me 
)se 
ind 
Ids 
hin 
for 
ing 
tes 
fa 
leo 
in 
the 
88, 
  
319 
2. THE TEMPLATE CONFIGURATION 
The structure of template comprises two parts. The first part is a white square which is adopted as the actual control point. The 
second part is a pattern referring to the control point number. The patterns are placed to the right side of the squares at a constant 
distance. It should be noted that this constraint was used in this project for achieving a minimum time delay for processing, and the 
place of the patterns according to the methods of photogrammetry can be designed. Figure 1 illustrates the templates and patterns of 
points numbered from one to ten. Table 1 introduces the patterns up to one hundred and gives their descriptions. 
In addition to distinctness and simplicity there are several reasons for adoption of large white squares for control points. The first is 
for the approximate determination of three rotation angles ( &, @, K ) using the sides of the squares. It is more obvious for an observer 
to recognise the effect of three rotation angles on a straight line than a curve. Therefore the sides of the square can be used to obtain 
the approximate value of the three rotation angles, and the values can be used to accelerate the computation and processing within 
real time photogrammetry. Figure 2 shows the perspective distortion effect of three rotation angles on the square. Most important is 
the relation of the length of the sides and the position of the top right corner of the square with the position of the numbers. As 
mentioned above, the numbers are positioned at right hand side of square with a constant distance. The constant distance is equal to 
half of the square's side. Additionally, the length of each vertical line is equal to the length of the sides of the squares. The distance 
between the vertical lines is equal to half of the side length. The thickness of the lines in the image should be at least one pixel. 
Therefore it is necessary to take into consideration object distance, camera resolution, and scale factor when determining line 
thickness. For good results the length of the lines should not be shorter than 10 pixels size in the image. These requirements 
complicate the problem, but guarantee a robust and reliable result. An algorithm for detecting and recognising the control points and 
the numbers referring to them is under development using the C language. The first step of the algorithm is the detection of the 
control points. If an object similar to a control point is recognised, the target will be tested and will have a weight assigned to it. 
Potential targets with the highest weights above a preset threshold are chosen. 
Methods for the recognition and detection of templates are different for different tests. For example, the method for the first test in 
close range photogrammetry can be used in real time digital close range photogrammetry. In this method, the coordinates of control 
points in the object coordinate system which are distributed in the image, will be automatically retrieved from a file. Eventually 
approximate position of control points in the image can be computed using collinearity equations. Next , the control points are tested. 
A control points should be a four -sided, closed polygon, and the average pixel value inside the polygon should be more than optimal 
threshold. When the control point is recognised the pattern will be detected. Various methods for pattern recognition can be found 
based on matching the pattern against a template. One method is such dynamic programming (DP) matching. Sim et al (1994) 
explained the application of DP for the recognition of Korean characters. The symbols which are used in the templates for referring a 
number to control points are not as complicated as Korean characters. Therefore, the method for patten recognition is simpler and 
faster than DP matching. This method is base on computing the percentage of the optimal square occupied by the pattern. It seems 
this method has two problems. The first problem is that in reality the optimal square is not known, especially when the image is 
taken in convergent close range photogrammetry. The second problem is that the percentage of some patterns are the same, for 
example number 10 and number 50. The first problem can be compensated for by transferring the pattern to an optimal square 
according to the three obtained rotation angles ($, €, K). For eliminating the second problem, the pattern will be tested with all 
possibilities of similarities, and the correct number will be extracted. Another method is scanning the whole of images with an 
optimal window. When the control points are recognised, the pattern will be recognised. 
In order to achieve a satisfactory detection and recognition of the template, in real time or for on line processing, it is necessary to 
use a calibrated camera with known position. The greatest advantage of the templates is their ability to extend to numbers over 
1000, and their applicability in other disciplines of photogrammetry, such as aerial triangulation and aerial stereo matching. 
3. PRACTICAL RESULTS 
This section will discus the practical results which have been obtained from digital close range photogrammetry and aerial 
photographic tests. The different experiments were conducted to test the template matching under different conditions. 
Automatic detection and position extraction of the control points for stereo matching for close range photogrammetry at real time, or 
atleast on-line, processing are the main justifications for creating of the templates. Verification of the process can only be obtained 
from testing on real images. For the close range test the distance between the camera and the object should not be more than a few 
metres. Two off-the-shelf CCD cameras with approximate resolutions of 750 by 580 were used to acquire images. 
In the first close range test images have been acquired from a design in which templates were distributed at the same range within a 
test field (see Figure 1). The CCD camera was set up at a distance of 75 cm from the test field. A number of images were taken in 
ambient room lighting conditions. The method of detection is based on prediction using the collinearity equations. Achievement the 
results in real time or on line is the major consideration in this project, so constraints must be applied in the initial state to simplify 
the search problem. For example, a calibrated camera should be used, and the position of camera should be approximately known. 
These initial conditions reduce the range of unknown parameters and therefore reduce time delay due to computer processing. Once 
the control points are recognised and identified at the first stage, then the position of control points on the image can be tracked. In 
this test images were deliberately taken in room lighting conditions and no enhanced mask was applied to the image, in order to test 
the templates under typical conditions and speed up the processing. The collinearity solution locates the selected control points. A 
window is opened to cover the whole of template, and a threshold is defined according average pixel value within the window. The 
number of the control points is identified and read according the threshold. This method is fast, but the calibration factors and 
camera position must defined initially. 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995