Sensor planning for novel view generation by camera networks

Abstract

This document describes a system for generating novel views of an indoor visual scene by gathering successive 2D images from a set of independent networked robotic cameras. Specifically, the robotic cameras work to seek out texture and geometric information needed to generate the specified synthetic view or views, aiming to-with each successive camera move-increase confidence in the estimates of the pixel intensities in the novel view(s). This system lays the groundwork for future explorations in multi-camera video recording for electroholography and image-based rendering.