Motion detection - a neural model and its implementation

A model for motion detection is presented. In this approach, motion is viewed as a stable pattern propagating over the image - a technique that makes the model unusually insensitive to noisy input sequences. The model is based on a neural network and requires spontaneous activation of the nodes in order to be operative. The model has been tested in computer simulations with over 300 000 nodes and more than 6 million connections. A compilation of results from these simulations is also presented