Abstract:
|
This report aims to compare the state of the art Omni‐directional image unwrapping
methods/algorithms in order to develop a fast and correctly transformed panoramic
image. Omni‐directional vision systems consist of a perspective camera adapted to a
hyperbolic mirror. It is known that the distribution of resolution from hyperbolic
mirrors is better than the one obtained with systems that use spherical mirror or
parabolic mirrors without the orthographic lens. There are several methods to unwrapping Omni‐directional image. The most
common method used, to obtain panoramic images, is by directly transforming the
polar coordinates into rectangular coordinates. The panoramic image is created by doing an inverse mapping of the pixels, for each pixel in the panoramic image a correspondence pixel must exist in the
Omni‐directional camera image. An improved version of this method can be done by including the mirror equation in the unwrapping process, so the vertical
distortion is null. The circular features from an Omni‐directional image after unwrapping become linear and image tracking/detection algorithms are simplified. Nevertheless, the image must be correctly transformed. Thus, to speed up the process of unwrapping the image in real time, instead of calculating the coordinates at all‐time, lookup tables are used to know which pixel from the Omni‐directional image have a direct correspondence to the panoramic image. |