Towards point-based acquisition and rendering of large real-world environments

Abstract

This paper describes a pipeline for the acquisition and rendering of large real-world environments. In the acquisition phase, we use a laser rangefinder to capture the geometry of an environment, and a digital camera to capture its colors. In the rendering phase, we use a cluster of commodity PCs to render high-resolution images of the environment at interactive frame rates. In this paper, we describe in detail our scanning hardware, the tools we use to minimize the acquisition artifacts in the 3D scans, the procedure to register the scans to each other, and how to map colors from a photograph to a scan. We also present a sequential, out-of-core rendering approach that uses multiple threads to overlap rendering, visibility computation, and disk operations. Finally, we show how to use the sequential rendering approach as a building block for a parallel rendering system that uses a cluster of PCs to drive a high-resolution, multi-projector display wall. Our acquisition approach allows us to capture environments that would be extremely difficult to model by hand, and our rendering approach allows us to use inexpensive PCs, instead of high-end graphics workstations, to visualize those environments at interactive frame rates.