Three points in 3-space define a plane . The images of these points, together with the epipoles, provide the four correspondences necessary to compute the compatible transformation for . That is, the projective transformation such that

transfers points coplanar with . Given it is then possible to distinguish points in 3-space on either side of , using only their image projections and i.e. a binary partition of 3-space.

- The three points need not actually correspond to
images of physical points, the method can be applied
to ``virtual'' planes.
- Points contained in
*regions*of 3-space can be identified using a set of binary space partitions for the planes which bound the region.

** Original images and points**

** First and second images and corners**

** BSP 1**

** Selected planar points, on one side and on the other side**

** BSP 2**

** More selected planar points, on one side and on the other side**

Wed Apr 16 00:58:54 BST 1997