Unsupervised 3D Scene Reconstruction by Stereo Pair Images

The article is devoted to the problem of unsupervised 3d scene reconstruction by stereo pair. A brief description of the stereoscopic system model is presented. The system parameters and methods for their evaluation are briefly described. Equalities are given for calculating three-dimensional points of the scene and building a depth map. The algorithm for searching for conjugate points on a stereo pair is described in more detail. A method for its multi-thread implementation on a GPU is also described. The algorithm is tested on images from open datasets and compared with competitors. The results of reconstructing a real 3D scene with distance measurements are presented.