A challenge for all machine vision systems is the inevitable requirement to coordinate the timeless environment of the computers required to process image data and the real world time frames of manufacturing machinery and their associated sensors and actuators. It is a divide that conventional machine vision systems have never adequately spanned. Consequently their outputs remain inherently uncertain and their decisions less than reliable.
Lacking a deterministic method of linking image data to the particular part to which it applies, camera and machine vision manufacturers have capitulated and universally rely on the “time” or “order” of arrival of images to associate them with the “right” part being inspected. Unfortunately neither time nor order have real or consistent meaning in modern computers. For all but the slowest inspections (< 1 per second) these approaches fail with some frequency on any multi-tasking operating system, including without exception “real time” OSs.
While this problem has persisted since multi-tasking operating systems became dominant in the ‘80s, it has been greatly exacerbated by multi-core computers that may reduce the processing time for images, but actually increase the level of uncertainty regarding which image data should be associated with which part to be inspected.