Abstract
This paper reviews the process of automated visual inspection and highlights the fact that product location is likely to involve considerably more computation than product scrutiny. The situation is illustrated by a number of case studies drawn broadly from the baking and cereals industries. It is common for baked products to be manufactured at rates of 10–30 per second, and inspection can realistically provide reliable control over quality for 100 per cent of products. For cereal grains, 100 per cent inspection is much more difficult. An alternative strategy is to take a substantial sample from each consignment: typically, ∼ 3 kg of grain should be inspected within 3 min—a target that has now been met. Where profit margins are low, the cost of inspection equipment has to be kept down, and serious efforts need to be devoted to cost effective real-time implementation. This paper describes sampling procedures that permit considerable speed-up to be achieved in software alone, without incurring any real loss of robustness.
Finally, the paper points to the difficulties of specification and delivery of algorithms against a backcloth both of conflicting parameters—sensitivity, robustness, accuracy, speed and implementation cost—and of lack of deep knowledge of the nature of the image data being processed. While this provides some scope for theoretical analysis, case studies are proposed for providing a positive way forward, because these can help workers to see more clearly where solutions to their problems might lie.