Estimation of camera 3D-position to minimize occlusions

Abstract

Occlusions are almost always seen as undesirable singularities that pose difficult challenges to recognition processes of objects which have to be manipulated by a robot. Often, the occlusions are perceived because the viewpoint with which a scene is observed is not adapted. In this paper, a strategy to determine the location, orientation and position, more suitable so that a camera has the best viewpoint to capture a scene composed by several objects is presented. The estimation for the best location of the camera is based on minimizing the zones of occlusion by the analysis of a virtual image sequence in which is represented the virtual projection of the objects. These virtual projections represent the images as if they were captured by a camera with different viewpoints without moving it.