REDUCING STRUCTURE-FROM-MOTION - A GENERAL FRAMEWORK FOR DYNAMIC VISION PART 2 - IMPLEMENTATION AND EXPERIMENTAL ASSESSMENT

A number of methods have been proposed in the literature for estimatin g scene-structure and ego-motion from a sequence of images using dynam ical models. Despite the fact that all methods may he derived from a ' 'natural'' dynamical model within a unified framework, from an enginee ring perspective there are a number of bade-offs that lead to differen t strategies depending upon the applications and the goals one is targ eting. We want to characterize and compare the properties of each mode l such that the engineer may choose the one best suited to the specifi c application. We analyze the properties oi filters derived from each dynamical model under a variety of experimental conditions, assess the accuracy of the estimates, their robustness to measurement noise, sen sitivity to initial conditions and visual angle, effects of the bas-re lief ambiguity and occlusions, dependence upon the number of image mea surements and their sampling rate.