Retina Camera for vision modeling and high-resolution imaging applications

Abstract

Summary form only given. A novel, high-resolution camera will be demonstrated with an integrated image processing module that captures functional models of retinal signal processing in compact, low-power hardware. The device uses a high dynamic range megapixel CMOS image sensor and near-sensor computation to duplicate robust retinal image formation and feature extraction mechanisms. Retinal signal flow and data representations have been translated into multi-core processing modules in programmable logic in order to create a real-time implementation suitable for feeding cortical computing modules for object detection, classification, and recognition. The Retina Camera increases system-level performance in real-time applications by exploiting multiple layers of adaptation and simultaneous multi-scale spatial-temporal signal representations to achieve greater invariance to illumination, contrast, scale, view point, signal fidelity and clutter. Live image acquisition and visualization of the output of multiple retinal processing layers will be shown during the demonstration.