Remotely sensed images are very useful for land-cover monitoring and land-use-change detection. Before these images can be interpreted or superimposed onto other maps, they have to be geometrically corrected and registered to the consistent datum. Current approaches for geometric correction rely on manually measurements of huge amount of control points, which is one of bottlenecks of the procedure. This paper proposed an ad-hoc Least-squares Model-image Fitting (LSMIF) algorithm to semi-automatically determine the orientation parameters of the remotely sensed images. The images are therefore geometrically corrected by 3D building models instead of control points. Since the operator only has to identify each 3D building model and their approximately position on the image, the efficiency of the manual process can be improved. A proto-type system is developed to verify the proposed algorithm and the semi-automated strategy. Several FORMOSAT-II satellite images taken over the coastal area in Tainan are selected as the experimental data. The 3D building models are generated from the existing 3D topographic map. The corrected images are evaluated by two means. First, the coordinates of the check points on the geometrically corrected image are compared to their coordinates on the topographic map. Second, the geometrically corrected images are compared to the level 3 and level 4 images corrected by current approaches. The experiment shows the proposed semi-automated approach does not only improve the efficiency but also achieve the required accuracy. It also shows the potential of applying to the upcoming FORMOSAT-V satellite images
ConferenceArchives
August 2013
Categories
All
|