基于无穷单应的大视场摄像机标定方法

在大视场摄像机标定中,常常会出现由于场景过于单一而很难达到自标定所需要的场景约束和运动约束条件、立体标定所需要的强立体条件或者平面靶板标定所需要的绝对共面条件,如指向高空区域的摄像机标定任务就很难满足上述要求,因而大视场摄像机标定需要较为弹性的标定算法。提出一种基于无穷单应的大视场摄像机标定方法,该方法最少只需要4个非共线控制点和摄像机粗略的位置即可求解无穷单应,并且提出一种坐标变换方法以保证线性求解和优化无穷单应时的稳定性。从无穷单应中分解得到摄像机参数初始值,通过Levenberg-Marquardt(LM)优化算法最终实现摄像机的标定。在优化过程中,通过假设图像中心为主点和采用一阶径向畸变模型,相对增加了优化过程中的剩余自由度,能够实现4个像点为观测值的参数优化。相比于强立体或共面的条件,此方法所需条件很容易满足。仿真和实际实验验证了此方法的正确性和高精度,以及重复测量实验的灵活易实施。

Calibration Method for Large Field of View Camera Based on Infinite Homography

During the calibration for a large field of view camera, it usually appears such a situation that the scene is too simplex to satisfy the condition of scene and motion constraints for self-calibration method, the strong three-dimensional condition for three-dimensional calibration and the condition of absolute coplane for flat target plate calibration. It is difficult to fulfill the mentioned conditions while calibrating a camera pointing to the upper-air area. So it calls for a more flexible algorithm when a large field of view camera is calibrated. A calibration method for large field of view camera based on infinite homography is pointed out which just needs approximate position of the camera and at least four noncollinear control points to solve the infinite homography. A coordinate transformation method is also proposed to ensure the stability during the linear solution and optimization process of the infinite homography. The initial camera parameters for Levenberg-Marquardt (LM) optimization algorithm are obtained by decomposing the infinite homography. During the optimization process, the hypothesis that image center is the principal point and the adoption of the first-order radial distortion model will lead to the augment of the redundant freedom, so that the parameters optimization can be realized by using just four image points. The condition needed by the proposed method is facile to satisfy compared with strong three-dimension or absolute coplane. The simulation and actual experiments not only prove the correctness and high precision of this method, but also testify that repeated measurement experiments can be easily and flexibly implemented by this method.