Fs sj 
[4 
  
    
Table 2: Ground position, speed and direction of the vehicles 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Vehicles ground position Vehicles ground position 
in MS-1 image in MS-2 image 
No. X, (m) Y,(m) X,(m) Y,(m) Displacement (m) Velocity (KM/H) Azimuth (Deg.) 
1 360661.9 | 5107772.9 | 360668.92 5107768.4 8.3 136.3 122.9 
2 360695.4 | 5107751.4 | 360701.42 5107746.9 7.5 1227 126.9 
3 360803.4 | 5107659.4 | 360808.42 5107654.9 6.7 110.1 132.0 
4 360832.4 | 5107653.9 | 360824.92 5107659.4 9.3 1522 306.3 
5 360822.4 | 5107643.9 | 360827.92 5107638.9 7.4 121.6 132.3 
6 360844.4 | 5107625.4 | 360849.92 5107619.9 7.8 127.3 135.0 
7 360883.9 | 5107594.9 | 360890.42 5107589.9 8.2 134.2 127.6 
8 360940.4 | 5107564.9 | 360934.92 5107568.9 6.8 111.3 306.0 
9 360947.9 | 51075624 | 360944.92 5107565.9 4.6 75.4 319.4 
10 | 3609424 | 5107544.9 | 360948.92 5107539.9 8.2 134.2 127.6 
11 361010.9 | 5107516.4 | 361006.42 5107519.4 5.4 88.5 303.7 
12 | 3609994 | 5107506.9 | 361006.42 5107500.9 9.2 150.9 130.6 
13 | 361042.9 | 5107466.9 | 361048.92 5107461.4 8.1 133.2 132.35 
14 | 361080.9 | 51074409 | 361087.42 5107434.9 8.8 144.8 132.7 
15 | 3610854 | 5107429.4 | 361090.92 5107424.4 7.4 121.4 132.2 
16 | 3611294 | 5107416.4 | 361123.92 5107420.4 6.8 111.3 306.0 
17 | 361156.9 | 51073754 | 361162.42 5107370.4 7.4 121.6 132.3 
  
  
  
  
to detect vehicles is minimized by utilizing the vehicles’ image 
positions detected from the previous section. This has increased 
both efficiency and accuracy of the Gentle AdaBoost 
classification algorithm in vehicle detection. Total 150 vehicles 
have been used to train the Gentle AdaBoost algorithm. The 
detailed description of this process is beyond scope of this 
paper. Once, vehicles from MS-1 and MS-2 images have been 
detected, vehicles’ centre positions are determined by 
calculating the centre of mass of the detected vehicles. This 
results in vehicles’ positions in MS-1 and MS-2 images. 
Speed calculation needs two ground positions of a moving 
vehicle. Vehicles’ ground positions can be computed using their 
image positions and the satellite sensor model (RPC). The HR 
satellites vendors such as QuickBird and WorldView-2 provide 
a rational polynomial coefficient (RPC) as their geometric 
sensor model. The RPC sensor model (Xiong and Zhang, 2008) 
is given as: 
Pi GC Z) 
  
C Pa (X, Y,Z) (4) 
Pa (X,Y,Z) 
BE © 
P(X,Y,Z) = X =o ko Un XYZ (6) 
Dem <3, O<m<3, 0Sm<3, 
Where (x, y) are the image coordinates, (X, Y, Z) are the ground 
coordinates and aj is the polynomial coefficients. 
The polynomial coefficients a; are provided by WorldView-2 
satellite. Therefore, by putting the vehicle’s images position in 
equation (4) and equation (5), the vehicle’s ground coordinates 
can be calculated. 
Figure 7 shows 17 vehicles selected for testing the accuracy of 
vehicle’s speed calculation. Vehicles from the MS-1 and MS-2 
images were detected using Gentle AdaBoost algorithm. Then 
centre positions of vehicles were determined by calculating the 
centre of mass of the detected vehicles. Next, the vehicles’ 
ground positions were computed using WorldView-2 RPC 
model. Finally, the vehicles’ ground coordinates were used to 
compute vehicles’ speed and direction. Table-2 shows the 
result. The vehicles’ speed shown in the result is normal on the 
highways. In the Table-2, speed of vehicle no. 9 is 75.4 Km/h. 
This is because; the vehicle is on slow lane. Similarly, speed of 
vehicle no. 11 is 88.4 Km/h because this vehicle has just joined 
the highway. Therefore speed calculated using the developed 
methodology seems realistic and can be used for traffic 
planning and management purposes. 
3. DISCUSSION AND ERROR ANALYSIS 
In this paper, it has been observed that the vehicles are more 
accurately detected on highways because there are wide roads, 
fewer trees and fewer manmade structures. The rate of false 
vehicle detection is high inside the city area. The accuracy of 
vehicle detection in the city will be improved if roads can be 
extracted from the Pan image before vehicle detection. 
The accuracy of vehicles’ speed computation is highly affected 
by the vehicles’ image position. If the vehicles’ image positions