Geoscience Reference
In-Depth Information
uential variables into the
map matching process, like the angle between the directions of two lines, the
connectivity of network topology and accuracy of the GPS measurements [
3
Recently, research appeared to incorporate more in
fl
5
].
From a methodological point of view, existing map matching algorithms vary from
ad hoc rules to advanced learning-based algorithms, i.e. fuzzy set, Hidden-Markov
algorithm, probability hypothesis, etc. For instance, Pyo et al. [
6
] proposed the
multiple hypothesis probability (MHP) algorithms for map matching. The method
was further adopted and extended to the application of matching with GPS data [
7
].
Quddus et al. [
8
] introduced fuzzy logic theory to map matching with road net-
works. Some of the important variables, like the search space based on the error
ellipse derived from the error variances, the perpendicular distance from a position
-
fix to the link, the bearing of the link, and the direction of the vehicle, are incor-
porated. Six rules were created for fuzzy inference. The model was shown to be
more accurate than previous models. However, it needs some expert knowledge to
determine the fuzzy set.
Ren [
4
] proposed advanced algorithms for pedestrian and/or wheelchair navi-
gation services. A Hidden-Markov model and a so-called multi-sensor approach,
which incorporates the accelerometer into map matching, were used. Bierlaire et al.
[
5
] proposed a probabilistic algorithm, which can be very useful for smart phone
data. The model showed good ef
ciency and provided an opportunity to measure the
uncertainty of the candidate road sets. However, the model needs some simplified
assumptions in real applications. White et al. [
9
] and Quddus et al. [
10
] provide more
detailed literature reviews on the main existing map matching algorithms.
Unlike the empirical methods, these advanced algorithms can potentially better
handle the complexity in map matching procedures. Nevertheless, a well-performed
algorithm should be
flexible enough to measure the uncertainty of matched roads,
and be capable to incorporate the major in
fl
uential factors to ensure prediction
accuracy. Therefore, it is still needs to further improve the performance of map
matching algorithms. Considering the increasing number of applications of Bayesian
Belief Networks (BBN) in different
fl
fields of research and their superiority relative to
other algorithms [
11
,
12
], it seems adequate to examine the feasibility of this
algorithm in map matching. Therefore, in this paper, we will develop a map
matching algorithm which incorporates the BBN model. The model incorporates
some major in
uential factors, including distance to road, connectivity between two
road segments, direction difference of two line objects, the accuracy of GPS mea-
surement and the direction difference between two adjacent links. The method is
evaluated using the GPS data collected in the Eindhoven region, The Netherlands.
fl
2 Algorithm
To identifywhether a road segment matched a GPS point, it is necessary to incorporate
the in
fl
uential factors. In the following discussion, we will
first illustrate the main
in
fl
uential factors we adopted, followed by a presentation of the proposed algorithm.
Search WWH ::
Custom Search