Upload
frederic-kaplan
View
1.022
Download
0
Embed Size (px)
Citation preview
1
Introduction to 3D digitization
technologies
Roberto Scopigno Visual Computing Lab.
CNR-ISTI Pisa, Italy
R. Scopigno, 3D Digitization - HW 1
Overview
o Digitization for visual presentation: 3D vs. enhanced 2D media
o 3D digitization technologies
2
2
Acquiring Visually Rich 3D Models Goal:
Build accurate digital models to clone the reality (shape + surface reflection properties)
Acquisition methodologies:
n Image-based Rendering o Panoramic images (2D) o RTI images (2D)
n Standard CAD modeling (manual process)
n Approaches based on Sampling o 3D scanning (active)
o 3D from images (passive)
3
Modelling vs. Sampling o Modelling
n Manual process [“redraw”]
n Accuracy is unknown
n 3D model is usually complete
o Sampling/scanning n Semi-automatic process
[“photography”] n Accuracy is known n 3D model is usually
uncomplete (many unsampled regions)
R. Scopigno, 3D Digitization - HW
3
R. Scopigno, 3D Digitization - HW 4
3D scanning devices Many different technologies, just two
examples:
o Laser or structured light, Triangulation n Small/medium scale artifacts (statues) n Small/medium workspace 20x20 ->
100x100 cm, distance from artifact ~1 m n High accuracy (>0.05 mm) n High sampling density (0.2 mm) n Fast (1 shot in ~1-2 sec)
o Laser, Time of flight n Large scale (architectures) n Wide workspace (many meters) n Medium accuracy (~4-10 mm) n Medium sampling density (10 mm) n Slow (1 shot in ~20 min)
R. Scopigno, 3D Digitization - HW 5
Active Optical Technologies
o Using light is much faster than using a physical probe
o Allows also scanning of soft or fragile objects which would be threatened by probing
o Three types of optical sensing: n Point, similar to a physical
probe o slow approach, lots of physical
movement by the sensor. n Stripe
o faster: a band of many points passes over the object at once
n Other patterns …
4
R. Scopigno, 3D Digitization - HW 6
Stripe-based scanning
R. Scopigno, 3D Digitization - HW 7
Optical Technologies - Triangulation
How do we compute the 3D coordinates of each sampled point?
o By triangulation, known:
n emitting point of the light source + direction (illuminant or emitter)
n the focus point of the acquisition camera (sensor)
n the center of the imaged reflection on the acquisition sensor plane ( P(a) ) Triangulation is an old, simple approach (Thales-Talete)
Issues: precision and price of the system
5
R. Scopigno, 3D Digitization - HW 8
Output: range map
R. Scopigno, 3D Digitization - HW 9
Triangulation-based systems
An inherent limitation of the triangulation approach: non-visible regions
o Some surface regions can be
visible to the emitter and not-visible to the receiver, and vice-versa
o In all these regions we miss sampled points è integration of multiple scans
6
R. Scopigno, 3D Digitization - HW 10
Scanning example
R. Scopigno, 3D Digitization - HW 11
Acquisition accuracy o Depends on sweeping
approach … o … on surface curvature
w.r.t. light direction …
o Laser syst.: the reflected intensity can be used as an estimate of the accuracy of the measure
7
R. Scopigno, 3D Digitization - HW 12
Acquisition accuracy
o … on the surface shape nearby the sampled point
o … and on surface reflectance [see Curless Levoy “…Space Time Analysis”, ’95]
R. Scopigno, 3D Digitization - HW 13
Optical Tech. – Time of Flight Measure the time a light impulse needs to travel from the emitter to the
target point (and back)
n Source: emits a light pulse and starts a nanosecond watch
n Sensor: detects the reflected light, stops the watch (roundtrip time)
n Distance = ½ time * lightspeed [e.g. 6.67 ns è 1 m ]
o Advantages: no triangulation, source and receiver can be on the same axis è smaller footprint (wide distance measures), no shadow effects
[Image by R. Lange et al, SPIE v.3823]
8
R. Scopigno, 3D Digitization - HW 14
Optical– Time of Flight
o Optical signal: n Pulsed light: easier to be detected, more complex to be
generated at high frequency (short pulses, fast rise and fall times)
n Modulated light (sine waves, intensity): phase difference between sent and received signal è distance (modulo wavelenght)
n A combination of the previous (pulsed sine)
o Scanning: n single spot measure n range map, by rotating mirrors
or motorized 2 DOF head
[Image by Brian Curless, Sig2000 CourseNotes]
R. Scopigno, 3D Digitization - HW 15
3D scanning – raw output data For the user, same type of output data :
n Range map: 2D map of sampled 3D points (640x480 -> 2M - 5M points)
n Can be managed as a point cloud or a triangulated surface chunk
9
R. Scopigno, 3D Digitization - HW 16
Why processing raw scanned data?
The acquisition of a single shot (range map) is only a single step in the 3D scanning process, since it returns a partial & incomplete representation
dal parziale al totale
We need algorithms and software tools for transforming redundandt sampled data into a complete and
optimal 3D model
3D Scanning Pipeline
R. Scopigno, 3D Digitization - HW 17
10
Note: New approaches appeared that use many redundant & overlapping images to produce results similar to those produced with active scanning devices
è 3D from images (passive methods)
R. Scopigno, 3D Digitization - HW 18
R. Scopigno, 3D Digitization - HW 19
Questions?
o Contact:
Visual Computing Lab. of ISTI - CNR
http://vcg.isti.cnr.it