Improving FOSS photogrammetric workflows for processing large image datasets

Background: In the last decade Photogrammetry has shown to be a valid alternative to LiDAR techniques for the generation of dense point clouds in many applications. However, dealing with large image sets is computationally demanding. It requires high performance hardware and often long processing times that makes the photogrammetric point cloud generation not suitable for mapping purposes at regional and national scale. These limitations are partially overcome by commercial solutions, thanks to the use of expensive and dedicated hardware. Nonetheless, a Free and Open-Source Software (FOSS) photogrammetric solution able to cope with these limitations is still missing.

Methods: In this paper, the bottlenecks of the basic components of photogrammetric workflows -tie-points extraction, bundle block adjustment (BBA) and dense image matching- are tackled implementing FOSS solutions. We present distributed computing algorithms for the tie-points extraction and for the dense image matching. Moreover, we present two algorithms for decreasing the memory needs of the BBA. The various algorithms are deployed on different hardware systems including a computer cluster.

Results and conclusions: The usage of the algorithms presented allows to process large image sets reducing the computational time. This is demonstrated using two different datasets.