Archival Aerial Images Georeferencing: A Geostatistically-Based Approach for Improving Orthophoto Accuracy with Minimal Number of Ground Control Points

• Main Authors: Manuela Persia, Emanuele Barca, Roberto Greco, Maria Marzulli, Patrizia Tartarino
• Format: Article
• Language: English
• Published: MDPI AG 2020-07-01
• Series: Remote Sensing
• Subjects: archival aerial photos; georeferencing process; geostatistical analysis; local accuracy mapping and assessment; ground control points
• Online Access: https://www.mdpi.com/2072-4292/12/14/2232

Georeferenced archival aerial images are key elements for the study of landscape evolution in the scope of territorial planning and management. The georeferencing process proceeds by applying to photographs advanced digital photogrammetric techniques integrated along with a set of ground truths termed ground control points (GCPs). At the end of that stage, the accuracy of the final orthomosaic is assessed by means of root mean square error (RMSE) computation. If the value of that index is deemed to be unsatisfactory, the process is re-run after increasing the GCP number. Unfortunately, the search for GCPs is a costly operation, even when it is visually carried out from recent digital images. Therefore, an open issue is that of achieving the desired accuracy of the orthomosaic with a minimal number of GCPs. The present paper proposes a geostatistically-based methodology that involves performing the spatialization of the GCP errors obtained from a first gross version of the georeferenced orthomosaic in order to produce an error map. Then, the placement of a small number of new GCPs within the sub-areas characterized by the highest local errors enables a finer georeferencing to be achieved. The proposed methodology was applied to 67 historical photographs taken on a geo-morphologically complex study area, located in Southern Italy, which covers a total surface of approximately 55,000 ha. The case study showed that 75 GCPs were sufficient to garner an orthomosaic with coordinate errors below the chosen threshold of 10 m. The study results were compared with similar works on georeferenced images and demonstrated better performance for achieving a final orthomosaic with the same RMSE at a lower information rate expressed in terms of nGCPs/km².