1€
PI
is
le
ind bs
Masafumi Uehara
On the other hand, we make the pattern of the building
boundary from a digital map like Figure 8. Trace the track
of the vehicle by GPS over the digital map. A
perpendicular line is drawn from the building's edge faces
the street toward this track. The record of intersection of
this line and trace is utilized to make the pattern of
buildings.
The next step of process is to match the histogram of
measured points by EPI analysis with the pattern of the
buildings made from a map, then judge the location of the
building boundary in the peak of the histogram.
3.2 DP matching
In this research, we apply DP matching method to match
digital map data with 3D measured data by EPI analysis.
The pattern of building boundary from digital map and the
prospective pattern of building boundary from the
histogram of the 3D measured points by EPI analysis are
utilized as the feature vectors. Figure 10 shows an
example of the path of DP matching. In this figure, the
vertical line is the prospective pattern of building
boundary from the histogram, and a horizontal line is the
pattern from the histogram of the 3D measured data by
EPI analysis. If these two patterns correspond completely,
the path of DP matching in Figure 10 becomes straight.
But windows or doors of buildings except building
boundary are detected with 3D point histogram, the path
of DP matching takes a zigzag course.
Matching urban scene information with a map data
utilized DP matching method has reported in [6]. In that
research, the boundary pattern made from obtained
panorama image of urban scene has matched with the
building boundary pattern made from a map. This research
utilizes the depth-data of buildings by EPI analysis.
4 EXPERIMENTAL RESULTS
In these experiment, we used 600 consecutive input
images built from the video image took by a car running
along downtown (see Fig. 11). The car equipped with the
gyro sensor and distance sensor in order to record
vibration and to obtain the moving distance. GPS was
used to record the location of the vehicle. Obtained image
sequence was normalized using distance sensor.
Because of real environment, it is necessary to revise
vibration for image sequence. As a result of measurement
of gyro sensor, we had known that the vibration of a car
running on the road is more influenced by a pitch than by
a yaw and roll. Therefore, we shifted one of the two
consecutive images up and down, and calculated the
correlation between two images is maximized. Figure
12(a) showed the slit image before vibration removed, and
Figure 12(b) showed the image after vibration removed.
Prospective pattern of
boundary from his
aE
in
X
À
d !
Bs Bn emm r = +
B + à d
Ha 3 à i i :
¥
À à ^c
Ds AZ
Patter of building boundary
Eom: digital msgs
Figure 10: Example of path of DP matching
Figure 11: Source image
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 915