Low-cost image and video processing using high-performance middleware in single-board computers with open internet standards

Abstract

Image processing is becoming ubiquitous in many activities. This kind of systems use industry-standard libraries, such as OpenCV, and GPGPU techniques such as CUDA and OpenCL. Nowadays, these are being ported to many computing platforms, offering significant processing power even in devices with limited resources. However, the only model that is truly ubiquitous, is the web itself. Modern browsers feature quite complex internals and offer sophisticated development and profiling tools, in order to offer the best user experience. Introduction of HTML5 allows Realtime video and image manipulation, in browser space, without any plugin. In addition, Wasm (web assembly) Javascript execution engine provides fastest possible performance by means of highly customized compiler and runtime, in almost any browser, including embedded ones. This paper presents an image processing system, architected as a modular web application, using only Raspberry PIs with a compact but fast middleware server, that performs all image operations in browser space by means of web assemblies. All components, including database support, can run in a single board, providing image and video processing speeds that match, or surpass, their native compiled C counterparts on the same platform. This solution has a very low cost, that fits with emerging markets, making it ideal for LATAM scenarios.