Automated geo referencing system
Scientific Research Institute of Precision Instruments, 51, Str., Dekabristov, Moscow, 127490, Russia
One of the critical problems of Earth remote sensing is data binding to images when necessity of on-line processing of large bulk of data with due attention to high precision and reliability arises.
Manual binding requires a great number of resources. Thus, a need to design an automated data binding system satisfying imposed requirements becomes relevant.
Data binding assumes imagery comparison, when one image is transformed to match another one. Precise automatic imagery comparison presents a complicated task and has a set of solutions, heavily dependent on the nature of photographed scenes. Creation of universal method seems to be impractical for several reasons. Thus, with allowance for a number of assumptions, we propose a method combing currently available methods.
The first stage of the proposed method consists in search area selection, based on a priori information.
In the course of the second stage, the search area is covered with temporary points.
Very often images of earth sensing contain uninformative temporary areas in the form of forests, fields, water bodies etc. Thus, we need a method allowing obtaining mostly clear cut objects among uninformative areas.
To achieve this goal we suggested implementation of Moravec operator based on calculating the mean value of image pixels values squared differences in different directions. Image fragment with considerable brightness changes gives greater value of Moravec operator.
For search area uniform coverage with temporary points, and multiprocessor systems performance optimization, the search area is split into sectors, and Moravec operator is applied to each sector.
The third stage consists in points positions refinement using areal methods. Prior to it, scale difference and rotation of compared images compensation should be performed.
Supporting points obtained undergo verification, and pairs of points with poor measure of concordance will be rejected. Then compared image is transformed according to selected transformation model, using previously found supporting points.
This work has been realized as part of the software package, built on a client-server architecture. Calculations are made on dedicated servers. The client part is used by operators, to generate jobs and sending them to the server for later execution, as well as performance monitoring, display and editing obtained results.
Keywords:geolocation, remote sensing of the Earth, image transform, informational content, correlation, image processing, image comparision
Jacobsen K. Orientation of High Resolution Optical Space Images, ASPRS Annual Conference, Tampa, 2007, pp. 7-11.
Dial G., Grodecki J. Block Adjustment of High-Resolution Satellite Images Described by Rational Polynomial, Photogrammetric Engineering & Remote Sensing, 2003, pp. 59-68.
Triggs B., McLauchlan P., Hartley R., Fitzgibbon A. Bundle Adjustment — A Modern Synthesis, ICCV ’99 Proceedings of the International Workshop on Vision Algorithms: Theory and Practice, pp. 298-372.