Robust contour extraction of fisheye images for image-based virtual reality

Image-based virtual reality (VR) is a good alternative to the 3D-based VR. The image-based VR can use a surrounding panorama to represent the environment. A fisheye lens with a field-of-view larger than 180 degree is usually the best choice to capture the images for creating a surrounding panorama. The contour extraction of the valid region of fisheye images is an essential step to create the panorama. The basic scan-line based contour extraction method can not find the correct contour if there are black objects on the edge of the valid region. To overcome this deficiency, in this paper, a new scan-line based contour extraction method was proposed; it moves eight scan lines on the fisheye image to form a circumscribed octagon of the valid region; the eight vertices of the circumscribed octagon is then calculated. Based on the eight vertices, the unconstrained optimization method is used to compute the contour. Finally, the robustness of the new method is validated by the experiment.