Geoscience Reference
In-Depth Information
WIFI, Bluetooth as well as applications of smart phones in data collection, the
requirement of an ef
cient and valid map matching algorithm to identify the exact
locations of the trace points on network data becomes extremely important.
Basically, to identify whether a node belongs to a speci
c road segment (link) is
not straightforward because of the difference between the spatial locations of the
node and the link, road network topology and the accuracy of geographical mea-
surement. This can be more complicated if a node is located in an area surrounded
by high density buildings where the strength of the satellite signals heavily in
u-
ences the accuracy of the location measurement. In addition, the variety in some
special cases, like U-turn, passing through a round-about, lane changing, also
increases the complexity to design map matching algorithms. A well-performed
model should not only handle these main research issues, but provide suf
fl
cient
robustness and uncertainty measurement in the model to be applicable in different
contexts.
Due to the fact that GPS data
fluctuate according to the contextual information,
like the weather, urban density, sensitivity of the GPS sensor, etc., the traces
measured on a temporal scale using the same device can have different accuracy.
Therefore,
fl
flexible enough to capture such
uncertainty. From a long-term view of perspective, it seems to be important that the
designed algorithm incorporates a learning function into the model in the sense that
the estimated parameters could be adjusted through an intelligent learning proce-
dure, which will then increase the overall accuracy of the map matching results with
more data coming in.
A common approach for map matching, which has been adopted empirically, is
by means of spatial analysis functions provided by geographical information sys-
tems (GIS) tools. The buffering area along with the geographical objects, i.e. lines
or nodes, are created
the algorithm is necessary to be
fl
first, and used to match other geographical data where various
criteria of overlapping
filters are set in advance. Such a method is popularly applied
because it is convenient to implement, but it needs to set the threshold of the
searching radius which can be different according to different accuracy of the
measurement. For example, Du and Aultmann-Hall [
1
] have empirically used a
10 m threshold to create the buffer and match the road network with GPS points.
Although this threshold value has been shown to have acceptable accuracy, the
method cannot meet the requirement when matching personal traces. In particular,
the variation of the coordinates leads to the dilemma that some nodes either cannot
be recognized as belonging to any roads in case of a small search threshold or are
overly matched in case of a big radius.
A prototype map matching algorithm was developed by using the distance from
node to node, the distance from nodes to lines and/or lines to lines [
2
]. These
algorithms are consistent in the sense the distance was considered as the only
decision variable. This can be problematic in real applications because of the
ignorance of other important variables, like the connectivity of road segments.
Although such algorithms have shown a good performance regarding the pro-
cessing speed, the matching accuracy is dif
cult to ensure.
Search WWH ::
Custom Search