Historical Images for Surface Topography Reconstruction Intercomparison eXperiment (Historix)

Historical film-based images, acquired during aerial campaigns since the 1930s and from satellite platforms since the 1960s, provide a unique opportunity to document changes in the Earth’s surface over the 20th century. Yet, these data present significant and specific challenges, including complex distortions in the scanned image and poorly known exterior and/or interior camera orientation. In recent years, semi- or fully-automated approaches based on photogrammetric and computer vision methods have emerged (e.g., Knuth et al., 2023; Dehecq et al., 2020; Ghuffar et al., 2022), but their performance and limitations have not yet been evaluated in a consistent way.

The ongoing “Historical Images for Surface Topography Reconstruction Intercomparison eXperiment (Historix)” project aims at comparing existing methods for processing stereoscopic historical images and harmonizing processing tools.

Within this experiment, participants are provided with a set of historical images and available metadata and invited to return a point cloud and estimated camera parameters. We selected two study sites near Casa Grande, Arizona, and south Iceland, chosen for their  good availability of historical images and variety of terrain types. For each site, we selected 3 sets of film-based images acquired in the 1970s or 80s, overlapping in space and time: aerial images with fiducial marks from publicly available archives and 2 image sets from the American Hexagon (KH-9) reconnaissance satellite missions acquired by the mapping camera (KH-9 MC) and panoramic camera (KH-9 PC). The submitted elevation data will be cross-validated across different image sets and participant submissions, as well as against reference elevation data over stable terrain. The spread in the retrieved elevations will be analysed with respect to image type, terrain type and processing methods to highlight the strengths and limitations of the different approaches.

In this presentation, we will introduce the experiment design, the selected benchmark dataset, the current methodologies and the preliminary results of the intercomparison. Finally, we will present some of the open-source code that exist or are being developed to process historical images.