Velocity as a Cue to Segmentation

Abstract
Although motion is one of the most fundamental aspects of the visual world, its use has not been widely explored in the field of scene description. This correspondence describes a program that demonstrates the feasibility of segmenting a scene into regions on the basis of velocity. A velocity measure is obtained by determining the amount of displacement of a reference feature between two pictures of the same scene. The pictures are taken only a short time apart so that the reference features can be correlated between the pictures on the basis of spatial position alone. This velocity measurement is associated with absolute reference points superimposed over the scene. It is assumed that all parts of an object have the same velocity measurement. Thus the velocity measurements associated with the reference points provide a means of classifying them as ``belonging'' to the same object. Points ``inside'' moving objects fall into either of two velocity measurement classes, which are easily correlated. Since the velocity of an object is independent of its color, a scene is segmented into regions that contain complete objects even if the objects are not monochromatic.