This corner detector is designed to select the best features for tracking inside of a Kanade-Lucas-Tomasi (KLT)
feature tracker . It selects features which have a low self similarity in all directions. The output
is an intensity image which indicates how corner like each pixel is. Point features are extracted from the
feature intensity image using
An important consideration when using this detector in some applications is that the selected corners will not lie on the actual corner. They lie inside of the corner itself. This can bias measurements when used to extract the location of a physical object in an image.
Detected features are square in shape. The feature's radius is equal to the square's width divided by two and rounded down.
 Jianbo Shi and Carlo Tomasi. Good Features to Track. IEEE Conference on Computer Vision and Pattern Recognition, pages 593-600, 1994