Multi-Camera Calibration

Abstract

3-D display technology is under development, and it has not yet become common among people. In this project, we use light field display to create a multi-view image. This type of display makes it possible to visualize objects from different perspectives on the screen by changing the viewing angle. The display does not require additional equipment from the viewer, unlike the older 3-D displays where the viewer can feel the depth of the image by using the polarizing 3-D glasses. The aim of this study is not to get into the technology behind the light field displays but do research on how to create tools to provide proper content for light field displays. In more detailed, this study focuses on methods in multi-camera calibration where the multi-view image data has to be parametrized accurately to know where and how these images have been obtained. Such requirements need accurate calibration.

In general, camera calibration consists of color, focus, distortions and position calibration in which the focus in this thesis is on the position calibration between cameras. In one camera calibration, the goal is to remove the distortion caused by the camera lens and components to which errors have occurred during the manufacturing process. Firstly, the distortions caused by the lens will cause straight lines in the real world to be curved in the image. Secondly, imperfect parallelism between the camera lens and the image sensor gives the image projective impression. Thirdly, slight distortions can happen by the component errors which make the image skewed or stretched.

The calibration of multiple cameras must take into account the relative translations and rotations of the cameras. The unknown parameters describing the position and distortions of a camera are estimated by minimizing the error between the model and the actual measurements. The final result of the minimization is finally seen on the light field display where the image should not have significant artifacts. To achieve this result, accurate methods are required to calibrate the translations and rotations of the cameras.

The study provides tools for more accurate calibration and illustrates the less accurate pair-wise calibration method. The calibration method can be seen in the image quality and on that basis it is required to use better methods of calibrating a multi-camera camera rig.