Helen Burman
Figure 4 Example of interest points in an intensity image.
A dense grid of 21x21 pixels with about the same resolution as the laser data, was defined around each interest point.
Approximate values were then again calculated using the finite element method and using all laser measurements within
the area of interest. Gradients were calculated by using the Sobel filter. Each laser measurement within the interest area
produces one observation equation (equation 7 or 8). All observations from all areas of interest were put into one least-
squares adjustment and the unknown orientation errors were solved. As the observation equations are linearised, the
process was repeated until convergence. To avoid forested areas (which produce high gradients), a criteria was put on
the matching areas that a majority of the grid points should be describing a flat surface. This was done through Laplace-
filtering of the interest area.
5 PRACTICAL TESTS
Test area 1: A runway with flat and open terrain with large intensity differences between grass and hard-made surfaces
and high reflecting painted lines. The area was covered four times from 60 meters height in four directions (east-west,
west-east, north-south and south-north).
Test area 2: An oblong building and small undulations in the rest of the area. Large intensity differences between grass
and hard-made surfaces. There is both open and forested terrain. Two strips flown at 60 meters height in two opposite
directions (east west and west east).
The matching grid was chosen to approximately correspond to the resolution of laser points on ground. For 60 metres
flying height, the grid had a resolution of 0.2 meter.
130 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.