Mapping matrix for perspective correction from fish eye distorted images

Abstract

Fisheye cameras are widely used in various applications like surveillance, robotic vision, automobile rear-view imaging systems etc. But the images captured by fisheye lens are hemispherical and objects away from the center of image look curved due to lens distortion. Hence it is a challenge for an end user to interpret the data correctly from these images. The solution lies in digitally processing these images to obtain distortion-free rectilinear images. The process should compensate for both fish eye and perspective distortions with minimum overhead. This paper presents a novel approach that corrects fisheye and perspective distortions from wide angle camera image. The basic geometry of hyperbola is used for fisheye correction; while perspective correction is done using a standard homography based transform. Both these correction techniques are combined to form a mapping matrix wherein each output pixel in the corrected image is related to the corresponding pixel in the input distorted image. This approach using a mapping matrix reduces the overall processing time and hence can be extended for video correction. This method finds application in vision based parking assist system; in order to provide a perspective view of the parking area from the input wide angle images.