Real-Time Vision Using a Smart Sensor System

This paper presents a real-time data processing concept for asynchronous "address event" image sensors in highspeed machine vision applications. In the implemented system, a dual-line temporal contrast vision sensor asynchronously responds to relative illumination intensity changes in the visual scene, and encodes the information in the form of "timed address event" representation (TAE) data. The TAE stream consists of the event generation time concatenated to the address event (AE), which encodes the coordinates of the pixel in the 2times256 matrix. The event data processing includes object detection, denoising and scaling, and feature extraction. The feature extraction step comprises circle fit and orientation estimation techniques that enable robust recognition of a wide range of objects. The data processing algorithms take advantage of the efficient information encoding in the TAE data protocol, yielding a flexible, compact and low-cost real-time machine vision system for high-speed shape detection, object orientation extraction and monitoring.