Accuracy Of Image Matching Research Articles

Triangulation of Well-Defined Points as a Constraint for Reliable Image Matching

This study demonstrates the utilization of the well-defined points to improve the reliability and accuracy of image matching. The basic principle is: (a) to triangulate a few well-defined points within the stereo model area to form a coarse triangulation; (b) to detect certain amount of corners within each triangle for further matching; (c) to propagate the matching of corner points from the reference points (i.e., the three triangle vertices) to obtain the best matching for each of these corners; (d) to dynamically update the triangulation by inserting the newly matched corner; and (e) to further detect corners and perform matching for them until a pre-defined criteria (the minimum size of triangle or the largest number of points matched) is reached. Experimental results reveal: (a) the false matching caused by the occlusion and repetitive texture is diminished; (b) the accuracy is improved, i.e., with a reduction of RMSE of check points (located in different types of terrain areas) by 12 percent to 62 percent, and a reduction of the largest error by up to two times; and (c) most building corners and boundary points of main objects could be matched directly and accurately.

Automated image matching with coded points in stereovision measurement

A coding-based method to solve the image matching problems in stereovision measurement is presented. The solution is to add and append an identity ID to the retro-reflect point, so it can be identified efficiently under the complicated circumstances and has the characteristics of rotation, zooming, and deformation independence. Its design architecture and implementation process in details based on the theory of stereovision measurement are described. The method is effective on reducing processing data time, improving accuracy of image matching and automation of measuring system through experiments.

Application of rational function model to satellite images with the correlation analysis

RFM (Rational Function Model) is universally applicable to any types of sensors. The RFCs (Rational Function Coefficients) can be solved with or without knowing the physical sensor models. If the physical sensor model is available, so called terrain-independent solution can be used. Otherwise, the terrain-dependent solution should be applied. Most researches carried out recently have concentrated on the terrain-independent method, assuming that the physical sensor modelling is available. Most of high resolution satellite imagery launched recently, however, do not provide pertinent satellite ephemeris information to perform physical sensor model. For this it is required to have a general sensor model being independent on sensor types. In this case, the fitting accuracy of RFM highly depends on the terrain relief, the number of GCPs (Ground Control Points), and their distribution. Moreover the terrain-dependent RFM can be deteriorated by the over-parameterization among the polynomial coefficients. This research focused on the development of techniques to improve RFM solution, a matching technique, and the DEM (Digital Elevation Model) generation through the correlation analysis using the terrain-dependent solution. As a result, for KOMPSAT stereo image pairs over-parameterization problem has been successfully adjusted by choosing the optimal RFCs with the intensive correlation analysis. Also, the object space image matching technique using optimal RFCs could reduce searching area to a line. This matching algorithm adopted piecewise epipolar line for each corresponding height interval. Results show that the accuracy of image matching is reliable. About 55,180 matching points were obtained from object space image matching technique, and a DEM generated from these points shows the comparable accuracy with that of the DEM generated from the rigorous sensor model. RMS errors in northerly and easterly directions are 18.107 m and 32.278m respectively, and RMS errors in height direction is 34.577 m.