Relative Orientation of Stereo-Pair Using 2D Projective Singular Correlation:Applications on Video Images and Scanned-Aerial Photographs

Abstract

Although the mathematical model of 2D projective singular correlation for relative orientation was introduced to the photogrammetric community more than fifty years ago, it does not penetrate the deep practice of analytical and digital photogrammetry. On the other hand, this model is widely accepted and used in the computer vision community, which is known as the fundamental matrix approach for relative orientation. The attractiveness of this model stems from two facts. First, it is a linear model. Second, it does not require any prior knowledge about the camera parameters. In other words, it can handle uncalibrated cameras. This paper offers a fresh look for the use of 2D projective singular correlation in the relative orientation of stereo-pair. Moreover, this paper argues the case for the use of this model in the daily practice of photogrammetry; and this is through the practical demonstration on rich data sets. In particular, this paper provides critical evaluations and new insights for this model in terms of its computational procedures, least-squares random errors modelling, external accuracy checking, and practical issues of implementation. For example, it presents a general formulation for external accuracy checking that uses the epipolar distance as a metric for the quality of the solution. This model is tested on three stereo-pairs that were obtained from handheld video camera, aerial video camera, and scanned aerial photographs. In all tests, subpixel accuracy was achieved from external checking of the average distances between the epipolar-lines and their corresponding conjugate points.