A Calibration Method for the Generalized Imaging Model with Uncertain Calibration Target Coordinates

Abstract

The developments in optical metrology and computer vision require more and more advanced camera models. Their geometric calibration is of essential importance. Usually, low-dimensional models are used, which however often have insufficient accuracy for the respective applications. A more sophisticated approach uses the generalized camera model. Here, each pixel is described individually by its geometric ray properties. Our efforts in this article strive to improve this model. Hence, we propose a new approach for calibration. Moreover, we show how the immense number of parameters can be efficiently calculated and how the measurement uncertainties of reference features can be effectively utilized. We demonstrate the benefits of our method through an extensive evaluation of different cameras, namely a standard webcam and a microlens-based light field camera.

Deflectometry for specular surfaces: an overview

“…In precision DM, generic and model-free approaches have been shown to reduce measurement errors even for paraxial imaging, so the topic has attracted a lot of attention [219][220][221][222][223][224][225][226][227][228][229][230][231][232][233][234].…”

Model-independent light field reconstruction using a generic camera calibration

“…Hence, the lack of flexibility and precision has led to the development of new camera models. Cameras are now described as generic imaging systems, which are independent of the specific camera type and allow high-precision calibration [2,6,13]. An imaging system is modeled as a set of photosensitive pixels, with all other optical elements not explicitly modeled but represented by a black box.…”

“…The calibration of the entire bundle of rays belonging to the camera is then usually realized with the help of a minimization of the ray projection error ε i . Meaning, the sum of the Euclidean distances of the rays l i to known reference points p ik in space is minimized [13]: arg min…”

“…A minimization of the commonly used ray reprojection error on pixel level is often not possible, because most generic models do not support a direct projection of points in space onto the pixel plane. See [13] for details. The advantage of this type of calibration is that there is no longer one global model that has to describe the camera over the entire pixel plane.…”

“…As a consequence, in recent years, new camera models were proposed that describe the camera as a generic imaging system. They are able to model the ray of each pixel individually and thus allow for a high-precision calibration [6,13]. However, the biggest disadvantage of the common light field reconstruction methods is that they are only applicable for a single type of camera, e. g. microlensbased light field cameras whose microlenses are exactly focused onto the sensor.…”