An integrated neuromimetic architecture for direct motion interpretation in the log-polar domain

•A bio-inspired vision system for motion interpretation in cortical domain.•A population of motion energy neurons for the computation of optic flow.•Relationships between the log-polar and Cartesian affine description of optic flow.•An architecture that exploits multi-core CPU and GPU for real-time processing.•Benchmarking on synthetic and real-world sequences.

A hierarchical vision system, inspired by the functional architecture of the cortical motion pathway, to provide motion interpretation and to guide real-time actions in the real-world, is proposed. Such a neuromimetic architecture exploits (i) log-polar mapping for data reduction, (ii) a population of motion energy neurons to compute the optic flow, and (iii) a population of adaptive templates in the cortical domain to gain the flow’s affine description. The time-to-contact and the surface orientations of points of interest in the real-world are computed by directly combining the linear description of the cortical flow. The approach is validated through quantitative tests in synthetic environments, and in real-world automotive and robotics situations.