Digitising, pre-processing and photogrammetric processing of historical aerial photographs for the production of high resolution orthomosaics and the study of geohazards

The acquisition of aerial photographs for cartographic applications started in the 1930s, and more intensively after World War II. Such old, often panchromatic, imagery offers metre to sub-metre scale spatial resolution over landscapes that have significantly evolved over the decades. Before the appearance of the first digital aerial camera systems at the end of the 20^th Century, surveys were performed with analogue metric cameras, with images acquired on films or glass plates and, next, developed on photo papers. In Europe and North America, several institutions hold unique collections of historical aerial photographs having local, national and, in some cases, colonial coverages. They represent invaluable opportunities for environmental studies, allowing the comparison with today’s land use land cover, and the analysis of long-term surface displacements.

Initially, the photogrammetric processing of analogue aerial photographs would require expensive equipment, specialised operators, and significant processing time. Thanks to the digital revolution of the past two decades and the development of modern digital photogrammetric approaches, the processing of this type of image datasets has become less cumbersome, time consuming and expensive, at least in theory. In practice, this is more complex, with digitising and processing issues related to the ageing and quality of conservation of the aerial photographs, the potential distortions created during the digitising process, and the lack of ancillary data, such as, flight plans, and camera calibration reports. The limited overlap between photographs, typically 60 % and 10-20 %, along-track and across-track, respectively, make their processing with Structure-from-Motion Multi-View Stereo (SfM-MVS) photogrammetry poorly reliable to accurately reconstruct the topography and orthorectify the images. Given the fact that some collections reach up to millions of historical aerial photographs, the digitising, pre-processing, and photogrammetric processing of these images remain a challenge that must be properly tackle if we would like to ensure their preservation and large-scale valorisation.

In the present work, we describe the mass-digitising, digital image pre-processing and photogrammetric processing approaches implemented at the Royal Museum for Central Africa (RMCA, Belgium) to preserve and valorise the collection of >320,000 historical aerial photographs conserved in this federal institution. This imagery was acquired between the 1940’s and the 1980’s, over Central Africa, and mostly D.R. Congo, Rwanda and Burundi. For the digitising, a system of parallelized flatbed scanners controlled by a Linux computer and a self-developed software allows speeding-up the scanning of the entire collection in only few years. A series of Python scripts were developed and combined to allow a swift pre-processing that prepare and optimise the digitised images for photogrammetric processing. Finally, a SfM-MVS photogrammetric approach adapted to historical aerial photos is used. Examples of application for geo-hydrological hazards studies in the western branch of the East African Rift are shown.