
Uncertainty: Uncertainty, in contrast to imprecision, depends on the object
being observed rather than the observing device. Uncertainty arises when
features are missing (e. g., occlusions), when the sensor cannot measure
all relevant attributes of the percept, or when the observation is ambigu-
ous [51]. A single sensor system is unable to reduce uncertainty in its
perception because of its limited view of the object [30].
As an example, consider a distance sensor mounted at the rear of a car
in order to assist backing the car into a parking space. The sensor can only
provide information about objects in front of the sensor but not beside, thus the
spatial coverage is limited. We assume that the sensor has an update time of
one second. This is a limited temporal coverage that is significant for a human
driver. Finally, the sensor does not provide unlimited precision, for example its
measurements could be two centimeters off the actual distance to the object.
Uncertainty arises, if the object behind the rear of the car is a small motorcycle
and the driver cannot be sure, if the sensor beam hits the object and delivers a
correct measurement with the specified precision or if the sensor beam misses
the object, delivering a value suggesting a much different distance.
One solution to the listed problems is to use sensor fusion. The standard
approach to compensate for sensor deprivation is to build a fault-tolerant unit
of at least three identical units with a voter [71] or at least two units showing
fail-silent behavior [42]. Fail-silent means that a component produces either
correct results or, in case of failure, no results at all. In a sensor fusion system
robust behavior against sensor deprivation can be achieved by using sensors
with overlapping views of the desired object. This works with a set of sensors
of the same type as well as with a suite of heterogeneous sensors.
The following advantages can be expected from the fusion of sensor data
from a set of heterogeneous or homogeneous sensors [10, 33]:
Robustness and reliability: Multiple sensor suites have an inherent redun-
dancy which enables the system to provide information even in case of
partial failure.
Extended spatial and temporal coverage: One sensor can look where oth-
ers cannot respectively can perform a measurement while others cannot.
Increased confidence: A measurement of one sensor is confirmed by mea-
surements of other sensors covering the same domain.
Reduced ambiguity and uncertainty: Joint information reduces the set of
ambiguous interpretations of the measured value.
Robustness against interference: By increasing the dimensionality of the
measurement space (e. g., measuring the desired quantity with optical
sensors and ultrasonic sensors) the system becomes less vulnerable against
interference.
Improved resolution: When multiple independent measurements of the same
property are fused, the resolution of the resulting value is better than a
single sensor’s measurement.
5