Historical Structure from Motion (HSfM): Automated processing of historical aerial photographs for long-term topographic change analysis

• DOI: 10.1016/j.rse.2022.113379
• Publication Date: 2022-12-07T09:54:11Z
• Authors: Friedrich Knuth; David Shean; Shashank Bhushan; Eli Schwat; Oleg Alexandrov; Christopher McNeil; Amaury Dehecq; Caitlyn Florentine; Shad O’Neel
• Keywords: Orthophoto; Structure from Motion; Aerial photography; Bundle adjustment; Geolocation

Precisely measuring the Earth's changing surface on decadal to centennial time scales is critical for many science and engineering applications, yet long-term records of quantitative landscape change are often temporally geographically sparse. Archives scanned historical aerial photographs provide an opportunity augment these with accurate elevation measurements that capture state Earth surface. Structure from Motion (SfM) photogrammetry workflows produce high-quality digital models (DEMs) orthoimage mosaics images, but time-intensive tasks like manual image preprocessing (e.g., fiducial marker identification) ground control point (GCP) selection impede processing at scale. We developed automated method process images generate self-consistent series high-resolution (0.5–2 m) DEMs orthomosaics, without GCP selection. The relies SfM correct camera interior exterior orientation a robust multi-stage co-registration approach using modern reference terrain datasets geolocation refinement. demonstrate North American Glacier Aerial Photography (NAGAP) archive collected between 1967 1997. present results two sites variable photo acquisition geometry overlap — Mount Baker South Cascade in Washington State, USA. corrects initial position errors several kilometers produces accurately georeferenced, orthoimages, regardless configuration, geometry, characteristics, DEM properties. average RMS reprojection error following bundle adjustment optimization was 0.67 px (0.15 261 contributing 10 final 1970 1992 Baker, 0.65 (0.13 243 18 individual 1997 Glacier. relative accuracy values stacks 0.68 m 0.37 Our products have reduced systematic improved compared generated Final measurement precision ∼0.7–1.0 over 30-year period, enabling study processes rates as low ∼1-3 cm/yr. potential this scalable rapidly archives imagery deliver new insights geodetic processes.