CAMIDES Project

Many new types of cameras are being used these days. These new types of cameras use special lenses, several types of mirrrors and special geometries. These cameras include:

cameras with fish-eye lenses;
cameras with panoramic annular lenses;
cameras with hyperbolic, parabolic and elliptical mirrors;
omnidirectional cameras with multiple centers of projection;
other geometries;

One of the goals of this project is the determination of the conditions for the identification and estimation of the camera model that generated a sequence of images. The problem to be dealt with can be decomposed into two main subproblems:

given an image sequence find out which kind of camera was used to acquire the images; this problem corresponds to the determination of the best camera model that “explains” the image sequence;
given the same sequence of images to estimate the parameters of the camera model that generated those images;

To deal with this problem it is considered that the images were acquired from the same scene. The correspondence problem will not be dealt with in this project. The techniques arlready available to solve for the correspondence problem will be used. To identify the type of model a predefined set of models will be initially used. One of the goals will be the use of camera models as general as possible. To estimate the parameters non-linear estimation and optimization techniques will be used.

To perform the identification the minimal conditions will be studied. For that purpose a set of predefined models will be used. For each of the models in the set the minimal conditions will be established. These conditions will be defined in terms of the minimum number of features (points, lines and/or conics) whose correspondence throughout the image sequence will have to be known. The identification may require some knowledge of the 3D scene. The project will also determine the amount of 3D information required, depending on the type of model, number of features and number of images available. The problem of the parameters estimation will also be dealt with. In this case the goal will be the incorporation of model restrictions in the estimation process so that optimization and estimation become robust.