Computational optimized 3D reconstruction system for airborne image sequences

Kurzfassung

In this paper, a computational optimized 3D reconstruction system for airborne image sequences is described and tested. The on-board system is designed to generate automatically and continuously the Digital Elevation Model (DEM) for large observation areas during the flight. The aim is to provide near real-time applications such as supporting rescue teams for prompt evaluation and timely reactions after natural disasters such as floodings, landslides or earthquakes. In addition, the proposed system enables the fast and accurate processing of 3D reconstruction for urban areas. The different optimization aspects and software-hardware methods are combined in the system as a compact solution to squeeze the potential performance of devices and to shorten the calculation time. Currently, most real-time stereo systems are applied only to small image sizes deriving a fast processing. Implementations on Graphics Processing Units (GPUs) are posed as image rendering. In this paper, a computational orientated concept for stereo processing using large images without special hardware is introduced. The data is partitioned according to the memory architecture of the programmable GPUs by parallel processing. This optimization is implemented on GPU with Compute Unified Device Architecture (CUDA). First results demonstrate the potential of the proposed real-time strategy.