Graphics Reference
In-Depth Information
Change detection methods are used to detect objects in the images. The background
is modelled based on a clustering approach. The velocity of the robot is adapted
according to its distance to a person in the workspace, and the path of the robot
can be updated continuously, where guided robot movements are also taken into
account. In addition to cameras, time-of-flight-based active range scanning sensors
are examined.
A similar approach for the three-dimensional detection of obstacles in the
workspace around an industrial robot is proposed by Henrich and Gecks ( 2008 ).
A difference image technique is the basis for a distinction between pixels belonging
to the static scene without unknown objects and pixels indicating the approximate
shape of unknown objects, where the classification is performed by a Bayes classi-
fier. The result is used by Henrich and Gecks ( 2008 ) to construct an image of the
current robot configuration and an image denoting the obstacle. Taking into account
future configurations of the robot, the system then performs a test to determine if
the robot may interfere with the obstacle. Regions possibly occluded by the robot
are marked as potentially belonging to an obstacle, which allows a refined collision
detection and a refined update of the reference image. The described system is also
able to determine a robot path which avoids collisions with obstacles in the scene
but nevertheless permits close encounters.
The image-based approach to the detection of obstacles in the neighbourhood of
an industrial robot is extended by Fischer and Henrich ( 2009 ) to a system consisting
of several colour cameras and range sensors observing the workspace, where the
data are processed in a distributed manner in on-board units associated with the
individual sensors. This system describes an object by its convex hull but does not
reconstruct parts of the human body.
For a safety system capable of monitoring an extended workspace around an
industrial robot or a machine and precisely predicting possible collisions with a hu-
man, a three-dimensional reconstruction of the scene is highly beneficial. As a con-
sequence, the camera-based SafetyEYE system has been created in cooperation be-
tween the Mercedes-Benz production facilities in Sindelfingen, Daimler Group Re-
search and Advanced Engineering in Ulm, and the company Pilz GmbH & Co. KG,
a manufacturer of safe automation systems (cf. Winkler 2006 , who provides a de-
tailed introduction to the SafetyEYE system, the cooperation in which it has been
developed and commercialised, application-oriented engineering aspects concern-
ing the integration of the system into automobile production facilities, and future
extensions). The stereo vision and camera calibration algorithms for the SafetyEYE
system have been developed in a research project led by Dr. Lars Krüger and the
author. The SafetyEYE system consists of a calibrated trinocular camera sensor,
which monitors the protection area around the machine (cf. Fig. 7.1 ), and two in-
dustrial PCs. Two different algorithms for stereo image analysis determine a three-
dimensional point cloud describing the structure of the scene being monitored. As
soon as a certain number of three-dimensional points is detected inside the protec-
tion area, the system initiates the protective measures necessary to prevent an acci-
dent, either by slowing down or by stopping the machine. The system is installed
by defining the three-dimensional virtual protection areas with a configuration soft-
ware. Setting up a traditional safety system consisting of several components such
Search WWH ::




Custom Search