High-performance computing in UAV photogrammetry

Mapping large areas of the Earth’s surface through Unmanned AerialVehicle (UAV) imagery poses challenges in terms of data storage, computational resources and, above all, processing time. The capability to acquire large, high-resolution datasets, together with the complex photogrammetric processing (Structure from Motion) needed for accurate mapping, requires the usage of high-performance computing resources (HPC). The time-sensitive nature of applications, as e.g. disaster response and environmental monitoring, exacerbates the need for near-real-time processing. In this context, the present study introduces the implementation of a distributed photogrammetric workflow based on a divide and conquer approach and on the HTCondor software framework to exploit theIBiSCo-ReCaS-Bari HTC/HPC cluster. Performance tests of the workflow, which leverages computing parallelism, hybrid bundle adjustment and GPU usage, for image matching and dense correlation, show a significant reduction in processing time for large UAV image datasets with up to 83% improvement over state of the art approaches.