Recognizing 3-D Motion

Abstract

What is Vision? Being more precise, how should we formulate the process of vision, if our goal is to equip machines with visual capabilities? Is it suitable to follow the paradigm suggested by Marr and consider vision as the recovery or reconstruction of the shapes, material properties and motions of objects in our environment? Or, should we view the process of vision as the recognition of particular situations, objects, pattern or instances in spacetime n~levant to the task a machine is engaged in? We argue for the latter point of view, and we demonstrate our position by proving that the famous problem of ego-motion estimation, which in the past has been addressed as an application of applied mathematics techniques, can be formulated as a problem of recognition . We show that local image measurements form global patterns in the image plane. The position of these patterns is related to the three dimensional motion parameters. By locating through a simple search technique some of these patterns, which depend on only subsets of the motion parameters, the 3-D motion parameters are found . The proposed algorithmic procedure is very robust. Since only the sign of the image velocity is employed, the direction of translation and the axis of rotation can be estimated with up to 100 % error in the image measurements.