Suitable Type of Ground Control Point for High-resolution Satellite Imagery

摘要

Highly accurate geometric correction is necessary for high-resolution data. Then, it is important to acquire the high accuracy Ground Control Point (GCP) and corresponded Image Control Points (ICPs) for geometric correction. The objective of this study is evaluation of GCP type for high-resolution satellite image. In this study, following two types of GCP was selected: l)Polygon in field surveying using GPS 2) Template image using aerial photograph Accuracy was evaluated using Root Mean Square Error (RMSE) which was calculated by comparing around GCP and validation points. In this study, IKONOS was used as high resolution imagery. 3D affine transform was carried out for the geometric correction of the IKONOS imagery. GCP polygon such as agricultural field were surveyed by kinematic method of GPS. Then, a center of gravity was calculated using the polygon data. The corresponded ICP polygon image was extracted from the IKONOS image by visual interpretation. A center of gravity was also calculated using the polygon image. Then geometric corrections were carried out using control points obtained from GCP polygon and ICP polygon. In the result, RMSE showed less than 0.56 pixels. GCP from gravity point became enough accuracy. For image matching using aerial photograph, ortho image was established using aerial photograph and DSM. And, template image was generated from the ortho image. Land cover in the image was much different between aerial photograph and IKONOS image. Therefore, the classified image was generated by normalized Euclidian distance form training data of road. Image matching was carried out by using the classified template image and IKONOS image. Classification image can be adapted for image matching, which showed 0.08 correlation and 0.189 pixel error. The original color image showed 0.07 correlation and 14.06 pixels error, for image matching. Image matching with classification image made reliable result rather than original color image.