EI I
[Esl
Ld
—
à
duc EN E
lard
ing
een
nge
ons
ion
ime
nge
da
nge
/ of
this
nge
om
2-3,
ent
age
ing
een
S-2
eT
(2)
-T,
(3)
| to
The above process results in two images; first image contains
moving vehicles of the MS-1 image and second image contains
moving vehicles of the MS-2 image.
Figure 2: Second principal component of MS-1 image (MS-1:
PCA2)
Figure 3: Second principal component of MS-2 image (MS-2:
PCA2)
Figure 4: Second principal component of MS image (MS:
PCA2). In this figure, two positions of the same vehicle are
available.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
2.6 Post-Processing
Although thresholds used in equations 2 and 3 significantly
reduced the outliers, still there are false alarms (structures other
than the vehicles) present in the resulting images. To eliminate
the false alarms, c/ean filter, majority filter, hole filling filters,
and opening/closing morphological operators discussed in
Sharma et al.(2006) have been used. In the above filters, (1)
Clean filter removes the one pixel size structure surrounded by
Os. (2) Majority filter sets a pixel to 1, if five or more pixels in a
3x3 neighbourhood are 1s; otherwise, it sets to the pixel to 0.
(3) Hole filling filter fills isolated pixels, such as centre pixels.
(4) Opening/closing morphological operation removes one or
two pixel wide interconnections between two pixel groups.
After filtering operations, Otsu threshold method (Otsu, 1979)
has been used to convert gray level image into binary (black and
white) image. An algorithm has been developed to count
vehicles from the generated binary images. The final result of
moving vehicle detection is shown in Figure 5 and 6. Figure 5
shows the moving vehicles detected from the MS-1 image and
Figure 6 shows the moving vehicles detected from the MS-2
image.
Figure 6: Result of vehicles detected from MS-2 image
2.7 Results of vehicle detection
The experiment results are based on WorldView-2 imagery of a
part of Moncton, a city in New Brunswick, Canada. A very
small part of the results have been shown in the figures due to
space limitation. The whole image was taken as input to the