Motion perception with recurrent self-organizing maps based models

Representation and processing of spatio-temporal motion information on different levels of granularity, is one key capability of the visual processing ability of the human brain. We introduce a multi-layered model consisting mainly of recurrent self-organizing maps and a neural associative memory for motion prediction. This processing structure is psychophysically and biologically inspired and some of the equivalent findings are discussed. The model is self-contained and can be used for motion planning and prediction and also for all kinds of context-sensitive information processing with demands on prediction ability.