Several models of statistical estimation of motion from visual input a
re derived and analyzed theoretically and experimentally. We study a w
ide variety of models, ones that use least squares and ones that use m
aximum likelihood, with several different assumptions (dependent and i
ndependent noise, isotropic and nonisotropic noise), spherical and pla
nar image surfaces, and different preprocessing (one based on correspo
ndence and one based on disparity). We do all this analysis using only
a few fundamental concepts from statistical estimation, so the relati
ve merits and shortcomings of all the methods become evident. The expe
rimental results provide a quantitative measure of these merits. (C) 1
994 Academic Press, Inc.