Machine vision is the use of cameras and software to inspect every item on a production line. The cameras photograph each item, and software judges it: reading barcodes, checking labels and parts, and spotting defects faster than a person can.
A vision system has four parts. A light shines on the item in a fixed way, because the picture must be the same every time. A camera takes the picture when a sensor says that the item has arrived. Software analyses the picture. And a signal goes to the machine: pass, or reject.
Consider a line that fills jars of pickle at 300 jars a minute. That is five jars a second, so each jar passes the camera for only about 200 milliseconds. In that time, the system checks three things: the cap is present and straight, the label is not torn, and the printed batch code can be read. A jar that fails is pushed off the line by a small air jet.
Setting the system up is a matter of balance. If the limits are too strict, the system rejects good jars, which wastes product. If they are too loose, it passes bad ones, which reach customers. The lighting matters as much as the camera: a shiny label can reflect light and look torn. So the plant tests the system with a set of known good and known bad items before it trusts the result.
No person could do this at that speed for a full shift. Machine vision is therefore one of the most widely used kinds of computer vision in industry. Newer systems run edge AI models, which learn what a good item looks like from examples, so that they can spot defects that no one has described.
The same cameras serve other jobs. They read the code that an inkjet printer has printed. They show an industrial robot where a part lies, which becomes AI robotic picking when the part is unknown. They check tiles in ceramic machinery and the joints of an assembly line. The results go to the MES.