A robust self-calibration algorithm based on three views

We present a new robust self-calibration route based on three views. First, we can achieve the initial guess of focal value by another form of Kruppa equation. Second, three affine parameters can be solved in linear equations according to the absolute dual quadric. Gauss-Newton method is applied to solve Modulus Constraint equation instead of continue method. At last the whole intrinsic camera parameters are obtained by Cholesky decomposition. Many experiments with both synthetic and real images show that our method is effective, robust and accurate.