3D processing and metadata ingestion at POLIMI, Presentation given by Gabriele Guidi, Sara Gonizzi Barsanti and Laura Loredana Micoli at the 3D ICONS workshop at the XVIII Borso Mediterranea del Turismo Archeologico conference in Paestrum.
The presentation describes the 3D digitisation carried out by Politecnico di Milano (POLIMI0 as part of the 3D ICONS project.
3D processing and metadata ingestion at POLIMI, Gabriele Guidi, Sara Gonizzi Barsanti and Laura Loredana Micoli
1. 3D Processing and metadata
inges1on at POLIMI
Gabriele Guidi*
Sara Gonizzi Barsan1
Laura Loredana Micoli
Politecnico di Milano ‐ Mechanical Engineering Dept.
2. Project
http://3dicons-project.eu
• 3 years EU-ICT pilot project
• Aim: supply Europeana with 3D items such as:
– Archaeological sites
– Architectures
– Monuments
– Artifacts
Including UNESCO World Heritage assets
3. Project numbers
• 16 Partners
• 11 Countries
• 3000 3D models + metadata
• 36 months (30 months acquisition phase)
• Project: 100 models/month (average)
• POLIMI unit: 537 3D models + metadata
• POLIMI unit: 18 items/month (average)
Massive digitization project, each production step !
has to be greatly optimized!!
4. Massive Digitization (Libraries)
Book
(physical object)
Bibliographic record
(descriptive metadata)
create/convert
2D scan
Digital object
storing
Repository
Metadata Record
(technical metadata)
Digital object url
5. Massive Digitization (3DICONS)
storing
Repository
(technical metadata)
CH Asset
(physical object)
3D capture
3D model
(digital object)
(descriptive metadata)
create/convert
Digital
object url
Metadata Record
6. Source of Heritage Assets for POLIMI
The Archaeological Museum in Milan
The architectural structure
Settled upon a complex stratification of
archeological ruins, tangible sign of the
ancient role of Milan as Capital of the
Western Roman Empire
The content
1000+ archaeological items including:
• epigraphs
• statues
• mosaics
• furniture
• potteries
related to Greek, Etruscan, Roman and
Medieval periods
7. Metadata creation
• Specific skill required
The complex ar1cula1on of data requires a high level of
exper1se in the archaeological/historical field
• Time consuming process
Collec1ng the informa1on required for arranging
suitable descrip1ve metadata might require more 1me
than allowed by the project dura1on
✗
Only Heritage Assets with pre-existing metadata
have been chosen: conversion instead of creation!
Descriptive metadata (85%)
3D acquisition and modeling (14%)
Technical metadata definition (1%)
8. Metadata conversion
• POLIMI source of metadata is SIRBeC
(Information System of Cultural Heritage of the
Lombardia Region)
• All records can be exported in xml format
• The SIRBeC data structure is compliant with
the CARARE metadata schema, used by
3DICONS as reference for structuring their
metadata
• Only metadata mapping has to be designed
10. 3D data collection
Image‐based
modelling
triangula1on‐based
systems TOF system
Small
texturized
objects
Small un‐texturized
objects
Buildings
(77%)
(14%)
(9%)
11. SFM
• SFM is therefore the most used technology in
this project
• but SFM is nearly a "black box" giving an
output with little of no way of intervention on
the final output
• the only controllable inputs are good quality
images
We need an optimized image acquisition protocol
in order to maximize the quality of 3D output!
12. Possible imaging problems
• Image blurring due to:
– Movement on shooting
– Wrong focusing
– Limited Depth of Field
• Lighting/dynamic range
– Backlights/mixed color temp.
– Light spots
– Highlights
• Confusing scene elements
– Painted walls/mosaics
– High contrast elements around the subject
14. Aperture tests on a small artifact
• Camera: Canon 5D mkII
• Sensor: Full frame
CMOS 21.1 Mpixel
• Lens: 50mm macro
• Manual focusing on the
left eye @ x10
• Camera-target distance:
22.5 cm
• Avg. GSD: 28μm
15 cm
7.5 cm
15. Aperture F 2.5
DOF @22.5 cm
2.4 mm
Focal Plane (FP)
FP + 23mm
20. SFM/matching at different apertures
Le^ Front Right
• Automatic identification of tie points
• Image orientation
• Dense color cloud image matching (high)
All processing was made
with AGISOFT Photoscan
21. Matchable points vs. aperture
F Tie
points
Dense cloud
size
2.5 200 420637
Focal zone: 22.4 cm – 22.6 cm
22. Matchable points vs. aperture
F Tie
points
Dense cloud
size
2.5 200 420637
5.6 408 1286552
Focal zone: 22.2 cm – 22.8 cm
23. Matchable points vs. aperture
F Tie
points
Dense cloud
size
2.5 200 420637
5.6 408 1286552
11 694 2827132
Focal zone: 22 cm – 23 cm
24. Matchable points vs. aperture
F Tie
points
Dense cloud
size
2.5 200 420637
5.6 408 1286552
11 694 2827132
22 848 3854555
Focal zone: 21.4 cm – 23.6 cm
25. Matchable points vs. aperture
F Tie
points
Dense cloud
size
2.5 200 420637
5.6 408 1286552
11 694 2827132
22 848 3854555
32 1033 4117599
Focal zone: 21.1 cm – 24.1 cm
30. HDR processing
• Enhances details in images containing both
overexposed and underexposed areas
• Allows therefore to increase the number of
points in image matching
• Since the SW manages jpegs only the full HDR
is tone mapped and converted in 24 bit RGB
• This allows to increase by the number of
matchable points in the darker areas
31. HDR Processing example
0.60s, f/32
Exposure OK
2.50s, f/32
+2 stop
0.15s, f/32
-2 stop
Tone mapped
HDR
• 4 groups of 3 shots have been taken from different orientations
• SFM with:
– Properly exposed shots
– Corresponding one mapped shots
32. Matchable points vs. processing
Processing Tie
points
Dense
cloud size
None 5467 5637516
HDR 5646 5858700
+4% matched points
37. A few examples - Archaeological Museum, Milan
Texturized mesh models from SFM
38. Good practices adopted
Issues
• Image blurring due to
– Movement on shooting
– Wrong focusing
– Limited Depth of Field
• Lighting/dynamic range
– Backlights/mixed color temp.
– Light spots
– Highlights
• Confusing scene elements
– Painted walls/mosaics
– High contrast elements
around the subject
Shooting/pre-processing solutions
– Tripod
– Manual focusing @ 10x
– Small apertures (16-32)
– Light shielding panels
– Mask post processing
– HDR/Polarizer filter
– Black/white background
• hides confusing elements
• speeds up masking
39. Conclusions
• The best practices of a massive
3D digitization project has been
shown
• Metadata conversion from
preexisting sources was needed
for quickly generating
searchable material
• SFM is a key technology for
shortening 3D digitization to a
sustainable level
• No many intervention is
possible on SFM, the only actual
action is improving image
quality
• Proper imaging protocols may
increase the 3D model quality
and the success rate, in possible
bad environmental conditions.
40. Thank you for your attention
gabriele.guidi@polimi.it