Improving Geolocation by Combining GPS with Image Analysis - Geoinformatics for Intelligent Transportation - page 214

replicate the same conditions of these experiments, this tests were performed without
rain but with a very cloudy sky (dif cult illumination conditions) at 9 of June 2013,

between 1 h 30 pm and4h30pm.Inthese experiments the CV-GPS method was
con

gured to trigger the image analysis inside a threshold distance of 300 m of the
POI (X parameter), or after 2 s without GPS information (Y parameter).
Some of the experiments had similar goals, although performed in different routes
and different Points of Interest. This way, the first, second and third experiments had

the goal of simply detect the POIs. The first and third experiments were repeated

three times (referred as 1.1, 1.2, 1.3 and 3.1, 3.2, 3.3 in the results table). The fourth
and fth experiment

the error was inclusively provoked, by searching for a POI that did not belong to the
performed route. The

fifth experiments targeted the detection of false positives. In the

fifth experiment was executed twice (5.1 and 5.2 in results
table). The sixth and seventh experiments targeted the simulation of urban canyon
situations. To test this situation, we turned off the GPS signal during partial or total
time of the route. Finally, the eighth experiment had the purpose of testing a situation
where a vehicle is close to a POI, but without seeing it. This happens for instance in
single way roads, where the POI is only visible in one way. This way, the vehicle
may be passing very close to the POI, but the system only should target it when it is
visible (this is not possible by a GPS only solution, because the distance to the POI
would probably be the same (or approximately) in both ways of the road, but is
possible with the CV-GPS method because the system only returns a valid geolo-
cation when the POI is detected in the image.

5 Results

The execution of the set of experiments described in the previous section allowed us
to extract results and take conclusions about the correctness and validity of the CV-
GPS system. Table 1 summarizes the results achieved.
The evaluated parameters were the Duration, Number of kilometres of the route
(km), Average Speed in which the route was performed (Av. Speed), Image
Analysis Percentage Time (IAPT) which is the percentage of time in which the
image analysis method was running, Number of Analysed Images (NAI), Number of
False Positives (NFP) and finally if the POI was detected in time or not (before the

vehicle with the assembled system passes by it). Some of the parameters measured
and presented on the table are in

c, sema-
phores, crosswalks, etc.). All the tests were performed in an urban environment
with the limit speed of 50 km/h and with normal traf

fl uenced by the traf c conditions (traf

c conditions. The routes were
performed in different average speeds and during an approximated period of 3 h
between 1 h 30 pm and 4 h 30 pm. During that period a slightly change of
atmospheric conditions occurred, causing a slight increase of the luminosity, but
maintaining a very cloudy sky. The (a) label in the ``POI detected'' column in the
test 5.1 refers to the building detected in the false positive, once it was not corre-
spondent to the one we were trying to provoke in this test, but to another building in