The last decade has witnessed the development of numerous models for computing an observer's direction of motion from the optic flow generated by that observer's motion through a stationary scene. A few of these model physiological processing (Perrone & Stone, 1994; Lappe & Rauschecker, 1993; Royden, 1997), i.e. they are based on the physiological properties of neurons in the primate visual cortex. All of the physiological models work under the assumption that the observer is moving through a stationary world. However, in real situations observers are often faced with a scene containing multiple moving objects which must be localized and identified and their direction of motion determined. Psychophysical studies (Warren & Saunders, 1995; Royden & Hildreth, 1996) show that people judge their heading well in the presence of a single moving object, showing a small bias in their heading judgments only when the object crosses the observer's path. There has been a small amount of research on detecting a moving object in a flow field (Royden et al., 1996, Royden et al., 2001), however much remains to be discovered. What are the conditions and visual cues that allow an observer to distinguish a moving object from the other parts of the visual field? How does the visual field segment moving objects and process their 3D direction of motion in the presence of an optic flow field? We have much to learn about how the primate visual system detects and processes object motion in the context of an optic flow field.

Recent results in my lab suggest that a model (Royden, 1997) based on the motion opponent properties of neurons in the Medial Temporal visual area (MT) of primates shows similar behavior to humans when tested on scenes containing a single moving object. When a moving object is present and crosses the observer's path, human observers show a small bias in heading judgments. This bias is in the direction of object motion for an object moving laterally with respect to the observer (Royden & Hildreth, 1996), and is in the direction of the object's focus of expansion when the object moves in depth toward the observer (Warren & Saunders, 1995; Royden & Hildreth, 1996). When tested on simulated flow fields for conditions identical to those tested in the psychophysical experiments, the model shows a bias in the same direction and of similar magnitude to that of human observers. This suggests that these motion opponent operators may be responsible for the perceived bias in heading judgments.

In the future, this model can be extended to detect moving objects in the scene. Hildreth (1992) showed that one can localize a moving object using the difference vectors generated at the border between the object and the stationary scene. These vectors will be inconsistent with the pattern of difference vectors generated by observer motion and thus will signal the presence of a moving object. Once located, image velocities within the object itself can be used to determine the object's 3D direction of motion. A combination of psychophysical studies and modeling can be used to create a physiological model that processes the motion for both the observer and object, leading us to a better understanding of higher order visual processing of motion in the visual system.

References:

Lappe, M. & Rauschecker, J. (1993). Neural computation 5: 374

Perrone, J.A. & Stone, L.S. (1994). Vision Research , 34, 2917-2938.

Royden, C. (1997) Journal of the Optical Society of America. A, 14 : 2128

Royden, C.S. and Hildreth, E.C. (1996) Perception & Psychophysics 58: 836-856.

Royden, C.S., Wolfe, J.M., Konstantinova, E. and Hildreth, E.C. (1996) Invest. Ophthalmol & Vis. Sci. 37: 299.

Royden, C.S., Wolfe, J.M. and Klempen, N. (2001). Perception and Psychophys. In press.

Warren, W.H. & Saunders, J.A. (1995) Perception , 24, 315-331.