Vision-based dimensional measurement

Machine vision is a useful and often necessary technology, whose performance meets industrial requirements and constraints. Its development has benefited from rapid and continuous progress in solid-state point, linear and matrix sensors, as well as in microcomputing and image-processing electronics. Machine vision can be defined as a stand-alone system implemented in an industrial environment and comprising a camera or other optoelectronic sensor sensitive to light wavelengths, and an image processing unit, both of which may be used to make automatic decisions.

One sector particularly concerned by machine vision is non-contact 2D or 3D metrology. Numerous non-contact position, shape and distance measurement systems are used in the automotive and aerospace industries, and especially in hostile environments (submarines, nuclear power plants, boilermaking, etc.). This type of measurement system is more compact than a coordinate measuring machine (probe) for the same resolution. In fact, the measurement accuracy achieved using calibrated industrial cameras is generally comparable to that of a coordinate measuring machine.

The purpose of this article is to review 2D vision systems for positioning, identification and measurement. If we limit ourselves to a 2D measurement system (surface appearance) in the visible range, the object to be inspected has three dimensions, one of which will necessarily be neglected in relation to the others. This third dimension can be checked by synchronizing several imaging systems and simultaneously studying the images they provide. This method makes it possible to control objects with complex shapes, where no single dimension can take precedence over the others. The object may be static, or moving in translation or rotation within the context of the production line. In some cases, this movement can be used to make several acquisitions under different lighting conditions, highlighting the volumetric aspect of the object. It is the position of the object in relation to the incident source that creates the contrast effect sought in an image. The different facets of the object must appear distinctly separate in the image dynamics, and must also stand out from the background.

3D vision techniques, which offer the possibility of solving specific problems that cannot be easily solved using a 2D approach (such as determining the dimensions of left-hand surfaces), will not be discussed here. Interested readers can refer to the article "Three-dimensional geometric perception in robotics" [R 7 750] in the Industrial Computing treatise.