Fourteen (14) fiducials are chosen from several standard aerial photographs (some of them can be seen in the previous segmentation section). In order to compare the automated location results from different location operators, all fiducials are measured manually on screen with zoom factor up to 5:1, It is reasonable to assume the measured results are close the true positions. The idea of our paper is trying to replace the human operator using the fiducial segemtation and location algorithms. It is reasonable to compare our results to the operator results. For a real fiducial mark, nobody knows exactly where the true location is. Table 1 shows 14 fiducial marks' location results. From the table, we found, in average, that less than 0.5 pixel accuracy can be reached by our location approach, while some larger location errors (around 2 pixel) existed in the fiducial image (prsm1g, prsm2g, see Table 1), the reason of the error is that the original fiducial is not a completed target due to scanning problem or, there is some obvious distortion on the fiducial.
We think the main problem currently in interior orientation is to fully automatically segment the fiducials from an area, we haven't concentrated on the location procedure so no thorough comparison has been made.
Table 1:
Fiducial mark location results (Unit is in pixel, Wong's:
Wong location operator results; Förstner's: Förstner location
operator results, Optimal's: Optimum Location results; Manual:
manually measurement results).