1 3D Scanning Pipeline Roberto Scopigno, Matteo Dellepiane Visual Computing Lab. CNR-ISTI Pisa, Italy R. Scopigno, 3D Digitization - HW 1 Overview Let us present the processing phases and algorithms required to transform a set of redundant & partial sampled dataset (range maps) into a complete, optimized 3D model
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Let us present the processing phases and algorithms required to transform
o a set of redundant & partial sampled dataset (range maps)
into
o a complete, optimized 3D model
2
Planning
Acquisition
Editing
Merging
Simplification
Texturing
Registration
3D Scanning Pipeline
MeshLab The stages of the 3D scanning pipeline are demonstrated with o MeshLab, an open-‐source tool,
developed by CNR-‐ISTI o More than 300K downloads in 2013
o Video tutorials are a very effec<ve documenta<on and training resource: n Delivered via YouTube:
http://www.youtube.com/user/MrPMeshLabTutorials
R. Scopigno, 3D Digitization - HW 3
3
R. Scopigno, 3D Digitization - HW 4
Acquisition Planning
o Selecting the set of views is not easy
o Very hard to scan all the surface
o An example: Scanning the Minerva n Bronze statue, Archeological Museum
Florence (under restoration), 155 cm n 4 acquisitions with different scanners
(2000-2002) n Last scan: Minolta laser scanner
(03/2002) o No. range scans: 297 o Sampling resolution: ~0.3 mm o Scanning time: 1,5 days
R. Scopigno, 3D Digitization - HW 5
Range map – Registration [1]
o Independent scans are defined in coordinate spaces which depend on the spatial locations of the scanning unit and the object at acquisition time
o They have to be registered
(roto-translation) to lie in the same space
o Standard approach: 1. initial manual placement 2. Iterative Closest Point
(ICP) [Besl92,CheMed92]
MeshAlign 1.0 (C) Visual Computing Group
4
R. Scopigno, 3D Digitization - HW 6
Pairwise Registration [2]
Initial registration with user intervention:
Mode 1) The user manually places a range map over another (interactive manipulation)
Mode 2) Selection of multiple pairs of matching points
ICP
R. Scopigno, 3D Digitization - HW 7
Automatic registration o Many people are searching new automatic
approaches to range maps registration
o Our approach works on series of consecutive acquisitions (circular or raster scanning order, overlap existing between rmi and rmi+1)
n Results on a complex X-Y scan of a bas-relief: 163 range maps aligned in 1 h 50 min (unattended)
…
5
R. Scopigno, 3D Digitization - HW 8
Merging Range maps o Producing in output a point cloud is
not acceptable Cons: visualization, data processing, …
o Surface reconstruction:
all [aligned] range maps are fused in a single triangulated surface (no redundancy, hopefully no holes)
o But consider that some holes are
unavoidable in 3D scanning is the object is complex
R. Scopigno, 3D Digitization - HW 9
Merging Range maps
Many methods/algorithms proposed:
o Old approach: build a patchwork
o New approaches: n Fuse the available samples (based on
distance field or interpolators) n Consider samples quality while fusing them
(to reduce noise and improve quality of the final mesh)
n Two merging modes: o Keep holes in the final model o Produce a water-tight model (no holes,
by interpolation)
6
R. Scopigno, 3D Digitization - HW 10
Optimization: Mesh Simplification
o 3D scanning tools produce huge meshes (from 5M faces up to Giga faces)
o Data simplification is a must for managing these data on common computers (PC, internet)
o Standard simplification approach: edge collapse with quadric-based error control (QEM) [GarHecSig97]
R. Scopigno, 3D Digitization - HW 11
Managing data complexity o Multiresolution encoding
can be build on top of simplification technology
o Goal: structure the date to allow to extract from the model (in real time) an optimal representation for the current view view-dependent models produced on the fly
o Note: the screen is limited (2M pixels), take this into account to reduce data representation complexity
CNR’s Nexus vcg.isti.cnr.it/nexus/ [“Batched Multi Triangulation”, P. Cignoni et al, IEEE Visualization 2005 + newer ideas]
7
View-dependent rendering
R. Scopigno, 3D Digitization - HW 12
• Mesh is denser in foreground
• Mesh is more and more coarse as we get farther from viewpoint
• Zones which are outside the view frustum are very coarse
Managing data complexity
R. Scopigno, 3D Digitization - HW 13
8
R. Scopigno, 3D Digitization - HW 14
3D scanning cost o Remarkable evolution since Digital Michelangelo times:
increased accuracy & speed, cost reduction
Minerva of Arezzo (1st) 150 range maps
1.5 months (2000) Angel, Duomo di Pisa
273 range maps, 7 days (2002)
Minerva of Arezzo (4th) 306 range maps, 5 days (2002)