Visual motion detection based on a cooperative neural network architecture
Robert Pallbo · Lund University Publications (Lund University) · 1993
A neural network architecture for visual direction detection is proposed. The approach assumes a continuous flow of visual stimuli as input. The output of the network will as a consequence be a continuous flow as well. Each node in the network signals when motion is occurring at their position. Other nodes make use of this information in their computations. This allows, from a computationally viewpoint, rather simple nodes. When a node detects a motion, this detection is propagated through the network in the direction of the motion. Such a propagation is easy to construct. The initial detection of a motion is carried out by spontaneous activity among the nodes. Hence, no motion detection is carried out by the nodes, but are an emergent property of the collaboration in the network. In this paper, the model is presented and results from a computer simulation of the process is discussed. Related models of direction selectivity are also discussed in relation to the proposed model.