Any
artificial agent
acting in a real-world scenario should perform the analysis and
interpretation of visual motion in a fast and reliable way. This task
is effectively solved by the primate visual system, specifically in the
dorsal cortical pathway, i.e. the ``action stream’’.
The dorsal pathway is organized in a hierarchical structure of layers
that process the visual signal to extract simple image features that
become more informative and complex in higher layers.
In this work, to fully exploit such an effective hierarchical approach
adopted by the mammals’ visual system, we propose a
bio-inspired
vision system that mimics the first stages of the dorsal cortical
pathway, i.e. the cortical motion pathway (namely V1-MT-MST).
In particular, we have considered the following processing stages:
- The front-end
stage of the proposed bio-inspired
vision system performs a space-variant image acquisition.
- The second stage
is the computation of the optic flow
through a distributed neural architecture that mimics the mechanisms
underlying motion analysis in the areas V1 and MT.
- The final stage
of the proposed vision system
performs motion interpretation through a first order analysis of the
optic flow that leads to orientation of surfaces and time-to contact
(TTC) estimation