Stratified self-calibration and metric reconstruction for zooming/refocusing circular motion sequences

Abstract

Self-calibration for imaging sensors is essential to many computer vision applications. In this paper, a new stratified self-calibration and metric reconstruction method is proposed for zooming/refocusing cameras under circular motion. With the assumption of known rotation angles, the circular motion constraints are first formulated. By enforcing the constraints gradually, metric reconstruction is retrieved up to a two-parameter ambiguity. The closed form expression of the absolute conic w.r.t. the two parameters is deduced. The ambiguity is then resolved with the square pixel assumption of the camera. The advantages of this method are mainly as follows: (i) It gives precise results by defining and enforcing the circular motion constraints; (ii) It is flexible that it allows both the focal lengths and the principal point to vary; (iii) It requires no scene constraint. Experimental results with both synthetic data and real images are presented, demonstrating the accuracy and robustness of the new method. © 2008 Springer Science+Business Media, LLC.