"We are researching machine vision for practical industrial applications such as inspection and robot control"

Professor Bruce Batchelor is a researcher in Machine Vision at University of Wales, Cardiff. For the last eight years, he has been developing a machine vision system, which integrates a hardware-based image processing system, with MacProlog. The aim of this work is to produce an intelligent vision system for commercial and industrial applications, such as inspection and robot control.

The system uses images supplied by an external processor which is in turn linked to a camera. At the heart of the system is a program written in MacProlog, which is linked to a specially extended version of Prolog called Prolog+. This integrates commands for the image processing hardware into MacProlog.

"Most food products, such as cakes and confectionery, have a specification, often expressed in metric and word form, which describes how they should look," said Prof Batchelor. He says Prolog simplifies the implementation of language interpreters used to control an inspection system, so that it can recognise an object, given a specification, rather than using a procedure to examining it. Prof Batchelor says his research is not detached academic study, but is aimed directly at solving real world problems that affect industry: "We are researching machine vision for practical industrial applications such as inspection and robot control. In its present form, the Prolog+ system is not fast enough for the most demanding applications, but it is very versatile, and provides a far higher level of intelligence than existing commercial packages."

He also says that the limitations of the present system are not due to Prolog. "We have not begun to realise the full potential of the combination," he said. "What we've done so far has been quite minor compared to what we will be able to do, when we get a faster interface between the image processing sub-system and Prolog." Despite the speed limitations of the present system, he has already performed some notable work, including research on inspecting cake decoration patterns.

Another recent project was concerned with the automation of the process of dissecting very small plants for horticultural micropropagation. In both or these projects, Prof Bachelor said that the research was made possible precisely because of the way in which MacProlog can reason about high-level symbolic information. Prof Batchelor believes that MacProlog offers numerous advantages for this type of work, compared to a procedural language.

The system he has developed at Cardiff already controls a variety of robotic devices and has been interfaced to a speech recognition sub-system. It has also been used to implement a neural network, capable of discovering for itself the statistical parameters of image texture patterns.

See Next Application Brief