Remote GPU-Accelerated Online Pre-processing of Raster Maps for Terrain Rendering

Abstract

We present a distributed architecture for accelerated pre-processing of remote sensing data for immediate terrain visualization. Interactive 3D visualization approaches for large terrain datasets employ level of detail techniques that require a multi-resolution data representation. The high computational cost of constructing these representations is often not viewed as a major drawback, as it is considered an o-line pre-processing step. This prevents the application of existing methods in the case of changing data, which is becoming increasingly important for a multitude of applications where datasets are being generated, transmitted and must be visualized immediately, such as in disaster management. Our system uses graphics processing units (GPUs) to accelerate the process of generating a multi-resolution representation, achieving sucient performance to enable on-line visualization on a front-end workstation communicating with a back-end cluster of machines equipped with GPUs. As a reference data structure, we use a quad tree decomposition of the so-called HEALPix sphere parameterization, which is well-suited for spherical terrain rendering. Our system correctly handles overlapping and unregistered mixed-resolution datasets. We demonstrate the ecacy of our approach by applying it to the surface of Mars using both the NASA Mars Orbiter Laser Altimeter and the ESA Mars Express High Resolution Stereo Camera datasets.