Myriad : a distributed machine vision application framework

This thesis examines the potential for the application of distributed computing frameworks to industrial and also lightweight consumer-level Machine Vision (MV) applications. Traditional, stand-alone MV systems have many benefits in well-defined, tightly- controlled industrial settings, but expose limitations in interactive, de-localised and small-task applications that seek to utilise vision techniques. In these situations, single-computer solutions fail to suffice and greater flexibility in terms of system construction, interactivity and localisation are required. Network-connected and distributed vision systems are proposed as a remedy to these problems, providing dynamic, componentised systems that may optionally be independent of location, or take advantage of networked computing tools and techniques, such as web servers, databases, proxies, wireless networking, secure connectivity, distributed computing clusters, web services and load balancing. The thesis discusses a system named Myriad, a distributed computing framework for Machine Vision applications. Myriad is composed components, such as image processing engines and equipment controllers, which behave as enhanced web servers and communicate using simple HTTP requests. The roles of HTTP-based distributed computing servers in simplifying rapid development of networked applications and integrating those applications with existing networked tools and business processes are explored. Prototypes of Myriad components, written in Java, along with supporting PHP, Perl and Prolog scripts and user interfaces in C , Java, VB and C++/Qt are examined. Each component includes a scripting language named MCS, enabling remote clients (or other Myriad components) to issue single commands or execute sequences of commands locally to the component in a sustained session. The advantages of server- side scripting in this manner for distributed computing tasks are outlined with emphasis on Machine Vision applications, as a means to overcome network connection issues and address problems where consistent processing is required. Furthermore, the opportunities to utilise scripting to form complex distributed computing network topologies and fully-autonomous federated networked applications are described, and examples given on how to achieve functionality such as clusters of image processing nodes. Through the medium of experimentation involving the remote control of a model train set, cameras and lights, the ability of Myriad to perform traditional roles of fixed, stand-alone Machine Vision systems is supported, along with discussion of opportunities to incorporate these elements into network-based dynamic collaborative inspection applications. In an example of 2D packing of remotely-acquired shapes, distributed computing extensions to Machine Vision tasks are explored, along with integration into larger business processes. Finally, the thesis examines the use of Machine Vision techniques and Myriad components to construct distributed computing applications with the addition of vision capabilities, leading to a new class of image-data-driven applications that exploit mobile computing and Pervasive Computing trends