Detecting Spatial and Temporal Route Information of GPS Traces - Geoinformatics for Intelligent Transportation - page 69

Fig. 5 Distances from the GPS log points of different days to road segments

The true sample data was prepared using the carefully recorded diary and the
road network data within GIS. The locations, which cannot match with any road
segments, were excluded. We divided the sample data into two datasets, one is for
training purpose and the other is for test purpose. The datasets were selected ran-
domly in terms of trips and road categories, by setting 75 and 25 % as the training
and test dataset, respectively.
Since the connectivity is considered as one of the determinant variables, it is
important to

first road segment of the starting log point (origin). Because of
the effect of signal issue in the activity locations, the log points which are closed.to
the true activity location might be

nd the

fluctuated. Moreover, a prompted recall cannot
ensure the mentioned location as a truth. Therefore, we add some additional

fl

filters

in assisting the identi cation of the origin. More speci cally, we include the

information of personal pro les and the validated location names in combination

with the activity type. In concrete, if the personal pro le is available in terms of the

activity type, we picked the coordinates of their personal locations directly from the
database of personal locations. Otherwise, the location names will be used for
online geo-coding to

find a valid coordinates through Google service. Finally, if

both the personal pro les and location names are missing or incomplete for some

reasons, we fetch the original GPS traces according to the start time and/or end time
of the activity. Finally, the coordinates of the origin are generated as to provide
references of map matching for the consequent GPS log points. A detailed
implementation procedure is depicted in Fig. 6 .