Machine vision essentially uses optical, non-contact sensors to automatically receive and interpret images of a real-world scene to obtain information for controlling machines or processes. Vision systems can be used independently, such as as inspection tools or as components in automated control systems. Early vision systems, like most other automated control devices, were expensive and difficult to use. In recent years, their costs have decreased significantly, their recognition capabilities have improved dramatically, and they have become much easier to use. Therefore, the application of vision has increased exponentially and is now widely used in many automated systems and manufacturing processes.
It is important to note that machine vision currently falls short of human vision in many aspects. Therefore, any vision application must be carefully considered. Machine vision is continuous and tireless; many vision devices can operate beyond the visible spectrum, function in harsh environments, and precisely execute pre-set programs. Human vision, in contrast, offers higher image resolution, can quickly interpret complex sensory information, and is highly adaptable; however, it is confined to the visible spectrum, is prone to fatigue, and is subjective.
Machine vision is suitable for workpiece identification, location finding, inspection, and measurement. Therefore, it is applied in various production environments, including cleanrooms and hazardous environments, such as monitoring on high-speed production lines, microscopic monitoring, and closed-loop process control. It can also be applied to precise non-contact measurement and robot guidance. This section will not discuss all these applications, but will focus on vision applications related to robotic systems. The main application of vision in robotic systems is guidance, including workpiece pickup and tracking, workpiece presence/absence inspection and defect identification, as well as workpiece identification, including optical characteristic assessment and barcode reading. These will be discussed in detail later.
First, it's worthwhile to introduce the main components and operations of a simple vision system. A typical vision system includes a camera, lighting equipment, processing hardware, and software. The software is specifically designed for the vision system and performs image analysis for a particular application. There are three main operations in a vision system: first, acquiring an image; second, processing or modifying the image data; and third, extracting the desired information. Each operation affects its next operation. For example, using an external light source for illumination in the initial operation can greatly simplify image capture, which in turn reduces the required processing and makes it easier to extract the desired information.
There are many types of cameras to choose from, and their key parameters are resolution, field of view, depth of field, and focal length. Focal length determines the nominal distance at which the camera produces a focused image. Depth of field refers to the range of sharpness between the in-focus and out-of-focus images. Field of view determines the size of the image at the focal length. Resolution is the number of individual grids into which an image is divided; it determines the smallest resolvable metric or feature.
Lighting equipment is of paramount importance. Many different technologies are available, including direct and diffused lighting from the front, rear, or side of the object, as well as structured and polarized light. Ambient lighting influences various sources, including daylight, factory lighting, and any other possible light sources. In particular, changes in ambient light must not affect the operation of the vision system. The purpose of vision system lighting is twofold: to highlight key features of objects and to eliminate any potential impact from changes in ambient light.
For example, in a welding guidance system, a vision sensor is mounted directly in front of the welding torch, aligned with the weld seam at a distance of only 25mm. To allow the camera to "see" the weld, infrared light generated by a laser provides illumination, and a filter mounted in front of the camera filters out all light other than that laser wavelength. Light from the welding process is thus filtered out from the image received by the camera so that the camera can "see" the weld.
Backlighting is helpful for workpiece positioning and measurement because it simplifies the image of an object to shadows stripped of all surface features, thus simplifying the task of the vision system. The background of the object is also important for distinguishing workpieces. A typical application of vision is providing position and orientation information when a robot picks up workpieces from a conveyor belt, such as packing chocolates into a box. We often use white conveyor belts because their color contrasts strongly with the color of chocolate.
Highlighting key features or removing irrelevant information from images significantly reduces the complexity and time required for image processing. Furthermore, the reliability of visual operations is improved. Eliminating the effects of ambient light changes also enhances reliability. To completely eliminate the influence of ambient light, it is necessary to enclose the image operations in an opaque enclosure.
In robotic automation systems, vision is most widely used in packaging, especially in the food industry. Products are often scattered on conveyor belts and then transported to robotic packaging workstations. Imaging systems determine the product's position and then feed this information back to the robot, allowing it to pick up the product from the conveyor belt and place it into a packaging box. These are common conveyor belts; therefore, the imaging system needs to continuously track the pick-up point's position throughout the entire robotic unit at the input end. These systems often involve multiple robots, so it's necessary to determine which robot should perform the pick-up operation to balance the workload among the robots. For these typical applications, there are standard solutions that simplify implementation and offer better cost-effectiveness.
The same vision system can also be used for quality control. For example, by checking the shape of the chocolates to be packaged, it ensures that all deformed products are rejected. Another example is the packaging of biscuits. During the packaging process, the vision system also checks the color of the biscuits. If the color is too dark, the biscuits are overcooked; if it is too light, they are undercooked.
In both cases, the small biscuit pieces must be rejected. Vision systems, especially those used in assembly systems, are used to inspect features or workpieces. They check the success of previous operations and ensure that assembly automatically stops when an unsuitable workpiece is encountered.
Vision is also used to inspect manually loaded fixtures, checking that all workpieces are pre-loaded before the next operation and ensuring all items are in the required positions. While this could also be achieved by mounting individual sensors on each workpiece, the vision method may be more cost-effective, especially when many different workpieces use the same fixture.
Vision can also be used to read characters on labels or barcodes that provide product identification. For example, palletizing systems can use vision to identify different boxes, ensuring they are placed on the correct pallets. In most applications of this type, barcode readers tend to be inexpensive; however, in some cases, vision systems are superior.
Machine vision enables the automation of applications by providing guidance, measurement, or quality control. The cost of vision systems continues to decrease, while ease of use and performance continue to improve. However, vision systems always require careful investigation to ensure operational reliability.
