Graphics Reference
In-Depth Information
In photometric stereo approaches, only one image of the face is captured and used to recover
the depth information. The face is illuminated with three colored light sources from three
different directions. The capture can be made in real time enabling 3D and 4D acquisition.
By knowing the surface reflectance properties of the face, the local surface orientation at
points illuminated by all three light sources are computed. One of the important advantages
of the photometric stereo approaches is that the points do not need to be registered. Thus, this
category of approaches does not suffer from the correspondence problem, providing high
performance for featureless surfaces as the human skin. On the other hand, the disadvantages
of this category of approaches are that they are indirect and practical only for applications
in which the illumination is carefully controlled.
Multiview-based approaches capture images instantly and provide high resolution 3D and
4D textured images. In addition, they have several advantages over active approaches. First
is the quality texture image. The acquisition phase does not require pattern projection, and,
so, there is a true one-to-one correspondence with every color pixel and every 3D point. The
original texture images are always of the highest quality. Second is the absence of holes in
the final 3D scans. Dimensional Imaging 7 proposes systems designed specifically to capture
high definition 3D surface images of the human face with highly detailed 20-megapixel
texture maps using four 10-megapixel cameras and up to 32 cameras with a resolution of
up to 21 megapixels.
ToF cameras are relatively new devices, as the semiconductor processes have only recently
become fast enough for such devices. The systems cover ranges of a few meters up to about
60 m. Another advantage of ToF systems is the high rate capture. In return, they have a
low resolution and a precision of 1 mm to 1 cm. The Mesa Imaging 8 SwissRanger 4000
(SR4000) is probably the most well-known ToF camera. It has a range of 5-8 m, 176
×
144
6 ×
6 field of view and operates at up to 54 fps.The PMD
pixel resolution over 43
.
34
.
Technologies 9
CamCube 2.0 is a less popular one. It has a range of 7 m, 204
×
204 pixel
0 ×
0 field of view. It operates at 25 fps.
resolution with 40
.
40
.
Exercises
AB
A B =
AC
A C =
BC
B C =
h ABC
h A B C
1. From Figure 1.18 prove that
; retrieve the Z formula
given in Equation 1.8.
2. We need to study the 3D scanning prototype given in Figure 1.19. It consists of a laser
source that illuminates the object to be continuously scanned and two cameras that look
at the same object. The projected laser stripe is seen by both cameras. The global sensor
calculates the depth information, as illustrated in the figure. To capture the full geometry
of the object, a manual scan of the surface is required.
Compute the Z 1 -coordinate together with Z 2 -coordinate.
Explain the triangles considered to calculate Z 1 -coordinate.
Why this prototype involves two sensors, each of them capable of measuring the depth.
Suggest a depth value Z as a function of Z 1 and Z 2 ; explain your choice.
7 http://www.di3d.com
8 http://www.mesa-imaging.ch/
9 http://www.pmdtec.com/
 
Search WWH ::




Custom Search