Maximum likelihood autocalibration
Image and Vision Computing
| Hi-index | 0.00 |
We describe a novel approach to the camera auto calibration problem. The uncalibrated camera is first moved in a static scene and feature points are matched across frames to obtain the feature tracks. Mismatches in these tracks are identified by computing the fundamental matrices between adjacent frames. The inlier feature tracks are then used to obtain a projective structure and motion of the camera using iterative perspective factorization scheme. The novelty of our approach lies in the application of semidefinite programming for recovering the camera focal lengths and the principal point. Semidefinite programming was used in our earlier work [1] to recover focal lengths under the assumption of known principal points. In this paper, we relax the constraint of known principal point and do an exhaustive search for the principal points. Moreover, we describe an end-to-end system for auto calibration and present experimental results to evaluate our approach.