Line correspondence for unknown camera positions
Date
1996
Authors
Vanderkooy, Geoffrey Edward
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The correspondence problem of matching features between images from different positions has not be solved for uncalibrated cameras in arbitrary positions. The goal of feature matching is to determine enough matches so that the camera positions can be determined. A line matching algorithm based on the cross-ratio invariant from projective geometry and the absolute intensities of the line border regions is developed. Practical issues of data reduction, invariant cross-ratio identification and signal-to-noise ratio improvement are discussed. The algorithm is tested on synthetic and real image data. The signal-to-noise ratio for typical images turned out to be much lower than expected. Although insufficient for recovery of camera positions, line matches with very low error are achieved for camera rotations about an object of 45 degrees.