Biological inspired CMOS foveated sensor: For neural network training

Abstract

In this work we describe the design and testing of a custom CMOS-based motion detection system that derives its functionality from biology. The system will beone component of a CMOS visual sensor system that will identify and foveate on objects. The system is inspired by different animals that perform similar functions and have similar retinal structures. For example, we mimic frog's eye using a hemispherical surface structure to provide a wider field of view. In addition, we mimic the human retina with a combination of fovea and peripheral light detectors. The result is a sensor array that combines these aspects to attend and foveate on objects of interest.