The principle of sensor fusion is to combine information from various sensing
sources since no individual sensing technology is ideally suited for vehicle automation
under all modes of use. The appropriate sensor will depend on the field
status at the time of operation. But even under a given field operation, the
availability of data from multiple sensors provides opportunities to better integrate
the data to provide a result superior to the use of the individual sensor. Fig. 2
illustrates one example of a sensor fusion system to combine information from
multiple sensors.
Benson et al. (1998) used GDS and GPS together for vehicle guidance based on
dead reckoning as a simple path planner. The system was tested at slow speeds
(1.12 m:s) and had an average error of less than 1 cm, which compared favorably
to GPS-based guidance. When implementing a 3-m step change in responses the
sensor fusion system had a maximum overshoot of 12%. Under GPS-under mode
the system experienced a 50% overshoot.
Noguchi et al. (1998) developed a guidance system by the sensor fusion integration
with machine vision, RTK-GPS and GDS sensors. An Extended Kalman Filter
(EKF) and a statistical method based on a two-dimensional Probability Density
Function were adopted as a fusion integration methodology.