Accuracy analysis of DEM extraction over Japan using ALOS PRISM stereo images

This study is to make an accuracy assessment of the DEM extracted from a stereo pair of ALOS PRISM images of Kanazawa area, Japan. First of all, we computed 14 linear coefficient parameters in 3D perspective transformation (3DPT), using 7 and 9 triangular points whose three dimensional coordinate values were given by Geospatial Information Authority of Japan. The DLT model characterizes the relationship between two dimensional image coordinate system and three dimensional object coordinate system. As for the tie point on the stereo pair image, we selected the GCP set L, the GCP set M, and the GCP set H, consist of 15 triangular points with low elevations (0m 200m), respectively. The three DEMs were generated with an aid of the OrthoEngine module (PCI Geomatica Ver. 10.3), by assigning these three GCP sets as the tie point sets. It is very important to input the correct set of pixel and line coordinate values for the DEM extraction. For this purpose, the pixel and line coordinate values for the Nadir view image and the Forward View image were calculated by the DLT with newly computed linear coefficients. The accuracy analysis of the extracted DEMs was examined at independent 10 Check Points (CPs). We found the most accurate DEM was generated using the GCP set M. The overall accuracy of the DEM with 2.5m spatial resolution was computed to be RMSE = 5.8m in the vertical direction by comparing the extracted elevation values with measured values at 10 CPs.