Event based visual attention with dynamic neural field on FPGA

Dynamic Field Theory (DFT) is an established framework for neuro-modeling or neuro-inspired computing, well suited for challenging perception and motion related tasks. However, their computational requirements, distributed storage and bandwidth needs make them difficult to design for real-world environments. In this paper, the digital hardware implementation of an event-based dynamic neural field for object tracking and attention is presented. To make computation less complex and hardware-friendly, some optimization on the weights and the neuron model were conducted on the Dynamic Neural Field (DNF) model under a spiking-based computation approach. In a proof-of-concept prototype we show how this derived Spiking DNF (SDNF) core can be interfaced to a Dynamic Vision Sensor (DVS) silicon retina and integrated into a more complex architecture able to perform selective attention.