Image Resolution in Machine Vision
Image resolution dictates the smallest physical feature a vision system can reliably detect and measure. It is a combination of the sensor's total pixel count and the optical resolving power of the lens. In industrial automation, specifying the correct camera means calculating exactly how many pixels are required to cover your field of view while still rendering your target defect clearly. Pushing for the highest megapixel count is rarely the right call, it reduces frame rates, taxes interface bandwidth, and demands more expensive optics to actually resolve the detail the sensor is capable of capturing.
Sensor resolution vs. spatial resolution
When comparing cameras, buyers often look first at the sensor resolution. This is the absolute number of pixels on the CMOS chip, typically expressed in megapixels. A camera outputting 2448 x 2048 pixels yields a 5 MP sensor resolution.
However, machine vision software algorithms do not measure megapixels. They measure physical space. Spatial resolution defines the actual real-world size that a single pixel represents at a specific working distance.
If your field of view is 100 mm wide and your sensor is 2000 pixels wide, your spatial resolution is 0.05 mm per pixel. If the goal is to detect a 0.02 mm micro-fracture on a machined part, that 5 MP sensor will fail regardless of its overall megapixel count, because the defect is smaller than a single pixel.
The tradeoff between resolution and speed
Moving more data takes more time. Every pixel generated by the sensor must be read out, transferred over a hardware interface, and processed by the host PC.
A 1.6 MP camera might effortlessly output 200 frames per second over a standard USB3 Vision connection. Swapping that camera for a 20 MP sensor will drastically reduce the maximum frame rate simply because the interface bandwidth becomes a bottleneck. System integrators must balance the need for detail against the mechanical speed of the production line.
How much resolution do you actually need?
The prevailing rule in machine vision is to use the lowest resolution that reliably solves the application. To determine that number, engineers calculate the minimum feature size they need to detect and ensure it covers a specific number of pixels.
|
Inspection Task |
Pixel Requirement (Rule of Thumb) |
Engineering Rationale |
|
Presence / Absence |
1 to 2 pixels |
You only need to verify a gross edge or a mass exists to pass or fail the part. |
|
Dimensional Measurement |
3 to 4 pixels |
Sub-pixel algorithms can accurately determine precise dimensions if the part's edge spans multiple pixels. |
|
Barcode / OCR |
3 to 5 pixels per narrowest element |
Text characters and 2D matrix codes require high spatial resolution to cleanly separate tightly spaced lines. |
Why your lens dictates your true resolution
Placing a 24 MP sensor behind a low-quality lens will not yield a 24 MP image. Lenses have their own physical limitations, typically measured in line pairs per millimeter (lp/mm).
If the lens cannot resolve detail fine enough to match the physical size of the sensor's individual pixels (the pixel pitch), the resulting image will be soft and blurry. High-resolution sensors generally feature very small pixel pitches. They demand high-quality, precision optics to ensure the incoming light is focused sharply enough to actually utilize the expensive silicon.
Frequently asked questions
Often, yes. To fit more pixels onto the same physical sensor footprint, the pixel pitch must shrink. Smaller pixels have a lower full well capacity and physically collect fewer photons. This lowers the quantum efficiency and overall sensitivity of the camera, requiring brighter industrial strobe lighting to compensate.
It is a software-driven mathematical technique. Advanced vision algorithms can interpolate data between physical pixels based on grayscale gradients. This allows the software to measure edges and geometries with a precision significantly greater than the camera's physical spatial resolution.
Yes. Most modern CMOS sensors allow you to crop the active sensor area in hardware. If you only need a 1000 x 1000 pixel area of a 20 MP sensor to track a specific part, activating an ROI drastically reduces the data payload. This significantly increases the frame rate without requiring you to switch to a lower resolution camera.