3)
re
re
Sergey Zheltov
right images by L(X, Y), R(X,Y) respectively. Subtraction of the orthophotos (Fig.3(c)) is performed by the following
way: D(X, Y) 2 L(X,Y) - R(X,Y).
(a)
i M
?Left Right v,
Figure 4. Invisible surface areas for left and right
images
(b) (c)
Figure 3. Example of orthophotos and their difference. (a) Orthophoto from left image. (b) Orthophoto from right
image. (c) Difference: (a) minus (b)
Pixel coordinates of (i,j)-th orthophoto point are calculated as follows:
(X,Y) = (X, +i *S, Y; +7 * Sy), where S, S, - grid sample distances along
X and Y axes respectively. The height Z(X,Y) is reconstructed from the
surface using the bilinear interpolation of four nearest surface values. To
assign a gray value to the (i,j)-th orthophoto pixel the point (X, Y,Z) is
projected to the image using collinearity equations. The gray value is
obtained by bilinear interpolation of four image gray values in pixels with
nearest to the projected point integer coordinates.
The basic principle of the developed detection method is the significant
difference in orthophoto due to different view positions of the cameras
(Fig.4). In the left image the surface area A» behind the object P is
invisible. The same is for the surface area A, in the right image. Thus the
gray values in invisible areas are taking from objects gray values. This
results in «projecting» the object to the invisible surface area.
5. DETECTION BY THE MATCHED FILTER
We assume that the object to be detected has rather contrast edges that are near to perpendicular to the surface. If the
stereosystem with respect to the surface is under an angle of view close to zero, then edges of the object become
elongated and enough rectilinear.
Let the object be a rectangle of height / and width w. Let the left-bottom corner of rectangle be located in surface point
(X, Y, Z) perpendicular to the surface and parallel to the image plane. Then the invisible area behind the rectangle is a
trapeze (Fig.5(a)) with dimensions satisfying the following equations (assuming the flat surface near the object
(Fig.5(b))):
b h v L zr - 1s] (7)
ben Bag
In these formula L is a distance from the camera to the object.
Im age Ms YZ)
1
Surface
a D (Y. 23 (b)
Figure 5. Object dimensions in orthophoto
For the following model parameters: h = 10cm., IZ- Zi — 1 m., L 2 50 m the parameter b is equal to 12.5 m. This shows
that Y-dimension of the invisible surface area dominates on other dimensions. So this fact actively used in the detection
algorithm, because the orthophoto difference contains clear triangles formed by object edges.
Intern
ational Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.
1045