Camera self-calibration with known camera orientation

Autori

• Jan-Michael Frahm

Parametri

• Computer, IT, programmazione

Maggiori informazioni sul libro

A fundamental challenge in computer vision is reconstructing a three-dimensional static scene from uncalibrated images. This has applications in robotics, architecture, biometrics, gaming, and film. Most reconstruction methods rely solely on camera images, estimating camera motion alongside reconstruction. However, if intrinsic camera calibration (focal length, aspect ratio, principal point) is unknown, the resulting geometry is projectively skewed, necessitating calibration for a meaningful Euclidean reconstruction. Self-calibration, the estimation of camera calibration from image data, is a significant research area. Relying only on scene images presents issues due to sensitivity to measurement noise and degenerative camera motions. The main contribution of this research is the development of self-calibration methods for cameras with varying intrinsics, utilizing camera orientation as an additional information source. One approach requires a purely rotating camera, while the other accommodates freely moving cameras, leading to linear methods that overcome limitations on varying intrinsic parameters. Both approaches compute self-calibration from an image triplet by leveraging relative orientations, with the requirement for known relative orientation being a minor limitation in many applications. These methods can enhance systems equipped with rotation sensors, such as video conferencing, surveillance, and automotive safety. T

Lingua: Inglese

Pubblicato: 2005

(In brossura)