3D Spatial Information Extraction from Stereo Combination of Two Different Satellite Images

This paper describes the geometric accuracy testing of the IRS-1D and SPOT-4 images when combining together as stereo, over a test field in Zanjan province in the west part of Iran. For reconstruction of the image orientation we have used two different models: (i) a rigorous bundle adjustment program using an orbital parameters model and (ii) a generalized sensor model as rational function model. These models have already been tested for stereo SPOT Level 1A, Level 1B, IRS-1C, MOMS-02, as well as IKONOS. These mathematical models and analytical photogrammetric solution are first described in brief. This is followed by the results of the various 3D geometric accuracy tests carried out with these images using different sets and combinations of control and check points. The GCPs for these tests are extracted from 1:25,000 scale topographic maps produced by National Cartographic Centre (NCC) of Iran using 1:40,000 scale aerial photographs. Finally an analysis of the results is given. Introduction These days the huge capability of satellite images, including their spatial, spectral, temporal and radiometric resolutions as well as their stereoscopic viewing, introduces them as a powerful source for Photogrammetry, Remote Sensing and GIS communities. To get more benefits from satellite images, one approach is to combine them together. Although there have been a lot of research on merging these images to obtain better radiometric and spectral information, the research on combining these images to extract 3D spatial information is less. The purpose of this paper is to report on (i) an investigation into the mathematical modeling of IRS-1D and SPOT 4 images and differences between rigorous and generalized sensor models, and (ii) an evaluation of 3D spatial information extracted from the combination of these images. Test Area and Materials One SPOT-4 Level 1A stereo-pair images taken on 14 May and 21 June 2000, and one single IRS-1D Pan in superstructure format taken on 4 May 2003 covering a part of Zanjan Province in the west part of Iran, have been used in this research. The crosstrack angles for the left and right images of SPOT stereo pair are +24 and -26.4 degrees respectively while for IRS-1D image it is just -1.8 degree. For this project 30 Ground Control Points (GCPs) were extracted from 1:25,000 scale digital topographic maps of the area. These maps were produced by NCC using 1:40,000 scale aerial photographs. The accuracy of the contours where a 10m interval has been used is estimated to be ±3.5m. In total 32 well distributed GCPs have been selected and measured. The Mathematical Models: (i) A rigorous bundle adjustment program using an Orbital Parameters Model : An orbital parameter model can be applied to the linear array stereo images such as SPOT-4 and IRS1D in order to determine their exterior orientation parameters. The model adopted here is developed by the first author to deal with cross-track configuration of linear array images such as IRS1D, along-track configuration such as (MOMS-02) as well as flexible pointing configuration such as IKONOS. The spacecraft is moving in a defined elliptical orbit. The position and attitude of the spacecraft are changing continually in a systematic way to keep the sensor pointing towards the centre of the Earth. An orbital resection method is developed to model these changes by finding the orbital parameters of the spacecraft during the period of its exposure of the stereo-pair. A bundle adjustment program based on the following collinearity equations has been developed to determine these parameters using GCPs: