A Pragmatic Approach for Lens Distortion Correction from a Fictitious Image

Session

Photogrammetry

Full Paper Review

Yes

Authors

Naa Dedei Tagoe, University of Cape Town, South Africa, tgxnaa001@myuct.ac.za
Heinz Ruther,Zamani Project, South Africa, heinz.ruther@uct.ac.za
Julian Smit, University of Cape Town, South Africa, julian.smit@uct.ac.za

Abstract

The paper offers a pragmatic alternative to the mathematical modelling of lens distortion. The method was developed as a proof of concept for application in the photogrammetric restitution of objects and buildings from photographic panoramas. The method quantifies the distortion characteristics of a lens-camera combination by comparing near distortion-free fictitious image coordinates with real (distorted) image coordinates.

Based on the differences between fictitious and real image coordinates, a distortion matrix, equivalent to a look-up table, is created which describes the distortion characteristics of the camera-lens combination. The distorted images are subsequently rectified using a backward pixel mapping strategy. As a check of the effectiveness of the proposed technique, a best-fit line function is used to verify that a straight line in object space appears as straight line on the image after rectification. The standard deviations before (180.6µm equivalent to 28 pixels) and after (17.5µm or approximately 3 pixel) fitting the best-fit line indicate that the lens distortion has been minimised.

To further characterise the performance of the proposed method, a subset of the control points was used to form the distortion matrix. The known locations of the remaining randomly selected control points were compared with their interpolated values. It was observed that the denser the control points the more precise the interpolated values indicating that the proposed method is dependent on the distribution and density of control points.

View This Paper