A novel approach for camera self-calibration from projective reconstruction

A novel approach for camera self-calibration is addressed in this paper. It is well known that one of problems for camera self-calibration is the matrix of the dual image of absolute conic (DIAC) is must positive definite. Then calibration matrix can be gotten by cholesky factorization from DIAC. In this paper, calibration matrix is directly optimized with nonlinear method which means that the solution of DIAC matrix is not necessary. It can help us avoid the positive definite problem. The algorithm builds on the basement of projective reconstruction, and it includes two steps. Firstly, the initial value of calibration matrix can be estimated from the manufacture explanation, then initial guess of infinity plane vector is searched out. Secondly, 8 parameters containing calibration matrix and infinity plane vector are optimized with Levenberg-Marquardt algorithm. Experiments validate the method.