>■ ‘ ■ v- 
The images were transferred to the computer by the 
software Kodak PhotoEnhancer using a cable connected to 
the serial port. Then the images were transformed from 
KDC format to TIFF. Row and column coordinates were 
measured by using PhotoFinish and the mouse cursor. An 
estimated standard deviation was set to 5 pixels due the 
zoom effect. Then the pixel coordinates were transformed 
to photocoordinates according to the following equations: 
x = (nc - nc 0 )-tpc + x 0 
У = (nl - nl 0 )-tpl + y 0 
where nc,nl are the numbers of the column and row, 
respectively; nc 0 ,nl 0 are the numbers of the central column 
and row, respectively (378, 252); tpc,tpl are the pixel size 
(45,6pm x 45,6pm), and x 0 ,yo are the principal point 
calibrated coordinates. 
There have been some difficulties to handle this kind of 
image specially when it comes to point selection and 
measurement due to the particular geometry. Considering 
figure 3, an image can be partitioned into fours triangles: 
middle up, middle down, left and right. Upper triangle 
comprises mostly the sky while middle down triangle 
covers mainly the street pavement. Left and right triangles 
contain the sidewalks, trees, gates, drives, walls, poles and 
other street elements that can be selected to function as 
pass or interest points. These limited portions of the images 
associated with high contrast and extreme scale variation 
bring difficulties to this method. 
ft , 
P 
\ 
L. -да 
r '■ 
Figure 3 - Two consecutive pairs of a photogrammetric 
traverse 
The photocoordinates (observations) and the interior and 
exterior orientation parameters (unknowns) were the input 
to a phototriangulation program (tftc) with the coordinates 
of the perspective center weighted according to the GPS 
surveying estimated precision. A bundle block adjustment 
was performed to compute the angular orientation of the 
terrestrial digital images. 
The object points were positioned in a hybrid system 
formed by UTM (E,N) coordinates and orthometric heights 
(H). The altimetric coordinates were computed based on a 
local geoidal model. The hybrid system was considered 
orthogonal because of the small size of the field area and 
then the approximated object coordinates were introduced 
directly in the collinearity equations without any coordinate 
system transformation. 
After accomplishing the complete exterior orientation, a 
simple intersection (based on a pair of images) or a double 
intersection (based on two pairs of images) can be applied 
to compute the final coordinates for the object points. 
These points represent a sort of distinct features like 
sidewalk and construction alignments, poles, trees, and 
garbage cans, and front land parcel limits. Each one of 
these categories was expressed in a separate layer. The 
alignment points were constrained by a straight line 
equation. 10% of the 150 alignment points had to be re 
measured in the images for the high discrepancies (up to 10 
m). In the street comers an arc equation constrained those 
alignment points. A topographic surface was interpolated to 
generate contour lines using Surfer 6.0 software. 
5 RESULTS 
A 1:2000 street line map was built. A display of it can be 
seen in figure 4. The accuracy of the final product was 
estimated by comparing the UTM map coordinates and 
GPS ground coordinates (transformed to UTM) of the 6 
check points. The computed root mean square error (rmse) 
was 0.76 m and 0.43 m for the N and E coordinates, 
respectively. A trend analysis was performed after 
computing the discrepancy average and standard deviation. 
Table 1 shows a summary of the figures involved. 
amausm 
l fte i** L**» £*5*чй U** te*»« la* is* 
"BQIsF 
i cmtm Oifew едаом 
ЬсЛёТГ 
■mwiomitB Wrp (Ллаикыи «uo 
gifc—v«-*Мм» I и» щя 
Figure 4 - 1:2000 street map built by photogrammetric 
traverse 
Table 1 - Trend analysis and accuracy statistics 
Parameter 
N 
E 
A (m) 
0.222 
0.069 
(m) 
0.797 
.0466 
t sample 
0.68 
0.36 
t(5,5%) 
2.02 
2.02 
Xa 
17.65 
6.03 
Xb 
6.35 
2.17 
Xc 
4.41 
1.51 
5A-5-3 
IШЩ l 
Si