Quantized reality

Abstract

Capturing and reconstructing real world environments in 3D has broad areas of application including virtual reality, entertainment, archiving, simulation, and training. Geometry-based methods typically estimate the geometry of the environment based on multiple 2D cameras or depth cameras. These methods are applicable when the environment consists of solid and opaque substance. However, they cannot deal with transparent or highly reflective materials, or environments with an ambiguous surface such as fog. Light field space is a method that captures space by directly recording the light field, without using geometrical information. It can solve the problem faced by geometry-based systems in which view positions are limited when a 2D camera array is used. This problem cannot be solved by introducing a 3D camera array, because in such a configuration, each camera may occlude the images of other cameras. In this paper, we propose a space acquisition method that uses an autonomous robot that moves around the environment with a camera array. By combining very dense spherical images obtained by this robot, real world scenes are captured and reconstructed without view position limitations.