Abstract:

Threedimension reconstruction based on images is one of important researches in computer vision. The reconstruction of single uncalibrated image of revolution surface is studied. The algorithm introduced by Colombo etc, which obtains threedimension reconstruction and texture acquisition at the same time, is improved and a new approach of reconstruction is proposed. Firstly, plane transmission is used to compute the equation of the ellipse with respect to the point of image apparent contour line. Secondly, the construction of the object is recovered by computing the location and size of each cross section of the revolution surface according to the property of quadratic curve. The method of the reconstruction process is simple and keeps the overall structure of the revolution surface more correct. Simulation and real image experiments demonstrate the correctness and validity of the algorithm.

Key words: revolution surface;metric reconstruction;camera calibration;planar homology;absolute conic