Global Positioning System Reference
In-Depth Information
encoders are used as motion sensors. The positioning accuracy is improved by compensating
for the error for each sensor. The error is determined by means of a KF, which is also utilized
as a fusion unit.
In Sharaf et al. (2005) an Artificial Neural Network (ANN) is chosen as a tool for detecting
errors and noises in INS measurements using a DGPS as a guide to the true location of the
vehicle during a training phase. The work reported in Sharaf et al. (2005) is similar to that
reported in Bouvet & Garcia (2000) in that preprocessing operations are performed on the
measurements before they are fused. An assumption that is made in this method is that
the DGPS data is always either available or unavailable due to an outage in satellite signal.
However, in urban areas, satellite signals are often available but quite often are contaminated
by multipath noises, which effects the quality of the ANN learning.
3. Fusion of landmark, INS, and GPS measurements
Detecting and recognizing landmarks provide spatial information related to the local
environment. It is therefore possible to integrate spatial information with localization
measurements from DR and GPS in order to improve localization accuracy Fuerstenberg &
Weiss (2005); Jabbour, Bonnifait & Cherfaoui (2006); Jabbour, Cherfaoui & Bonnifait (2006);
Rae & Basir (2007); Weiss et al. (2005). Two approaches for detecting and augmenting
landmarks to vehicle localization systems are presented next along with another localization
technique that attempts to detect visible satellites for use in the positioning process.
3.1 Laser scanners, digital maps, and GPS/DR
Due to the accumulated error caused by the long satellite outages in GPS/DR localization
systems, digital maps are utilized to perform localization during such outages Weiss et al.
(2005). A laser scanner mounted on a vehicle scans major objects in the vehicle environment.
The system matches these landmarks with other landmarks in the digital map that represent
the region of interest. If there is a match, the vehicle location is estimated by correlating the
identified landmarks.
However, segmentation is not a trivial job specially in situations where landmarks are merged
with background objects. Moreover, the system must be trained by having it traverse the
regions of interest Fuerstenberg & Weiss (2005) to extract landmarks (features, such as traffic
signs and the posts of traffic lights) that can later be used as a reference points.
In Jabbour, Bonnifait & Cherfaoui (2006), a vehicle equipped with an autonomous navigation
system and a laser scanner is reported. The laser scanner is used to detect the edges of
sidewalks and estimate the distance between the edge of the sidewalk and the vehicle.
Distance measurements are utilized to improve the accuracy of a localization system that
comprises GPS, DR, and Geographic Information System (GIS). The GIS data contains
digitized information such as abstract road maps, road edges, and other landmarks.
Landmark information is created through a learning stage. During the testing stage, the EKF
fusion technique produces an innovation value from which the system determines whether
to accept the fusion location estimate. If the GPS data is corrupted by multipath signals or
is unavailable, only the DR location estimate utilized. The vehicle location estimate is used
to select the region of interest from the GIS database that contains the landmark information.
To improve the vehicle location estimate, a matching scheme is performed to compare the
GIS-extracted landmarks (i.e., sidewalk edges) with those extracted by the laser scanner, and
Search WWH ::
Custom Search