Cloud architecture for industrial image processing: Platform for realtime inline quality assurance

Abstract

Cloud computing offers the opportunity to minimize the evaluation time of complex algorithms — e.g. needed for computational imaging — by horizontal scaling of the available computing resources. By this way, new image analyzing algorithms can be employed in weak real-time conditions, like inline quality analysis in production with time stamps in the order of several tens of seconds. The cloud offers a platform to merge sensor data of all production processes to analyze quality data comprehensively, e.g. for methods like predictive maintenance. Typically, cloud environments are applied for the Internet of things (IoT) or Big Data analysis. But IoT-applications usually generate very small data packages (like sensor values with a size much less than 1 megabyte), while BigData applications deal with very high data volume (terra-or petabyte). Image processing requires an environment, which is optimized for medium size data streaming, composed of images with a size in the lower megabyte range. In this paper, a sensor-to-cloud architecture as a platform for image processing is described. This approach is upward compatible, because it is not necessary to change the sensor hardware, e.g. if algorithms with considerable higher computing complexity are desired (like for a smart camera), so algorithms can be exchanged in the cloud without interrupting the production process.