A network for motion perception

A locally connected artificial neural network based on physiological and anatomical findings in the visual system is presented for motion perception. A set of velocity selective binary neurons is used for each point in the image. Motion perception is carried out by neuron evaluation using a parallel updating scheme. Two algorithms, batch and recursive, based on this network are presented for computing the flow field from a sequence of monocular images. The batch algorithm integrates information from all images simultaneously by embedding them into the bias inputs of the network, while the recursive algorithm uses a recursive-least-squares method to update the bias inputs of the network. Detection rules are also used to find the occluding elements. Based on information on the detected occluding elements, the network automatically locates motion discontinuities. The algorithms need to compute the flow field at most twice. Hence, fewer computations are needed and the recursive algorithm is amenable to real-time applications