EXPERIMENTAL ESTIMATION OF ERRORS IN RECONSTRUCTING THE STRUCTURE OF THE OBSERVED SCENE FROM A SERIES OF IMAGES BY VARIOUS CAMERAS

Abstract

The article is devoted to the study of the influence of using various mathematical models of cameras, and therefore models of scene image formation, when restoring the 3-D structure of a scene from a set of 2-D images during camera movement (restoring the structure from motion, hereinafter referred to as LEDs). A comparative assessment is carried out for two camera models: the classic central projection camera model and the relatively new omnidirectional camera model. The article provides a brief description of the mathematical model of an omnidirectional camera, the described model is used during experiments, and also describes ways to represent images from omnidirectional cameras. Additionally, a description of the mathematical model of the classical camera of the central projection is given. The described model is also used during experiments. The analytical calculations used in solving the problem of restoring structure from motion are briefly mentioned in the article. An algorithm for obtaining 3-D coordinates of the points of the observed scene from a sequence of images in motion is also described. The experiments carried out as part of the study are described in detail in this article. The process of setting visual landmarks and determining their true 3-D coordinates is revealed. The steps for the formation of data sets for obtaining comparative estimates are described. At the end of the work, an analysis of the experimental results is given, models are identified that reduce the errors in restoring the 3-D coordinates of the observed visual landmarks