Choosing the right hardware for a machine vision system
While most vision systems include a similar collection of hardware, there are specific choices to be made for each component – and they all need to work seamlessly together.
The exact configuration of any machine vision system is dictated by the job requirements: by starting with the application, you can work with a vendor to determine the precise setup you need.
Mono vs. color sensors
Machine vision image sensors are monochromatic, meaning each pixel only detects the intensity of the light falling on it, not what color it is. They can capture color images using red, green, and blue filters, but in the vast majority of industrial applications, monochromatic imagery is a more effective choice. Since color filters reduce light intensity, they’re only used when color is absolutely required for a specific task.
Image sensors
At the center of any vision system, an image sensor converts light energy into electrical signals that can be analyzed by software.
Image sensors are solid-state semiconductor chips comprising millions of photodetectors called pixels, which convert light into electrical signals. The two main technologies used are charge-coupled device (CCD) or complementary metal oxide semiconductor (CMOS). CMOS technology is newer and taking an increasing share of the market, thanks to lower costs and faster processing speeds, but CCD sensors are still used in applications that require extremely high-quality images.
Rolling vs. global shutters
Machine vision cameras use electronic shutters to determine how the pixels are exposed to light. Despite the use of the word “shutter,” as in an old mechanical camera, the shutter is part of the way a sensor processes the light falling on it, not something separate.
A rolling shutter image sensor exposes lines of pixels sequentially. This kind of sensor is lower cost, but if the target being imaged is moving rapidly, the image can be distorted.
A global shutter image sensor can expose all of its pixels at once. They can be more complex to manufacture, but prices have come down and they are widely used in industrial applications.
Additional sensor features
Sensors range widely in a number of characteristics, including size, resolution, pixel size, frame rate, sensitivity, and dynamic range. Each choice can affect other characteristics, as well as cost. Sensors can also have a wider spectral range than the human eye and detect light into the infrared (IR), useful in some industrial applications.
There’s no single answer to “which sensor should I use?” Sensors should be chosen based on their specific purpose, whether for a fast-moving assembly line, where a high frame rate might be essential, or detecting tiny details in a complex product, where resolution and dynamic range might be more important.
Machine vision lenses
Lenses collect and focus light onto image sensors. They are matched with the sensor so they completely illuminate the photosensitive area with a resolution equal to or greater than the sensor’s pixel size.
Manual lenses need to be refocused if the objects being observed change, or if the camera moves.
But this can be time-consuming if done frequently, or if there are a large number of cameras that need to be adjusted. In these situations, autofocus lenses are optimal.
Autofocus lenses adjust their focus automatically as needed, saving time if frequent changes are expected.
Liquid lenses are a specialized subset of autofocus lenses that adjust their shape on the fly to maintain sharp focus. They’re also smaller and more resistant to shock and vibration than mechanical autofocus lenses of similar spec.
Covers
A camera cover, also called a housing or enclosure, protects your expensive machine vision camera from the dust, oil, water, food particles, impacts, and other hazards that come with operating in an industrial environment.
Covers can also dissipate heat, protect cable connections, and prevent tampering.
Covers are rated by how well they protect the camera, with IP67 being the usual rating for a dustproof camera in an industrial environment. An IP67 rating also means the enclosure is waterproof, so the operating area can be washed down safely.
Additional vision system hardware requirements
Another factor in choosing the type of processing is the required hardware. Because of the amount of parallel processing it requires, deep learning generally needs to run sophisticated parallel processors called GPUs. Edge learning requires less processing, and rules-based algorithms the least.
Some industrial cameras have onboard or embedded processing. Such cameras can typically do both rule-based and edge learning processing entirely on their own. Deep learning may require a more powerful processor located in a separate PC.