Polarimetric dynamic vision sensor p(DVS) neural network architecture for motion classification

Abstract

Abstract The purpose of this study is to introduce efficient bio‐inspired vision architectures, integrating human cognition capabilities, such as computation and memory emulating neurons and synapses, together with polarization of light properties, that would ultimately revolutionize and give rise to the next‐generation highly efficient augmented intelligence vision systems. The system consists of a polarimetric dynamic vision sensor) and a spinning wheel, whose motion is externally modulated and controlled by means of a signal generator, varying its motion pattern and speed, respectively. A neural network has been designed to classify different object speeds and motion patterns. The neural network utilizes a limited number of events within a given time window, instead of full‐frame images; as well as two classifiers, which practically take a single input and independently classify both its speed and motion pattern. As a result, both high computational efficiency and motion classification accuracy are achieved.

Keywords

• Image recognition
• Computer vision and image processing techniques
• Neural nets