terminology and then discuss motivations and limitations of sensor fusion. Sec-
tion 3 presents a survey on architectures for sensor fusion. Section 4 describes
algorithms and methods like the Kalman Filter, inference methods, sensor fu-
sion in robotic map-building, and the construction of reliable abstract sensors.
The paper is concluded in section 5.
2 Principles of Sensor Fusion
There is some confusion in the terminology for fusion systems. The terms
“sensor fusion”, “data fusion”, “information fusion”, “multi-sensor data fusion”,
and “multi-sensor integration” have been widely used in the technical literature
to refer to a variety of techniques, technologies, systems, and applications that
use data derived from multiple information sources. Fusion applications range
from real-time sensor fusion for the navigation of mobile robots to the off-line
fusion of human or technical strategic intelligence data [59].
Several attempts have been made to define and categorize fusion terms
and techniques. In [72], Wald proposes the term “data fusion” to be used
as the overall term for fusion. However, while the concept of data fusion is
easy to understand, its exact meaning varies from one scientist to another.
Wald uses “data fusion” for a formal framework that comprises means and
tools for the alliance of data originating from different sources. It aims at
obtaining information of superior quality; the exact definition of superior quality
depends on the application. The term “data fusion” is used in this meaning
by the Geoscience and Remote Sensing Society
1
, by the U. S. Department of
Defense [69], and in many papers regarding motion tracking, remote sensing,
and mobile robots. Unfortunately, the term has not always been used in the
same meaning during the last years [64]. In some fusion models, “data fusion”
is used to denote fusion of raw data [18].
There are classic books on fusion like “Multisensor Data Fusion” [74] by
Waltz and Llinas and Hall’s “Mathematical Techniques in Multisensor Data
Fusion” [34] that propose an extended term, “multisensor data fusion”. It
is defined there as the technology concerned with the combination of how to
combine data from multiple (and possible diverse) sensors in order to make
inferences about a physical event, activity, or situation [34, page ix]. However,
in both books, also the term “data fusion” is mentioned as being equal with
“multisensor data fusion” [34].
To avoid confusion on the meaning, Dasarathy decided to use the term
“information fusion” as the overall term for fusion of any kind of data [20]. The
term “information fusion” had not been used extensively before and thus had
no baggage of being associated with any single aspect of the fusion domain. The
fact that “information fusion” is also applicable in the context of data mining
and data base integration is not necessarily a negative one as the effective
meaning is unaltered: information fusion is an all-encompassing term covering
all aspects of the fusion field (except nuclear fusion or fusion in the music world).
1
http://www.dfc-grss.org
2
