ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002 
  
record 
Figure 2 Choosed coordinate system 
The method of random shift 
The method is based on record averaging. 
The principle of the method is based on the as best as 
possible detection of relative positions of the record 
according to one selected record (usually the selected 
record is the first snapped) with the accuracy of the 
one tenth of the pixel. 
Using of records positions information they 
are translated and rotated in the way that in selected 
coordinate system all the objects of the scene have 
identical coordinates for each snapped record of the 
scene. When imagined the records as overlays, then it 
is the same as translation and rotation of the overlays 
to get a feeling that it is one overlay. 
The neigbouring pixels parts sizes of the 
record (label — the contributions) which cover the 
actual pixel are determined and the weighted average 
of color values of that contributions are calculated. 
Figure 3 illustrates one of the possible cases. S is the 
area of actual pixel, S;, S», S;, S, are areas of record 
pixles parts which cover the actual pixel. I is the 
contribution to resulting color value of actual pixel. /;, 
I”, I; 1, are color values of record pixels overlapping 
the actual pixel. Then it is valid: 
  
  
  
SS TSUCS TS (1) 
I =[*S; + 12S, + I* S; L* S; (2) 
nd | 
/ i 
/ ™ D. 
J MN cM 
, 
1 T > l4 
$47 A p 
| "^ K 8 IL y 
\ > | 4. M 
a Tm 2 L^ um 
  
  
Figure 3 Contribution of the record to resulting color of 
actual pixel 
When by this means achieved record 
contributions to the resulting intensity of actual pixel 
are averaged through all suitable records, the actual 
pixel color value is achieved. 
For determination of relative position of the 
record according to the first snapped two detection 
objects are placed in the scene. After thresholding of 
the records and successive localization of these 
detection objects their centers of gravity are 
calculated. This will help to determine the relative 
positions of the records. 
The determination of detection objects 
In the next it is supposed that every detection 
object is created by pixels with darker color as their 
neighbours in the vicinity of detection objects. 
The first thing to do is to isolate the detection 
objects from the background. The one of the effective 
and easy implementable methods fairly solving this 
problem is the method of thresholding. All the pixels 
lighter than certain threshold value are filled with 
white color (the background color). The rest of them 
have original color which is used in the calculation of 
the weighted average in the determination of the 
detection objects centers. 
I suppose, that every detection objects is 
continuous. Because the thresholded image can 
contain in addition to detection objects others objects, 
it is necessary to find the way how to ignore them. In 
my application the detection objects are selected from 
all the detected objects by interactive user 
determination of one point from the detection object. 
For detection of the areas of the detection objects 
which are bounded by the background color I use the 
method of the area filling (seed fill). 
When I know which pixles belong to 
detection objects I can determine its centers. Because 
the quality of the result strongly depends on the 
accuracy of the determination of that centers, for their 
computation I use several methods. 
1. The method of simple average, where the center is 
the arithmetic average of the coordinates of all 
detection object pixels. 
1 
TAX. 3 
nz ; (3) 
T -the coordinates of the detection object center 
N - the number of pixels of the detection object 
X; - the coordinates of the i-th pixel of the detection 
object (17 1, ... N) 
Q - the set of all detection object pixels 
2. The method of simple weighted average, where 
the center coordinates are calculated as the weighted 
average of all detection object pixles and the weight of 
every point of detection object is its color. E.g., in my 
application the darker is the point the more important 
it is in the calculation of the center. 
1 
FE== VAS (4) 
v A 
T - the coordinates of the detection object center 
N, - the sum of weights of all the pixels of the 
detection object 
v;- the weight of the i-th pixel of the detction object 
X; - the coordinates of the i-th pixel of the detection 
object (17 1, ... N) 
Q - the set of all detection object pixels 
A - 259