Algorithms for Quickly Reconstructing the Earth's Surface Using Parallel Processing of Several Aerospace Images

摘要

A method of the design and accuracy characteristics of algorithms aimed at the software-digital reconstruction of three-dimensional scenes and developed by the authors is outlined in this paper. The method, based on the simultaneous combined processing of an arbitrary number of planar stereo projections, was comprehensively examined by processing the real images acquired from high-Earth orbital satellites during their flight over certain areas of the Earth. Using information about satellite location and the camera's optical axis direction, we managed to minimize the number of couplings between the unknown external and internal camera's parameters and carry out a directed search through these parameters. At the same time, for unknown coordinates of the three-dimensional scene (relief), we succeeded in formulating an overdetermined system of linear equations, which can be solved by the method of least squares. Such an approach made it possible to accurately process up to four stereo projections with a reconstruction error lower than the increment fraction of the optical system's resolution. Designed algorithms are insensitive both to various types of noise accompanying any real survey and to the instability of the detection-channel parameters, as well as to errors in setting the initial conditions.