The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3b. Beijing 2008 
561 
Figure 7. Car tracking by normalized cross correlation of a group of three cars detected in the first image of a sequence (right) to 
the second image (left). Clipped images were taken from the scene shown before at the motorway A 96. 
correlation score is stored, if the correlation factor is greater 
than 0.9. We introduced this border in order to avoid hits, if the 
car from the first image is not anymore present in the second 
image. A border of 0.9 turned out to be the best value to track 
vehicles present in both images, and to avoid errors in tracking 
if the vehicle is missing in the second image. With the 
maximum of the correlation score, the image coordinates in the 
second image as well as the corresponding geocoordinates at 
the location where the maximum in correlation appeared are 
stored. This gives the location of the vehicle matched from the 
first image into the second image. With this position stored for 
each vehicle, we are able to repeat the application of the car 
tracking module for an image pair consisting of the second and 
third image of an image sequence, and afterwards it is applied 
to the third and fourth image. Hence, we are able to track a 
detected vehicle over the whole image sequence of one 
exposure burst. Moreover, we are able to track vehicles from 
one image sequence into another image burst, if the overlap is 
sufficient. However, if the overlap is in the range of 10 % to 20 
%, only few vehicles are situated in more than one image 
sequence, so that we consider reasonable to restrict vehicle 
tracking only to image pairs obtained within one exposure 
sequence. 
4. RESULTS 
We tested our processing chain based on the data take from 
30.04.2007 as described in Chapter 2. For that, the 
completeness and correctness of vehicle detection and tracking 
are determined on data of several resolutions, obtained from 
different flight levels. 
4.1 Road Detection 
Road detection was performed using two different modules. It 
turned out, that detecting roadside markings for determining the 
road area is a good strategy on images taken at a lower flight 
height of 1000 m resulting in a resolution of 15 cm GSD. 
Nevertheless, at higher flight levels (for instance at 2000 m) 
road extraction works well with the module searching the edge 
between blacktop and vegetation. Fig. 3 shows typical results of 
road extraction using roadside markings. Top image shows the 
line extraction, whereas in the image below the finally extracted 
roadsides after smoothing and closing gaps are shown. 
4.2 Vehicle Detection 
In order to quantify the vehicle detection efficiency, test data 
were processed and the results of the automatic vehicle 
detection were compared to a manual car detection. Table 1 
shows the results of the comparison between automatic and 
manual car detection. On a flight height of 1000m (15 cm GSD), 
vehicle detection performs well on motorways with a 
correctness of around 80 % and a completeness of 68 %. In a 
complex scene like the city ring road we can proof that car 
detection delivers respectable results with a completeness of 65 
% and a correctness of 75 %. However, at a flight height of 
2000 m (GSD=30 cm) performance drops down to 56 % in 
completeness but correctness is still high with 76 %. 
Site 
correct 
false 
missed 
correctness 
completeness 
Motorway 
(1000m) 
85 
22 
41 
79% 
68% 
Motorway 
(2000m) 
95 
30 
76 
76% 
56% 
City 
(1000m) 
47 
16 
25 
75% 
65% 
Table 1. Results on testing vehicle detection on data 
obtained at several test sites (from different flight 
heights). Counts of correct vehicle detections, false 
alarms and missed detections, as well as correctness and 
completeness in percentage are given. 
Figure 6 shows examples of vehicle detection performed on 
images obtained at a flight height of 1000 m. Upper image was 
taken on highway A96 near exit Munich-Blumenau, lower 
image shows part of the circular road “Mittlerer Ring” in 
Munich city. Only few false alarms were detected. 
4.3 Vehicle Tracking 
Vehicle tracking was tested on the same data takes obtained at a 
flight height of 1000 m (15 cm GSD) and at a flight height of 
2000 m (30 cm GSD). Figure 7 shows a typical result on 
tracking vehicles from the first image of an image sequence into 
the second exposure of the sequence. On images with a 
resolution of 15 cm GSD, vehicle tracking on motorways 
performs perfectly well, with a correctness of better than 95 % 
and a completeness of almost 100 % on each image pair. On 
images obtained from higher flight levels (>30 cm GSD) 
tracking still works fine with a completeness of 90 % while 
having a correctness of 75 %. We attribute the good tracking 
performance on low flight heights to the fact that with a 
resolution of 15 cm GSD vehicle details like sunroof, 
windscreen and backlite, and body type go into the correlation 
which simplifies finding the correct match. However, these 
details are not anymore seen at higher flight levels.