Project 4, Recognition/tracking
In this project, you will implement a feature detector.
You are given an image of an object, and a movie in which there is both
camera and object motion. The output from your program should be another
movie where the object is marked in each frame in which it appears.
(e.g. by coloring the book red, or otherwise.) Notice that the book is
sometimes only partially visible.
You may implement this project in any way in which you like. Notice,
however, that there is substantial radial distortion in the camera,
meaning that even though the object you are searching for is
approximately planar, simply doing an exhaustive search for the object is
unlikely to be successful. We suggest a feature detection approach
along the lines of the following paper:
D. Lowe, "Object Recognition from Local ScaleInvariant Features", ICCV,
Kerkyra 1999
Which may be downloaded from here
thebook_cropped.bmp
themovie.avi
Describe clearly the approach that you have taken. If your algorithm
fails in some situations, explain as precisely as you can what those
situations are, why it fails, and what improvements you think might fix
the problem.
Web Accessibility