image of a stereo-pair by image matching. 
There are several techniques available to 
perform this task, which fall into two 
categories: area-based and feature-based 
matching. The area-based least squares 
image matching (LSM) technique has been 
selected for this development. LSM derives 
a shaping function from one image window 
to another by minimizing the gray level value 
differences between the two matching 
windows. Briefly, this is done by forming an 
observation equation for each pixel being 
matched at a particular position of the target 
array as it moves in the search area. Each 
observation equation compares the gray level 
values of the pixels being matched to 
estimate geometric and radiometric 
differences between the two images 
[Rosenholm, 1986]. 
LSM has several advantages: 
* High precision can be obtained. 
Precision of up to 0.05 pixel has been 
reported in the literature. 
» The geometric and radiometric 
differences between two images can be 
modeled. 
* The solution is highly redundant, even for 
small window size, which provides a 
strong least squares solution. 
* The information obtained at one matching 
position can be used as a guide for the 
next iteration, which avoids a systematic 
search. LSM .is. therefore 
computationally efficient and is suitable 
for a real time solution. 
LSM also posses certain problems: 
* The algorithm, which has been developed 
by the use of Taylor's expansion to 
linearize the observation equations, 
excludes the non-linear higher order 
terms from the solution process. 
Therefore, a good initial approximation is 
needed for the matching position . 
*  Multi-solution (non-unique solution) may 
occur in areas where repetitive patterns 
and high frequency texture exist. 
In the development of the pseudo-stereo 
mapping technique these problems have been 
largely overcome by predicting the close 
vicinity of the matching position. This is 
accomplished by applying the epipolar 
geometry as a constraint, and introducing a 
close approximation for the elevation of the 
140 
feature point to be mapped. In the absence of 
other information, the average terrain 
elevation, calculated from the control points, 
is accepted as the initial value. A contour 
map and familiarity with the general trend of 
the topography could facilitate the making of 
a better selection. This elevation is later 
refined as the correlation and the 
determination of the object space coordinates 
progresses and more exact elevation values 
are generated [Chen, 1993]. The procedure 
followed in the computation of the object 
space coordinates of the digitized points is 
shown in Figure 2. 
3. PERFORMANCE TEST 
3.1 Test Materials and Method 
The pseudo-stereo mapping technique was 
tested on a stereo pair of black-and-white 
photographs, covering the City of Fredericton 
and vicinity. The photographs were taken at 
a scale of 1:35 000, with a wide angle 
camera. The downtown area, which spreads 
along the south shore of the St. John River, is 
essentially flat at an elevation near sea level. 
From there the terrain has a steady incline 
and reaches an elevation of 130 m at the city 
limit. 
Digital images were generated by scanning 
the paper prints in a Hewlett-Packard ScanJet 
Plus document scanner. The contrast and 
brightness parameters of the scanner were 
adjusted before scanning to equalize the 
mean gray level values in the two images. 
Considering the limited disk space available 
in the computer, the scanning resolution was 
restricted to 300 dots-per-inch (dps) or 118 
dots-per-cm, which resulted in a pixel size of 
85 um in the image and 3.0 m on the ground. 
The radiometric resolution was set to 256 
gray levels. 
GCPs for the determination of the exterior 
orientation elements were obtained from 1:1 
200 scale maps. The contour interval is 1.0 
m overall, but in areas where the elevation is 
below 15 m, the interval is 0.5 m. 
A total of 120 feature points of different 
types were digitized on the screen in Photo 1. 
The features selected were residential streets, 
buildings, a bridge, railway line and a sports 
field. The ground coordinates of each were 
then con 
space in 
with the 
scale ma 
3.2 Test 
The resi 
Table 1. 
The RM 
of the p 
the pseu 
respect t 
maps. T 
the perc 
differen 
photogra 
block fl 
Therefor 
based on 
map acc 
scale. 
The acci 
and Resc 
and Wat 
states [N 
Ninety p 
features 
accuracy 
those w 
adversel 
Accurac 
area) ce 
between 
the digit 
represen 
Althougl 
87% of 
satisfied 
tolerance 
generall 
America 
90% con 
5.0 m for 
within th 
The follo 
put the re 
I. They 
m on 
2.5m 
stand 
2. The s 
8.5 x