Omnidirection image restoration based on spherical perspective projection

Because of its advantage of panoramic sight with a single compact visual scene, omni-directional vision technology has extreme application value in many fields. One of the most effective ways to establish omni-directional vision system is using fisheye lens. However it brings a strong unavoidable inherent distortion while it provides hemispherical field of view. But it can be corrected by some image processing techniques. A method for correction of such distorted image is derived in this paper. After introducing two kings of perspective projection model briefly, we will transform the fisheye image to perspective projection image by using spherical perspective projection model and establish a mathematic transform model from fisheye image to perspective projection image. In addition, a simple method to find out the optic center and the radius of the projection sphere from the property of the fisheye image which needed in rectifying the fisheye image will be derived. Usually the algorithm is implement by software on PC, but its processing speed is too slow for some applications, which need the real time image information. Considered these, we implement the algorithm of rectification on FPGA to make the processing speed to fit the requirement of the real time application. The experiments demonstrate the correctness and effectiveness of the model and the results are satisfactory.