The Scene Structure Construction Algorithm

Abstract

One of the most important problems in the technical vision systems development is the problem of a static scene high-quality spatial reconstruction in accordance with images captured by moving camera. 3D-structure reconstruction is required in different areas where human activity is being actively robotized: medicine, industry, autonomous vehicle control systems, and many others. Traditional methods of spatial reconstruction are based on the images local features search, features matching, the fundamental matrix evaluation according this correspondence and the spatial structure of the scene is then calculated. The quality of the feature based methods generally depends of the feature sample quality for the fundamental matrix evaluation. Based on the variational optical flow model with the additional epipolar geometry constraint algorithm of a scene structure reconstruction from a sequence of images is proposed. The proposed approach of the spatial reconstruction is build on the optical flow and epipolar geometry joint functional minimization with the simultaneous scene points coordinates determination. The method of one gathered functional for all variables minimization provides higher accuracy of their joint computation and, accordingly, better scene spatial structure reconstruction. Experiments show that the method is a good alternative to the feature matching based approaches and their 3D-position computation based on triangulation. A sequence of a static scene images is required as input data and the internal parameters of the camera are assumed to be known.

Keywords

• optical flow
• epipolar geometry
• features correspondence
• fundamental matrix
• spatial reconstruction