Automated Camera Calibration for Augmented Reality

Background

Camera calibration is an important field in computer vision. It is the process whereby we calculate the perspective projection of any 3D point in the world to it's 2D position in the camera image. An excellent resource including software is provided by Jean-Yves Bouguet:

Jean-Yves Bouguet's home page

Camera calibration is a vital component of systems that provide 3D reconstruction and also those that aim to mix virtual and real structures with correct alignment - a process known as augmented reality.

Project Details

This project will examine a number of ways that this process might be improved upon. Firstly, the implementations available on Bouguet's page involve manual marking of the corners of the calibration pattern. If a large number of images are taken (for example a video sequence) this can be rather laborious. The identification of the calibration pattern could be automated in a number of ways. For a video sequence the corners can be marked in the first image only and be tracked in subsequent images. It may also be possible to automate the corner identification in any individual image by a number of techniques (Hough transform, spatial Fourier transform, grid extrapolation). The accuracy of the calibration itself may be improved upon by using further information in each of the images. At present just the detected corner points are used. Use of full linear features or comparison with a rendering of the calibration pattern should increase the amount of image information used and hopefully improve accuracy. Finally, an alternative parametrisation of the camera is proposed that reduces the interdependence of the internal and external parameters. This should certainly result in improved error estimation and may also increase the accuracy of the overall calibration. Existing libraries can be used for some parts of the project, but algorithm development by the student will be required. Software will be written in either Matlab or C++ with OpenGL for 3D rendering. If successful the output of this project could be incorporated into the freely available software which would significantly raise the student's international profile.
Eddie Edwards
Last modified: Wed Oct 17 19:10:37 BST 2007