| Fisheye camera is one kind camera with very short focal length and more than180degree perspective camera lens. In the research of fisheye optical, the conventional imaging principle is no longer applied. Researchers have introduced one principle called "no-similar imaging" which will bring "barrel distortion" for FOV range increasing. Therefore, instead of conventional lenses linear model, a nonlinear model has to be used for fisheye image distortion correction. There have several models been proposed by larger numbers of scholars, in case of radial distortion fisheye images correction. Also, model parameter estimation and optimization have become the focus of fisheye model research. This paper focuses on model establishment and parameters optimization, the main work are outline as follow:(1) We proposed a new fisheye camera model. First simplify the imaging process of the incoming rays by the optical refraction and derive the imaging model from relationship between incident angle and refraction angle of the incoming rays. Next, the justification of the model is tested with synthetic data through fitting the projection functions of the designed fisheye lenses. Unlike the previous fisheye camera model, our model has the parameters with a definite physical meaning. After compare validation of our model with common used fisheye model, model feasibility is given. Simulation experiments and real fisheye images of the scene experiments prove model reliability.(2) The model parameters estimated from homography constraint using second-order cone programming (SOCP) are presented. A suitable linearization of the fisheye camera model and of the homograph constraint p’=Hp (with p and p’are corresponding points of two images) between two spherical surfaces induced by the same space plane is investigated in order to arrive at a second order cone programming problem. Experiments with real fisheye images validate the approximated fisheye camera and the estimated algorithm of the model parameters. |
PDF Full Text Request