The methodology of the archival aerial image orientation based on the SfM method

Authors

  • Artur Karol Karwel Warsaw University Technology
  • Jakub Markiewicz Warsaw University of Technology

DOI:

https://doi.org/10.55627/smla.001.02.0015

Keywords:

aerial images, archival images, bundle adjustment, RANSAC, reprojection error, Structure-from-Motion, SIFT detector

Abstract

Nowadays, archival images find increasingly finding their way into geospatial applications, namely, among others, multi-temporal analysis, documentation reconstruction or change detection. It is, therefore, necessary to determine the images' external orientation elements that allow the images' position to be reconstructed in the assumed reference system. This paper aims to present a methodology for the extended evaluation of the automatic orientation process of archival images based on the commonly used Structure-from-Motion (SfM) approach. The work carried out presents: (1) the influence of parameter selection on the accuracy, number and distribution of tie points in the descriptor matching process at the pairwise image bundling stage using the descriptor matching approach together with the use of Random sample consensus filtered triangulation (RANSAC), (2) analyses of the reciprocal orientation quality of the images on detected points (control points) in the bundle adjustment process using simultaneous verification of the matching quality on check points, and (3) analysis of the external orientation accuracy. Points detected and matched using the SIFT algorithm on archival images of a fragment of Warsaw from 1986, 1994, and 2014 were used as reference data. A comparative analysis of the obtained results with the data obtained using the algorithms implemented in the Agisoft Metashape software (standard approach) shows that the relative orientation reprojection RMSE is about 4 time better, and detected points are even more robust.

Downloads

Published

2022-12-01

How to Cite

Karwel, A. K., & Markiewicz, J. (2022). The methodology of the archival aerial image orientation based on the SfM method. Sensors and Machine Learning Applications , 1(2). https://doi.org/10.55627/smla.001.02.0015