Omnidirectional robot vision : correction and restoration of fish-eye lens images

The purpose of this study was to describe techniques for geometric and blurring restoration of fish-eye lens images. The history of extreme wide angle imaging is briefly reviewed, with most emphasis on previous experiments and theories developed on the fish-eye lens subject. Fish-eye lens images, which can include a 2π steradian field of view, appear with an unavoidable inherent distortion which can be corrected with image processing techniques. A method for geometric restoration of such distorted images is derived, based on a simple pinhole camera model, and its resolution is examined. In a more detailed examination of the model, it is shown that the fish-eye image can be transformed from a space variant system into a space invariant one. An example is given of the application of this transformation to "undistort" a photograph made through a fish-eye lens. After geometric restortion, the image can be further enhanced by applying deblurring techniques to improve resolution. An example is given of the results of deblurring applied to the geometrically restored image from the previous example. Finally, an image correction system that would interact with a computer was suggested for hardware implementation.