Transformation of omni directional image into perspective view

Department of Computer Graphics and Multimedia (DCGM)

Igor Potúček

Program description

You can transform video sequence or single images. Opened video or image is displayed in the higher image window. You must select the center of projection by mouse clicking. Then you can adjust size of the inner and the outer radius of the transformed circle. Setup of the rotation serves to shift the perspective image beginning.  Another track bar serves to setup the image overlapping. You can also change output image size. The program output can be video sequence or single transformed image.

Method description

An image obtained using the hyperbolic mirror can be transformed into a standard perspective image. It is used simple transformation that presumes linear pixel distribution along the radius direction. The coordinates of the panoramic view are   and . We must transform these coordinates to omnidirectional image. The real world elements are projected on a cylinder whose radius is equal to d. The Axis of the cylinder is identical to the mirror and camera axis. The horizontal size of the panoramic view is a perimeter of cylinder. The coordinates of the video camera image   are the following:

  (2.1)
  (2.2)
where angle .

The calculated pixels in the camera image do not correspond “one to one” to the pixels on the projected image so sub pixel anti-aliasing methods should be used. In this case, it is sufficient to use weighted average of neighbor pixels as the size of the output pixels is comparable to the input.

Fig. 1: Transformation into panoramic view.

It is suitable to crop part of image that contains the center of the omnidirectional view. This usually displays part the image of the camera itself anyway.

Original and unwrapped image

Download

Image Transformer (500kB).