高分解能衛星画像のためのベクトル図形基準点の精度検証

抄録

Nowadays, high resolution satellite imageries can be used in various fields. The imageries are expected to measure precise coordinates. Then accurate geometric model is required to calculate image coordinates (u, v) from ground coordinates (x, y, z). Generally, RPC model is applied as the accurate geometric model. However, the RPC model has a systematic error with several pixels. To correct the RPC model, information of GCP (Ground Control Points) should be used. GCP are consisted with precise ground coordinates and corresponded image coordinates. An accuracy of ground coordinates is required less than 1/5 of ground sampling distance of satellite imagery.
GCP database in Shikoku Japan has been established by authors. Over 550 points were finished to survey within 4cm accuracy by GPS VRS observation. However, point based GCP doesn't have enough accuracy for high resolution satellite imageries. Because, image coordinates are selected by visual interpretation. Poly-lines or polygon type GCP should be prepared to select accurate image coordinates by image processing.
In this study, Poly-line type GCP were suggested. Target satellite imageries are ALOS PRISM which are panchromatic imageries with 2.5m ground sampling distance. In the test area, 5 intersections and 4 curves of road were surveyed, and Poly-line type GCP were prepared. Corresponding coordinates of Poly-line type GCP in PRISM imageries can be computed by image matching process. Least squares matching method was applied in this study. Then, precise image coordinates can be obtained.
Poly-line type GCP were validated using bias corrected RPC model. The root mean square error showed 0.5 pixel. On the other hand, point based GCP were also validated. In the case of point based GCP, corresponding coordinates were selected by visual interpretation. The root mean square error showed around 1.0 pixel. The results showed poly-line type GCP were effective for precise image registration.