Projective Reconstruction of 3D-curves from its 2D-images using Error Models and Bundle Adjustments

In this paper, an algorithm for projective reconstruction of general 3D-curves from a number of its 2D-images taken by uncalibrated cameras is proposed. No point correspondences between the images are assumed. The curve and the view points are uniquely reconstructed, modulo projective transformations. The algorithm is divided into two separate algorithms, where the output of the rst is used as input to the second. The rst algorithm is independent of the choice of coordinates in the images and is based on orthogonal projections and aligning subspaces. The ideas behind the algorithm are based on an extension of aane shape of nite point conngurations to curves. The second algorithm uses the well-known technique of bundle adjustments, where an error function is min-imised with respect to all free parameters. The errors in the detection of the curve in the images are used in the error function. These errors are obtained from a proposed model of image acquisition and scale space smoothing, making it possible to analyse the errors in a simple edge detection algorithm. Finally, experiments using real images, have been carried out and it is shown that the results are superior to previous approaches.