An Automatic Method for Adjustment of a Camera Calibration Room

Ehsan Khoramshahi, Eija Honkavaara, Tomi Rosnell

Forskningsoutput: Kapitel i bok/rapport/konferenshandlingKonferensbidragVetenskapligPeer review

Sammanfattning

Machine vision is becoming a part of our everyday lives. Measurements by cameras are a fundamental part of this automation process. During past few years revolution has taken place in the development of the camera technologies. High resolution multi-spectral sensors with highly accurate lenses are more accessible than any time. Such technologies also exist in many everyday use devices such as smart phones, DSLRs and tablets. In order to position the measurements by cameras, the geo-referencing is required. The basis of geo-referencing is the accurate calibration of internal parameters of a camera which involved focal length, principal point and lens distortions estimation. The determination of geometric characteristics of systems is becoming essential because different sensors have very different operating principles. Finnish Geospatial Research Institute (FGI) is investigating new aspects of camera calibration by developing a new calibration room and calibrating novel cameras. The essence of accessing a camera calibration room to facilitate estimating a camera’s internal-parameters suggests developing a fast and reliable approach for automatic calibration-room adjustment. Such a room could be employed to calibrate planar, multi-planar, fish-eye or multi fish-eye cameras. By the context of “automatic target adjustment” we mean estimating fixed targets with good spatial coverage in the local coordinate system of a calibration room with the possibility of automatic reading of their image observations in a short time (10-30 min). In this work we demonstrate such development for a calibration room by employing fixed-location coded-targets designed for good visibility in short range (30cm-4m). Our easy-to-read coded-targets ensure automatic and robust measurements of ties. A minimum-constraint bundle block-adjustment is employed to optimize the cost function and propagate errors from observations to unknowns. Our results demonstrate that the locations of targets are estimated with high precision (0.1-0.4 mm std.) in a distance of 30cm-4m. All object points are re-projected for residual investigation; sub-pixel image residuals are consequently observed for all coded-targets which ensure the quality of adjustment. Finally, repeated measurements (6 data sets) confirm the correctness of the proposed (95%) positional confidence intervals of the object points.
Originalspråkengelska
Titel på gästpublikationFIG Working Week 2017, Proceedings : Plenary and Technical Sessions
FörlagInternational Federation of Surveyors, FIG
Utgivningsdatum1 maj 2017
ISBN (elektroniskt)978-87-92853-61-5
StatusPublicerad - 1 maj 2017
MoE-publikationstypA4 Artikel i en konferenspublikation
EvenemangInternational federation of surveyors working week - Helsinki, Finland
Varaktighet: 29 maj 20172 jun 2017

Vetenskapsgrenar

  • 113 Data- och informationsvetenskap

Citera det här