Human motions are colorful, different motions often
represent different meanings. In many application cases, such
as behavior monitoring and motion analysis, human motions
hope to be presented comprehensively, but it is difficult in 2D
space. However, if we can recognize human postures in realtime
and display recognition results in 3D space, we can restore
human postures more accurately and vividly which is
convenient for people to observe and learn human motions. For
this objective, it is necessary to find a method to recognize
human postures in 3D space. In this particular paper, a novel
method is proposed for recognizing 3D human postures by
using Kinect sensor.
Many researchers have proposed different methods for
human postures recognition and reconstruction. Posture
probability density [1] has been used to reconstruct human
postures. This approach is very important at the early stage of
the posture modeling. However, it’s impractical utilizing
motion capture data for the daily-life posture modeling for the
entire human. Multi-camera system [2] has been used to
recognize 3D human postures. This method requires at least
three CCD cameras to work at the same time and needs a realtime
background subtraction method to extract human
silhouettes from color images accurately. Increasing the
number of camera is necessary to improve the accuracy of 3D
reconstruction which would aggravate the amount of
computation and increase the cost. The method [3], called
modeling of human postures using stereo camera, obtains
various human postures automatically by using a stereo camera
system. This method has proposed a technique for acquiring a
human posture by fitting a human skeleton model to a
recovered 3D human surface data. A stereo camera system
which can be mounted on a robot is employed for 3D recovery.
However, that presented human skeleton model only models
the upper part of a human body. Obviously, it isn’t enough only
to recognize human postures of the upper body in practical
application. Hence, a more compact device and a more
thoughtful method need to be considered to acquire entire
human postures.
Therefore, in this paper, we provide a novel method of
recognizing 3D human posture by using Kinect. Kinect [4, 5] is
a new game controller technology introduced by Microsoft in
November 2010. Since its launch date it manifest great
potential that it can’t be only used in computer gaming but also
many other applications like robotics and virtual reality. Kinect
(Fig. 1) includes a RGB camera, a depth sensor and a multiarray
microphone. It provides full-body 3D motion capture,
facial and gesture recognition. The depth sensor consists of an
infrared laser projector and a monochrome CMOS sensor.
Kinect is capable to capture color and depth image at the same
time. The resulting point cloud can be loaded to a computer
using open source libraries which enable Kinect to be operated
with Windows, Linux or Mac. Data connection to a computer
is through USB interface.