical entities...
 Acquired by Google some years ago...
 distributed now as a semi-free tool
http://sketchup.google.com/
Sketchup – the Luni Temple
 An ideal tool for very regular buildings... like this
one
 (regular does not equal new)
http://sketchup.google.com/

5. 03/10/2013
5
Sketchup Photo Match
 Assited modeling from a SINGLE photo
 Calibration: axis and vanishing points markup
 Modeling: 3D drawing by axis/reference reporting
 Partial calibration, with only a single photo, only the axis
can be recovered.
 SketchUp can be used to model by reporting/referencing
Example 1 – Example 2
Photogrammetry
 Perspective & stereo
 Common reference points are marked on
 multiple images…
 From these correspondences it is
 possible to calculate camera
 position/parameters and 3D
 location of the marked
 points

6. 03/10/2013
6
ImageModeler
 Photogrammetry commercial tool
 Points are marked on input images, camera are fully calibrated
using these
 points, camera are calibrated and modeling can be done using
the
 recovered 3D points
 Acquired by Autodesk... now it costs
 three times the old price (with the same features)
PhotoModeler
 Photogrammetry commercial tool
 The tool for the professionals... Two steps: camera calibration
(with markers
 and references) and camera pose estimation. Modeling and
measuring with lots of different tools
 Very, very, very complex to use...

7. 03/10/2013
7
MULTI-VIEW STEREO
MATCHING
1 - Automatic matching of images
 The entire process is based on finding matches between
images.
 Record your pictures not too far apart, so the computer
can match them easily!

8. 03/10/2013
8
2 - Camera Calibration
 No prior knowledge about camera calibration is
available, so all information must be recovered from
the images
 It is therefore important that enough information is
present in the images!
 Important factors:
 Motion of the camera
 General structure of the scene
 Enough overlap (only points that are visible in at least 3
images are useful)
 What you want reconstruct and how you get the photos
have great influence on the final reconstruction (!!)
Calibrated Cameras: what you can do with them
Photo Tourism
 Having a set of (even etherogeneous) images, you can
navigate the photo collection in a “spatially coherent” way.
It evolved into PhotoSynth (see later).

9. 03/10/2013
9
3 - Dense Matching
 After recovery of the
camera calibration,
dense depth maps
are computed
 These contain the
depth of every pixel
and a quality
measure (how
confident we are of
each particular pixel)
Depth Map
Photosynth Toolkit
 The Photosynth toolkit is the result of the
work of people from a Blog (Visual
Experiments)
 The code of Bundler (SIFT+Camera
Calibration) has been made available recently
 In alternative, it’s possible to use the
Photosynth service
 Moreover, further code for the surface
reconstruction (CMVS -> PMVS)
 The Toolkit is available and works locally

10. 03/10/2013
10
Photosynth Toolkit
 Also the Photosynth toolkit can be integrated with
MeshLab, so that the final result can be improved!
SfMToolkit
 It’s the same toolkit as PhotosynthToolkit,
but it’s completely local
 http://www.visual-
experiments.com/demos/sfmtoolkit/
 Good: completely local, no upload, control on
parameters
 Bad: completely local, sometimes it crashes…

11. 03/10/2013
11
Python Photogrammetry Toolbox
 Developed by Arc-Team, open source and free, for Debian and
Win (32 and 64bit)
 http://www.arc-team.com/
 Good: completely local, interface, control on parameters, video
tutorial
 Bad: completely local, a bit tricky to install…
Autodesk 123Dcatch
 Very well engineered tool...
 Works on a remote server, like Arc3D
 Produces a complete, textured model
 http://www.123dapp.com/catch
 It is free (for now), and works very very well.
 It is fast, works on difficult datasets and the
results looks good. However, not really high
resolution, and there is less control over the
process. It is a good tool to start with...

12. 03/10/2013
12
Autodesk 123Dcatch
 3
PhotoScan
 Commercial, low cost tool: 59 € for educational license,
179€ standard license. (win, mac & linux)
 Fast, work on local machine, directly produce textured
model. Very robust and reliable... We have used it with
good results on many diverse datasets.
 They also have a free tool to process images taken with
stereo (two lenses) cameras....

13. 03/10/2013
13
PhotoModeler - dense
 Photogrammetry commercial tool, 2nd version
 The tool is the same as before... after doing the camera
calibration, instead
 of using only user-picked points, the system does a dense-
matching.
 The result is quite similar to a range map... you need to do
standard
 processing in order to obtain a 3D model...
Model creation with MeshLab
After the 3D model is exported in MeshLab, there’s
a procedure to enhance it a bit and have a final
result:
 Cleaning
 Sampling
 Poisson reconstruction
 Poisson Model cleaning
 Vertex Color Transfer
 Cleaning
 Scaling
 Saving!

14. 03/10/2013
14
Conclusion
The methods for the acquisition of 3D content from
images have become an interesting alternative to
3D scanning.
Probably they won’t take the place of 3D Scanning,
but they could be a valid solution in a number of
practical and professional applications.
Nowadays, they are already used in archeologists
everyday work, and they are undergoing a sort of
standardization process.
Knowing these yechnologies is key to be able to deal
with 3D and CH.
Wanna know more?
Check the webpage of my University course:
http://vcg.isti.cnr.it/~dellepiane/Corso.html
(page in italian, slides mostly in English)
You’ll find slides, datasets, links etc etc
Do you want to learn how to use MeshLab?
Start from Mister P. Video tutorials
http://www.youtube.com/user/MrPMeshLabTutorials
Several playlists available