Brzezinska ef. al., 2003). A sample high-resolution and a 
video image are shown in Figure 6. 
EES 
zb 
en 
  
Figure 6. Sample images of the selected intersection: 4K by 
4K CCD (a), video (b). 
To support the vehicle matching and tracking in the image 
domain, the images were first orthorectified, i.e., all 
distortions due to surface and different camera pose have 
been removed. The processing, in general, benefits from the 
ortho domain; for instance, vehicle extraction can be done at 
true object scale and detection of moving objects can be 
easily obtained by simple image differencing, as shown in 
Figure 7. The vehicle extraction process includes edge 
extraction, intensity-based thresholding, profile analysis and 
morphological filtering (further details are in Paska and Toth, 
2004; Toth et. al., 2003b). The limitation of the 4K by 4K 
digital camera system, used in the Tucson, AZ flight, 
unfortunately allowed only for a 6-sec image acquisition rate, 
which was too low to obtain an adequate sampling of vehicle 
positions, and thus, resulted in unacceptable automated 
vehicle tracking performance. In contrast, the automation of 
video imagery resulted in excellent performance for small 
area tracking (see Mirchandani er. a/., 2003). 
852 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
  
  
  
  
Figure 7. Detecting moving objects in the ortho domain. 
5.3 Velocity Estimates 
The information on the vehicle counts and locations 
represents only the density aspect of the traffic flow, as the 
velocity, at least the average velocity, is needed to obtain the 
true flow data. For the LiDAR data, the observed vehicle 
sizes compared to the actual sizes can provide a basis for the 
velocity estimation. However, the problem is that only the 
major vehicle categories can be identified, and thus, the true 
vehicle size is unknown, as only the size range is known for 
a given category. Nevertheless, the coarse estimates for 
individual vehicles can lead to an acceptable average 
velocity estimate. Table 3 shows the statistics of the road 
segment shown in Figure 1. The imagery, in general, can 
provide a better source for velocity estimates. In our test data, 
good velocity estimates were obtained from the video, while 
the slow image acquisition rate of the 4K by 4K camera 
could deliver only coarse average velocity estimates. 
[Intersection flow volume of the test area in Figure 5 is shown 
in Figure 8. 
  
  
  
  
  
  
Lane £ Spacing Velocity Density Flow 
[m] [km/h] |vehicle/km] | [|vehicle/h] 
1 32.8 81.6 30.5 2487 
2 24.1 76.8 41.6 3187 
3 23.7 75 42.2 3164 
Average 26.9 77.8 38.1 2946 
  
  
  
  
  
  
Table 3. Traffic data. 
  
Balance of inbound and outbound traffic flow 
a 
a 
  
Vehicles 
n e Un 
-10 
-15 
-20 d LEE esaet til intact dtl as 
  
  
m East —— West North South | 
  
  
  
  
Figure 8. In bound and out-bound traffic flow of the selected 
intersection. 
In 
fe 
  
TT 
  
Uu Ld 
- Un