Geoscience Reference
In-Depth Information
5 Summary, Conclusions and Discussion
Identifying the location of a GPS point on a road segment has been an important
research issue in transportation planning, transport modeling and environmental
analysis. In route choice simulations, the GPS data provides the merit with more
accurate and detailed real-time information, which is impossible to approximate in
traditional stated choice experiments. Moreover, the matched roads in combination
with the time information will also contribute to providing more possibilities in
microscopic simulation of traf
c
fl
flow and environmental analyses.
Development of an ef
cient map matching algorithm is of high importance in
different research
fields. Due to the fact previous methods are not ef
cient enough
in the aspect of
flexibility and learning capability, in this paper, we proposed an
algorithm using Bayesian belief network model for map matching of GPS data.
Variables employed in the model are distance to road, difference of directions,
difference of angles of two adjacent links, connectivity, number of satellites and
PDOP. The algorithm is investigated by using the data collected in the Eindhoven
region, The Netherlands. Simulation results showed that the BBN model shows a
good performance in recognizing the road segments with an accuracy of 87 %
(correctly classi
fl
ed instances).
As one may argue that the accuracy of GPS data is questionable, which are a
common issue in this study as well as others related to map matching algorithms,
such data should not give main in
uences on the matching results. Although the
extreme unrealistic data appear not much frequent in our GPS data, it is always true
that the effect of the noise data should be handled well. Therefore, when applying
this algorithm in real applications, a process to
fl
filter such data out is necessary.
Moreover, the algorithm needs to be improved by designing it more
fl
flexible with
respect to these inaccurate data.
As presented in the paper, the performance of the algorithm was evaluated
through a carefully recorded dataset. This was considered as a feasible way to
confirm the validity of this algorithm, because the validation of an algorithm in the
context of map matching is not much straightforward. There needs additional effort
to provide example data where suf
cient ground truth and various special cases
should be included. In spite of the acceptable ef
ciency of Bayesian network in map
matching (87 %), one cannot say it is a better algorithm than others at this moment
because of the difference in the GPS data used. Future research will go further to
examine some other algorithms by using the same data set to see the possible
superiority of the proposed algorithm.
As to test the ef
ciency of this algorithm, in this paper, a prototype of an enhanced
map matching algorithm was proposed and examined through a small sample data.
However, it is necessary in future work to include more sample data to further check
the generality in large scale applications. Since the current algorithm spent much
time in referencing the geo-objects in GIS, future research should consider the
processing speed to make it realisable real-time applications. In a more general
context, various special cases, which are common in GPS data, like passing through
Search WWH ::
Custom Search