A spike-based neuromorphic stereo architecture for active vision

Abstract

The problem of finding stereo correspondences in binocular vision is solved effortlessly in nature and yet is still a critical bottleneck for artificial machine vision systems. As temporal information is a crucial feature in this process, the advent of event-based vision sensors and dedicated event-based processors promises to offer an effective approach to solve stereo-matching. Indeed, event-based neuromorphic hardware provides an optimal substrate for biologically-inspired, fast, asynchronous computation, that can make explicit use of precise temporal coincidences. Here we present an event-based stereo-vision system that fully leverages the advantages of brain-inspired neuromorphic computing hardware by interfacing event-based vision sensors to an event-based mixed-signal analog/digital neuromorphic processor. We describe the multi-chip sensory-processing setup developed and demonstrate a proof of concept implementation of cooperative stereo-matching that can be used to build brain-inspired active vision systems.