A system for the development of machine vision systems.

The development of hardware and software for a High resolution interactive Image Processing System (HIPS) is described, the purpose of which is to facilitate the investigation, development and optimisation of dedicated machine vision systems. The hardware for HIPS is built around a Multibus backplane with a 16-bit central microprocessor. The operating system of HIPS consists of customised modules for image capture, storage and display, and is built on a commercial real-time multitasking executive, iRMX86. All the resources required for images are created, managed and released by the operating system on behalf of user programs, and a consequence of this is that the resource requirements of those user programs are greatly reduced, the programs themselves are easier to use interactively, and are easier to develop. The size and number of images that can be created in HIPS depends on the amount of memory pool available to the operating system - memory that is shared with other programs. A unique feature of HIPS is that all access to the display and image stores must be through the operating system, which helps to prevent developers from presuming the nature of the final, dedicated machine vision system. A detailed performance model of HIPS is developed that assists the designer in predicting the speed of the dedicated system. The application of HIPS to the development of algorithms for the detection of shape and simple surface defects in Kiwifruit is described. The algorithm for detecting shape defects correctly discriminates between the various grades of fruit, while executing at about one third real-time speed on HIPS - a factor which is easily overcome with relatively minor hardware reconfiguration. The surface defect detection algorithm gives satisfactory results, but executes in a time roughly two orders of magnitude too slow. A performance analysis of the algorithm reveals that the precise bottleneck is the use of a convex hull step, used to even out illumination over the fruit. It is proposed that this step be abandoned completely, or implemented directly in hardware. A strategy for the investigation, development and optimisation of dedicated machine vision systems is detailed, built from the experience in the literature, the development of HIPS, and the applications of this machine to the Kiwifruit defect detection problems