Procrustean methods in Photogrammetric Computer Vision

with FabioCrosilla and Eleonora Maset


Overview

This stream of research focuses on applications of Procrustean analysis to computer vision and photogrammetry problems. Procrustes analysis is a well known least squares technique used to directly perform transformations among corresponding point coordinates belonging to a generic k-dimensional space. Our research aims to develop new analytical tools based on Procrustean methods for solving classical Computer Vision and photogrammetric problems. In [1] we derived the solution of the Anisotropic Extended Orthogonal Procrustes Analysis (AEOPA) with row-scaling and applied it to perform the exterior orientation of one image. We also provided [2] a robust version of this algorithm based on Forward Search, which proved to be highly effective and accurate in detecting outliers, even for small data size or high outliers contamination. In [3] we formulated the point-line registration problem, which generalizes absolute orientation to point-line matching, as an instance of the AEOPA model and derived its solution. The same formulation solves the Non-Perspective-n-Point camera pose problem, that in turn generalizes exterior orientation to non-central cameras, i.e., generalized cameras where projection rays do not meet in a single point. A generalized version of AEOPA leads instead to the Procrustean solution of the classical bundle block adjustment, developed in [4]. Moreover, we introduced [5] a robust variant of the algorithm based on Iteratively Reweighted Least Squares (IRLS), that achieves reliable results also in the presence of a percentage up to 10% of outliers.

References

  1. Garro, V.; Crosilla, F. and Fusiello, A. Solving the PnP Problem with Anisotropic Orthogonal Procrustes Analysis. In Proceedings of the 2nd Joint 3DIM/3DPVT Conference: 3D Imaging, Modeling, Processing, Visualization and Transmission (3DIMPVT), pages 262-269, 2012. pdf 
  2. Fusiello, A.; Maset, E. and Crosilla, F. Reliable Exterior Orientation by a Robust Anisotropic Orthogonal Procrustes Algorithm. In Proceedings of the Workshop: 3D Virtual Reconstruction and Visualization of Complex Architectures (3D-ARCH), pages 81-86, Trento, Italy, ISPRS Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences XL-5/W1, 2013. pdf 
  3. Fusiello, A.; Crosilla, F. and Malapelle, F. Procrustean point-line registration and the NPnP problem. In Proceedings of the International Conference on 3D Vision (3DV), pages 250-255, IEEE, 2015. pdf 
  4. Fusiello, A. and Crosilla, F. Solving Bundle Block Adjustment by Generalized Anisotropic Procrustes Analysis. In ISPRS Journal of Photogrammetry and Remote Sensing, 102: 209-221, 2015. pdf  doi 
  5. Fusiello, A. and Crosilla, F. Fast and Resistant Procrustean Bundle Adjustment. In Proceedings of the XXIII ISPRS Congress, ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences , 2016. pdf