Camera self-calibration with varying parameters based on planes basis using particle swarm optimization

This work proposes an approach of cameras self-calibration with varying intrinsic parameters from two images of any 3D scene. The present method is based on the projection of several unknown planes of the scene by their homographies in the two images taken with different viewpoints. Each plane is characterized by a Euclidean reference which is defined on two points of the scene whose their projections are well localized, then a global reference of the scene is chosen from the considered references. Then, the self-calibration equations are formulated by the expression of each homography which is defined by the intrinsic cameras parameters and the coordinates of the two points on which the Euclidean references of the planes are chosen. These equations are introduced into a nonlinear cost function, the optimization of this cost function allows to estimate the intrinsic cameras parameters. The strong points of the present approach are: a minimum images number (two images are sufficient), only five matching points and a new formulation of the cost function which does not require too much computing to converge towards an optimal solution of the cameras parameters. Extensive experiments on synthetic and real data are presented to demonstrate the performance of the proposed technique in terms of simplicity, precision, stability and convergence.