Stratified reconstruction of uncalibrated objects based on 3-parameter model

According to the formation principles of the orthogonal vanishing point in perspective projection, and the camera's intrinsic parameter matrix invariance, a approach of stratified reconstruction is discussed through the use of the structural information of scene. Firstly, using corresponding points in a pair of images solve the fundamental matrix. In order to improve the precision of fundamental matrix, a rank 2 matrix is given to approximate the fundamental matrix. Secondly, the infinity homography matrix is calculated by the orthogonal vanishing points to implement affine reconstruction. Finally, the camera's intrinsic parameter is solved by the vanishing points constraint to obtain the image of the absolute conic, so as to realize metric reconstruction. The experiment results show that this approach is feasible.