| Abstract: |  In this thesis, the algorithm LR73L for fundamental matrix estimation from 3 correspondences of local ane frames, i.e. the 3LAF2F problem, is proposed. The theoretical advantages of the approach are discussed rst and the approach is justi ed. Among other alternatives, LO-RANSAC is chosen as the basis for the proposed algorithm. The LO-RANSAC consists of parts, which were considered with respect to the 3LAF2F problem. The considered parts of LO-RANSAC are: sampling, hypothesis generation and local optimization. The proposed algorithm di ers when compared with the preceding algorithm for the 3LAF2F problem from [4]. The most important di erence is the use of the 7-point algorithm with the priority for longer arms of LAFs for the F calculation. This method provides higher quality fundamental matrices and lower computing time when compared with the 9-point method used in the preceding algorithm. The suggested algorithm LR73L was experimentally compared with its predecessor and
the traditional method for the fundamental matrix estimation from [13]. A speed-up of an order of magnitude was achieved when compared with the predecessor algorithm. The LR73L algorithm reaches quality and computing time of the traditional approach, however only for image pairs with more than 70% inlier ratio. Finally, it is stated that despite the improvements proposed in this work, the traditional approach remains the
better option for general use.
|
|---|