MAP UPDATING USING IMAGE POINT TRANSFER TECHNIQUES

Abstract

Invariance theory has been recently used widely in the field of Image Understanding (IU) and Computer Vision (CV). This research presents the invariance techniques which deal with image point transfer and their applications on map updating. The main advantage of the invariance image transfer techniques is that they are linear with respect to the unknowns. The image transfer process is accomplished by measuring ihe coordinates in the three images only. Therefore, the process is performed without knowledge of the 3D objects (ground coordinates) nor the camera parameters. Two tasks, image point transfer from two images to a third one and to a map, are investigated using four different approaches. The first approach is based on the fundamental (F) matrix without applying constraints (linear solution). The second two approaches dea! with applying constraints on the fundamental matrix. One of them applies the determinant constraint, and the other applies the determinant and epipolar constraints on the fundamental matrix. The last approach is based on the four trilinearity equations. The mathematical derivation and analysis of these four approaches are provided in the research. Experimental results are presented in tables and figures for both simulated and

real data sets using convergent and normal geometry. Consequently, close range and aerial photographs are used, along with existing maps, for real data map updating experiments. The experiments show that using the fundamental matrix without constraints produces a large amount of error. However, increasing the number of reference points improves the solution. Applying constraints on the fundamental matrix stabilizes the solution. Finally, using trilinearity equations produces the best results of the four used approaches. Conclusions, recommendations and relevant appendices are provided at the end of the research.