Graphics Reference
In-Depth Information
After registering all the scans, the data is divided into sub-objects such as lamp-
posts, trees, and cars, which are polygonized and reduced in point density to fit the
standard requirements of visual effects data processing pipelines. Some skill and
practice is needed to know what data will be needed to accurately recreate each
object while discarding redundant information.
8.6
NOTES AND EXTENSIONS
In the days before laser range finding, themainway to acquire highly accurate 3Ddata
was by means of a coordinate measuring machine (CMM), a bulky system in which
a user pressed a pen-like probe against the surface of an object to record
(
X
,
Y
,
Z
)
locations.
Outside of visual effects, one of the most striking applications of 3D data acqui-
sition was Levoy et al.'s Digital Michelangelo Project [
275
]. Piecewise scans of
ten Michelangelo statues were painstakingly acquired using a custom laser stripe
scanner and registered into highly detailed and textured models for use in art
history and preservation. This project highlighted the many practical challenges
of scanning priceless objects on a tight timeline. LiDAR technology is frequently
used for cultural heritage applications in architecture (e.g., [
13
]) and archaeology
(e.g.,
http://cyark.org/
)
. In construction applications, LiDAR is important for qual-
ity assurance that an as-built building conforms to an original blueprint [
207
].
Finally, many of the autonomous vehicles in the recent DARPA Grand Chal-
lenges (
http://archive.darpa.mil/grandchallenge/
)
used LiDAR for real-time terrain
mapping.
Since the data in laser-stripe scanning is usually acquired as a temporal sequence
of stripes, keeping track of this sequence can help with filling in missing data or
fixing poor returns. That is, a human in the loop can fix or recreate a bad stripe by
interpolating the 3D contours acquired just before and after it. This approach was
taken for human body scanning in the
Star Wars
prequels and the early
Harry Potter
movies.
An early classic paper by Bouguet and Perona [
56
] described a clever structured-
light-inspired systeminwhich images of the shadowof a pencilmoving across objects
on a desk acted as the “stripe” for producing 3D measurements. They obtained sur-
prisingly good, sub-millimeter-accuracy results for small objects with this simple
technique. Fisher et al. [
143
] presented a similar idea using a special striped wand
that also had to be visible in the camera image. Boyer and Kak [
57
] were among the
first researchers to propose a one-shot, color-stripe-based structured light technique
using an empirically derived pattern of red, green, blue, and white stripes. They used
a region-growing approach to expand the list of identified stripes froma set of reliable
seeds.
Salvi et al. [
419
,
418
] gave excellent overviews of the state of the art in structured
light pattern design. They discussed several techniques not mentioned here, in par-
ticular the class of methods based on extending the idea of locally unique color
subsequences to two-dimensional patterns. For example, the scene can be projected
with a pattern of colored dots, such that each 3
3 neighborhood of dots does not
repeat anywhere in the pattern. Such patterns are called
pseudorandom
or
M-arrays
,
and a good example of their application was described by Morano et al. [
334
]. One
×
