The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B3h. Beijing 2008 
texture signature is calculated. Selecting which texture 
signature as the measure of ATS, it should judge by the road 
conditions. After a lot of experiment, we conclude that, taking 
variance, strand deviation or entropy as the measure of ATS if 
the road has salient characteristic d) while taking mean as the 
measure of ATS if the road has obvious characteristic e). 
ATS 
3.4 Compute compactness ant j move f orwar( j one s j e p 
ahead 
Calculate the ATS compactness by Eq. (4). If the value is larger 
than predefined threshold , it tells us that the ATS is not fit 
any more to track road, and it needs least squares template 
matching instead. Otherwise the direction of the limit is 
regarded as the road direction, and moves the road trajectory 
one step. 
3.5 Compute the change of the curvature 
Calculate the curvature of the new added road centreline by Eq. 
(5), compare it with the curvature of last point, if the difference 
is larger than predefined threshold T 2 , delete the new point 
from the road trajectory, and resort to manually plot. Otherwise, 
predict the next road position by parabola equation and iterate 
from 3.3 if the trajectory doesn’t reach to another trajectory or 
the boundary of the image. Once the user accomplishes one 
road segment or the tracker reaches to one tracked road or the 
boundary of image, initializes another road segment and restart 
from 3.2 again until all roads are tracked. 
From the operator point of view, the procedure is as follows: 
the operator has to initialize the tracker by three input points to 
indicate the starting, the moving direction and the width of the 
road segment, and then the tracker is triggered. Whenever the 
internal evaluation of the tracking tool indicates that the tracker 
might lost the road axis or be no longer fit, it demands for 
interaction of the operator. Then the operator has to confirm the 
tracker or the user must edit the extracted road and put the 
tracker back on the road. 
4. COMPARISION OF FOUR TRACKERS 
To verify our algorithm, we make a comparison between least 
square template matching, least square profile matching, snakes 
and our tracker on a same Quickbird image of Huai’rou County 
in Beijing, China, whose size is 355 by 1066 pixels. On this 
image, there is a brighter centreline on the homogenous darker 
road surface with a brighter background. The results are shown 
in Fig. 3. All trackers extract the road centreline with different 
precision in red colour. For profile matching, the front part of 
the path is quite good but the last part of the extracted road 
trajectory has a larger deviation due to the larger curving of the 
road. For template matching, the extracted road trajectory is 
good but with some larger deviate points along the trajectory. 
For snakes, if there is only two seed points on centreline, the 
extracted road trajectory is quite wrong, as shown in Fig. 3(c), 
the up line; but if there is 5 road seed points, the result are quite 
good, as shown in Fig. 3(c), the down line. For ATS taking 
variance as measure, there is some acceptable deviation in the 
middle part of the road. For ATS taking mean as measure, the 
extracted road trajectory is very good. We also record the time 
needed by each tracker. Profile matching is so fast that it 
finishes immediately after initialization. The snakes is slower 
than profile matching. The template matching is slower than 
snakes but faster than ATS. We can get the conclusion that our 
proposed algorithm is more robust than other trackers. 
5. EXPERIMENT AND EVALUATION 
Fig. 3 Comparison of algorithms (a)The result of profile 
matching (b) The result of template matching (c) The result of 
snakes (d) The result of angular texture signature tale standard 
deviation as measure (e) The result of angular texture signature 
take mean as measure 
The algorithm proposed here was tested by one Quickbird 
image over Hefei City, An’hui Province, China. The image with 
2181 by 1998 pixels contains many different road types such as 
straight roads, curves, and crossings at different orientations. 
And for each segment, the surface material is same, but there is 
sudden change in radiometry, as shown in Fig. 4. The roads 
have different disturbing objects such as shadows of trees and 
542