Abstract
We propose a method of computing two-dimensional-image motion that combines spatiotemporal-filtering and feature-matching schemes. In our two-stage model, several velocity candidates are first detected as the intersection of the various constraint bands, each of which represents the possible solutions on the velocity plane. Each constraint band is obtained by using the spatiotemporal filter proposed by Watson and Ahumada [ J. Opt. Soc. Am. A 2, 322 ( 1985)]. The second stage determines a unique velocity by the feature-matching scheme, in which several kinds of feature vote for the velocity candidates, and the velocity with the highest score is assigned to each pixel as the best estimate. The filtering stage reduces the computational load and the number of false matches in the matching stage by limiting the range of correspondence. The accuracy problem of the filtering stage is overcome in the matching stage. We applied our two-stage model to some natural-image sequences and obtained accurate estimations.
© 1992 Optical Society of America
Full Article | PDF ArticleMore Like This
Steven J. Nowlan and Terrence J. Sejnowski
J. Opt. Soc. Am. A 11(12) 3177-3200 (1994)
Edward H. Adelson and James R. Bergen
J. Opt. Soc. Am. A 2(2) 284-299 (1985)
R. Jasinschi, A. Rosenfeld, and K. Sumi
J. Opt. Soc. Am. A 9(11) 1865-1879 (1992)