Real time ego-motion estimation with neuromorphic analog VLSI sensors

An analog VLSI (aVLSI) sensor with on-board focal plane smart sensor pixels for the analysis of visual motion is used for estimating ego-motion. The sensor has remarkable properties: the on-chip information processing is performed in parallel and in real time, the sensor is small, inexpensive and consumes very little power, incorporating mechanisms similar to those found in neural systems. In some applications this sensor shows a performance superior to machine vision approaches. Due to the limited accuracy of the sensor, parameters of simple velocity fields can be recovered. We show that nonetheless rather complex tasks such as recovery of ego-motion for navigation can be accomplished. This is demonstrated by recovering the heading after rotation of the sensor.