Emerging New Trends in Hybrid Vehicle Localization Systems - Global Navigation Satellite Systems: Signal,Theory and Applications - page 280

a GPS receiver as the sole vehicle localization measurement source may turn to be unreliable,
especially in urban canyons and other areas where the satellite signal can be distorted or lost.
A number of solutions have been reported in the literature that proposed augmenting GPS
measurements with information about the vehicle's motion in order to improve localization
accuracy. In what follows we provide a summary of a number of such solutions.

2.1 Dead Reckoning (DR) and GPS integration
A DR is a localization method that estimates the next location of a mobile object over a series
of short time intervals, given the object's direction, speed, and previous location. DR is simple
and known for producing incremental error and hence needs to be reset periodically.

It is

therefore suitable for use over short periods of time.

One approach to resetting the accumulative localization error is to combine DR with GPS
whereby GPS measurements are used to reduce the DR accumulative error; when the GPS
measurement is unavailable, the DR estimates the location using sensors such as wheel
odometers, a flux-gate compass, a gyroscope, and an accelerometer Kao (1991).

2.2 Inertial Navigation System (INS) and GPS fusion
Basically, INS operates as a DR system. INS employs a computing unit and motion sensors to
estimate its location without relying on any external reference once it is initialized using for
example a GPS measurement. To avoid the accumulated error caused by the measurements of
internal sensors in INS, the INS location estimate is fused with measurement data from other
sources. As discussed in Skog & Handel (2009) fusing INS and GPS can take the form of a
loosely or tightly coupled system architecture.

An example of a system that fuses INS and GPS is the real-time kinematic global positioning
system (RTK GPS) Bouvet & Garcia (2000) which uses an Extended Kalman Filter (EKF) to
fuse data. In this system, GPS latency is defined as the time required for the satellite signals to
travel to Earth and the time required for the computation of the location; GPS latency varies
with the number of observed satellites. Therefore, the GPS latency is encapsulated in the EKF
state so that the fusion of the INS and GPS data is synchronized with the readings of the
sensors.
It is possible to fuse standard GPS and INS by means of a KF as well Honghui & Moore (2002).
In this case the computational complexity of the EKF can be reduced by preprocessing the INS
measurements and inputting them into the KF as a linear component. However, preprocessing
the INS measurement adds to the computational cost of the solution.

2.3 Other motion sensors and DGPS fusion
Integrating the INS of a dynamic model with a DGPS is also investigated inRezaei & Sengupta
(2005). To deal with the nonlinearity of the dynamic model, an EKF is used. Due to the
accelerometer noise other motion sensors, such as six wheel-speed encoders, a steering angle
encoder, and an optical yaw rate gyro, are used instead. Localization accuracy of 0.9 m on 100
m driving track was reported for situations where the system relies on the dynamic model
more than it does on the GPS measurements. The multipath effect is not addressed as the
experiment was conducted in an open space environment.
In Aono et al. (1998) a method of positioning a vehicle on undulating ground by fusing DGPS
data and motion sensor data is proposed. A fibre optic gyro, a roll pitch sensor, and wheel