Images 44 and 45 show a portion of the south expressway in urban area. The database is the same as previously and has a maximum error of 45 meters (6 pixels). But, we also use a search area of 10 pixels on each side of the original vectors.

Figure 44: Original image (Channel P)

Figure 45: Original image (Channel XS3)

ISEF has been tuned to extract thick valleys on channel XS3 ( Figure 46); the Laplacian filter ( Figure 47) is not tunable.

The dynamic programming is diverted in B by a shape close to a large curve, in A by the bridge over the GARONNE. The error in D is small enough to be corrected by the snake.

Figure 46: I.S.E.F. valleys (SPOT XS3)

Figure 47: Laplacian valleys (SPOT XS3)

On the EMDRO image ( Figure 48) the dynamic programming is diverted by an airport takeoff strip, the snake cannot correct this type of error. The conjunctive fusion ( Figure 49) cancel most of the problems, the remaining ones in A and B can easily be corrected by the snake.

Figure 48: DUDA Road Operator EMDRO (XS3)

Figure 49: Conjunctive fusion

The result of the snake shown on image ( Figure 51), regarding to original vectors ( Figure 50) shows a slight difference on the left part of the bridge over the GARONNE. In this part the original vectors are correctly placed, the snake has been diverted on a railway parallel to the expressway on a few meters on the bridge.

Figure 50: Vectors from Data Base

Figure 51: Snake result