Improving the Accuracy of Di erential{Based Optical Flow Algorithms

This work deals with the design of the low{level stages of any di erential{ based optical ow algorithm. The problem is the accurate estimaton of the spatio{temporal derivatives of the moving image, to be used in the solution of the gradient constraint equation. We claim that the \traditional" techniques for the partial derivatives estimation can be improved by i) exploiting all the information provided by both elds within a frame in the case of interlaced scanning systems, ii) adopting prolate spheroidal lters instead of gaussian lters in order to get rid of noise and aliasing, and iii) using larger size di erentiators, designed by a weighted least squares technique.