A fundamental goal of work in recognition is to discover easily-computed visual features which are efficient indices of members of the class which is to be recognized. The hypothesis behind work in motion-based recognition is that features describing motion in the input can be efficient indices for large classes of objects and activities of interest to the computer vision and biological vision communities. Motion-based recognition encompasses the recognition of objects, object movements, situations, etc., when motion information is used as the primary cue for recognition. For example, optic flow can be used to recognize an imminent collision situation, without any prior recognition of objects or how they are moving. Similarly, objects could be recognized using characteristic motion parameters without prior determination of shape, texture, etc.