Boosting camera calibration performances

The aim of this thesis is to provide a general method for accurate camera calibration (resectioning). The task consists in finding a set of parameters to describe the camera according to a predefined reference model. The problem was studied since the beginnings of computer vision. Until now, photogrammetric (marker-based) camera calibration outperforms self-calibration techniques in terms of reprojection error minimization. It could be considered solved for most applications, but for the higher-demanding ones and in the presence of strong nonlinear distortions more accuracy would be desirable. There are some practical aspects that lead to difficult procedures: accurate camera calibration remains a topic for expert, trained users. The principal ideas behind nonlinear lens distortion modelling have been described, for both pinhole and fisheye lenses, without exploiting color nor motion information, together with general camera calibration issues and a projective bias exempt marker: a grid of ring markers solves the problem. Faster alternatives to the assisted camera calibration algorithms available in literature were researched. Estimators optimality criteria and matrix operations properties have been exploited to obtain computationally cheap ways to evaluate poses, allowing to have a denser pose space. We proposed and tested three new criteria for pose suggestion.