Coarse-to-fine fundamental matrix estimation by line correspondence based on the motion barcode

摘要

Estimating the fundamental matrix is one of the important techniques for 3D reconstruction using the images. For this purpose, a method using image sequences captured by cameras based on the correspondences of the lines which are randomly generated in each image was proposed. In this paper, we propose a coarse-to-fine method for fundamental matrix estimation using line correspondences based on the motion barcode to improve the accuracy. First, in the coarse step, our method generates the limited number of lines in each images and estimates the epipole in each image. Then, in the fine step, the lines passing near the epipole estimated in the coarse step are generated in each image near the lines selected in the coarse step and the fundamental matrix is calculated. In the both steps, the corresponding pairs of lines are selected based on the similarity of the motion barcodes of the lines and color around the moving objects on the lines.