Pose Estimation using Point and Line Correspondences

The problem of object pose from 2D to 3D correspondences has received a lot of attention both in the photogrammetry and computer vision literatures. Various approaches to the object pose (or external camera parameters) problem fall into two distinct categories: closed-form solutions and non-linear solutions. Closed-form solutions may be applied only to a limited number of correspondences [1,2]. Whenever the number of correspondences is larger than four, closed-form solutions are not efficient and iterative non-linear solutions are necessary [3,4]. The latter approaches have two drawbacks: (i) they need a good initial estimate of the true solution, and (ii) they are time consuming. Therefore, such approaches can not be used in tasks that require high speed performance (visual servoing, object tracking, ...) [5,6,7,8]. To our knowledge, the method proposed by DeMenthon and Davis [9] is among the first attempts to use linear techniques, associated with the weak perspective camera model in order to obtain the pose that is associated with the perspective camera model. The method starts with computing the object pose using a weak perspective model and after a few iterations converges towards a pose estimated under perspective.