Generation of digital terrain model and image interpretation by using EROS-Al stereo pair images observed from multiple directions
This study focuses on the improvement of automatic DTM (Digital Terrain Model) generation by using the stereo pair images of high-resolution satellite data. There has been a problem how to precisely correct the registration of corresponding points that are used to estimate the elevation in stereo pair images. DTM is usually generated by a set of stereo pair images. In this study, the authors propose an automatic DTM generation using high-resolution satellite EROS-Al images observed from multiple directions. The image registration is preliminary examined in WGS-84 coordinates system derived from GPS observation to set up ground control points for the rectification of each stereo pair image, in addition, the precise registration is completely performed using the camera model of EROS-Al image. As for the matching of stereo pair images, an automatic image correlation processing is performed by means of searching in each pixel based on correlation coefficient of test window. In addition, the comparison with correlation was performed on every combination of stereo pair images observed from multiple directions of different azimuth and pointing angle for evaluating and improving of accuracy of DTM. DTM is successfully generated by means of the above procedure proposed in this study. As the result, the residual error of registration was estimated 1.15 pixel in the mean rms at pixel direction and 1.47 pixel in the mean rms at line direction on each stereo pair image, and the residual error of DTM was distributed from + 11 m to -14 m approximately corresponding to GPS data on the ground. Finally, the authors verified that the stereo pair images observed from multiple directions have not only a capability to precisely generate DTM, but also a capability of improvement of image interpretation for identifying the structure of land covers, facilities and the geographical features by generating the color composite images consisted of each stereo pair image. 2003 COSPAR.