Structured Light Based Camera Calibration Methods for Computational Vision and Video Systems

Abstract

Modern computer vision, robot vision and computational video systems require the normalization and compensation for the image sensor and lens distortions. It is definitely a crucial part in stereo-, omnidirectional- and other overlapping view multi-camera systems.

At present, most conventional calibration methods use checkerboards and mathematical models [1,2] to calculate the extrinsic and intrinsic parameters, such as focal length, principal point and so on. Due to the popularity of toolboxes available in Matlab and OpenCV, checker board patterns are currently very much used for camera calibration purposes. When wide angle fish eye lenses are used, the accuracy of the camera calibration in the off-center and corner regions, is often difficult to achieve.

In this paper, we introduce a look-up table based method to undistort captured images. The method makes use of Gray code structured light patterns and introduces the notion of mapping files to create look-up table. There are two methodologies to acquire the LUT: the forward mapping and the inverse mapping. The key idea is to find the correspondences between the original pixels and the projected pixels. The effectiveness and high accuracy is confirmed by the comparison of the two methods.