A model of motion computation in primates

A neural network model of motion-field computation in the visual system of primates is proposed. The model assumes that local velocities are computed first in the primary visual cortex using information available from the sustained and transient pathways originating in the retina. A variation of the optical flow of algorithm B.K.P. Horn and B.G. Schunck (1981) and direction-selective representation of velocities found in the primary visual cortex and middle temporal area are used to solve the aperture problem. The resulting network has its connection matrix independent of measured data. The authors adopt D.H. Hubel and T.N. Wiesel's model (1977) of orientation selectivity, and generalize D. Marr and S. Ullman's XYX model (1981) of direction selectivity to compute local velocities. Preliminary fusion of edge and motion information is also proposed. Results of computer simulation, including the E.H. Adelson and J.A. Movshon experiment and the phenomenon of motion capture, are presented.<>