Modify our sample code “planewarpdemo” in the matlab sample code section of our course to
generate a higher resolution output than it currently does, for example, by setting the output
(destination) image be comparable in number of rows/cols to the input (source) image. Also,
note that one deficiency of this code is that the user has to guess what the shape of the
chosen rectangle is when specifying the output. Write a new version that does not rely on user
input, and that generates a top–down view of the floor plane that is accurate up to a similarity
transformation (rotation, translation and isotropic scale) with respect to the 2D X–Y world
coordinate system in the floor plane Z=0. Hint: how can you relate ground plane X–Y
coordinates to 2D image coordinates in the source and in the destination images, given the
known camera parameters of one or both views? Explain how you are generating your top–
down view. Also, with regard to the resulting output image, what things look accurate and
what things look weird? Could this kind of view be useful for analyzing anything about the
performance of a person as they move around in the room?
