Plenary speech III: visual appearance modeling and perception with retinal and cortical signal processing

Summary form only given. This talk focuses on the problem of shape estimation using multiple views from a land based mobile robot equipped with CCD cameras and a laser range finder that can compute the range of a target along a fixed horizontal plane. The talk surveys the problem of shape estimation from optical flow of points, lines and algebraic curves and emphasizes the fusion of camera and range sensors. Inspired from Neuroscience, the talk also introduces the role of cortical flow to the problem of encoding visual input signals and subsequently decoding these inputs using maximum likelihood estimates. To end the talk, we model the appearance of an object using principal component analysis and argue the role of appearance dynamics as an alternative to optical flow based algorithms.