Machine learning and natural language processing - The Computing Universe: A Journey through a Revolution - page 288

Fig. 14.3. In the last decade, progress in
object recognition in natural images has
been achieved with machine-learning
algorithms and very large labeled
training sets.

Fig. 14.4. The diagram illustrates the
operation of a Kinect 3D Camera. An
infrared laser illuminates the scene
with a random dot pattern. By using the
images of these dots, the camera sensor
can determine the relative distance of
objects in the scene.

Object at
depth d 2

Object at
depth d 1

y

z

x

Imaging
plane

Optic centre
of IR laser

Optic centre
of camera

Baseline

framework to solve the problem of face detection. Their system is now widely
used in the face detection software in digital cameras.
In 2008, the team working on the Microsoft Xbox game console met with
vision researchers at Microsoft Research's laboratory in Cambridge, England.
The Xbox team's ambitious goal was to develop human body-tracking software
that was powerful enough to be used for playing computer games without using
a game controller. Alex Kipman, from the Xbox team, had taken a new approach
to the problem of three-dimensional motion capture by using depth informa-
tion from an infrared three-dimensional camera. The infrared camera worked
at a resolution of 320 × 240 pixels and generated images at thirty frames per
second ( Figs. 14.4 and 14.5 ). Cambridge researcher Jamie Shotton wrote:

The depth accuracy really got me excited - you could even make out the nose
and eyes on your face. Having depth information really helps for human pose