Multi-volume high resolution RGB-D mapping with dynamic volume placement

Abstract

We present a novel method for creating high-resolution 3D reconstructions using RGB-D cameras, such as the Microsoft Kinect. We demonstrate how such a technique can be used both for large scale mapping and high-resolution object detection. Our work consists of two core contributions. First, we present an algorithm to generate 3D reconstructions of arbitrary resolution, up to that of the limitations of the camera, with a linear increase in computation time as a function of resolution. We the build upon this method to provide a general infrastructure for scanning regions of variable shape with variable resolution across the scanned regions. We then show quantitative improvements in resolution in a virtual environment, show qualitative improvements in resolution through images, and provide runtime performance details.