Graphics Reference
In-Depth Information
Sensor cost
ToF
1,0
0,8
Spatial/temporal
resolution
Face details
(wrinkles, etc.)
Multiview
0,6
0,4
Photometric stereo
0,2
SL-camera
Accuracy
measurement
Intrusivity/
User cooperation
SL-consumer camera
Laser
Illumination changes
Motion speed
Figure 1.17
Evaluation of different 3D face modeling techniques according to
Extrinsic factors
(Motion
speed/illumination changes/intrusivity and need for user cooperation/face details) and Intrinsic factors
(spatial and temporal resolutions/accuracy measurement/sensor cost)
work in office environment lighting conditions. Most of the facial details (wrinkles, scars,
and other person-specific markers) are reproduced in the virtual model.
Structured-light (SL) techniques provided an attractive alternative to the expensive laser-
stripe scanning technique. In fact, projected light(s) intend(s) to replace the laser scan. The
ARTEC MHT 3D scanner
4
is one commercial system which projects a permanent light
pattern and produces 3D video of a rate around 15 fps. Frame resolution is about 500,000
points. The working distance should be in the interval of 0.4-1 m. Depth-measurement
accuracy is comparable to laser scanners and is about 0.1 mm. Texture channel is also
captured but only when needed. The sensor is cheaper than the laser digitizers.
Structured-light (SL) consumer depth cameras (which are much cheeper) as MS Kinect
5
and
Asus Xtion Pro Live
6
have been recently developed and have been an attractive alternative
for expensive sensors. Kinect is based on the permanent projection of one infrared-laser
pattern. It was primarily designed for natural interaction in a computer game environment.
In fact, the sensor is less intrusive and only a near-frontal position of the user is needed.
However, the characteristics of the data captured by Kinect have attracted the attention of
researchers in the field of computer vision and computer graphics. The camera provides
depth and texture video with 300,000 points in every frame. The 2D and 3D videos have
got a rate of 30 fps. The depth measurement produced by the Kinect was not so accurate,
which means that it achieves a coarse reconstruction of the 3D face.
4
http://www.artec3d.com/3d scanners/artec-mht
5
http://www.xbox.com/fr-fr/kinect
6
http://www.asus.com/Multimedia/Motion Sensor/Xtion PRO LIVE/
