Skip to main content

Vision-Doctor

Camera

The eye of the inspection system

Resolution of sensors -
What resolution is required?

The typical sensor resolution of industrial cameras has increased over the years. Thanks to improved manufacturing processes, it is now possible to produce smaller pixel structures with comparable light sensitivity to the larger sensor pixels of the past. The quantum efficiency of the pixels has increased significantly.

It was the development of CMOS sensors that made such high resolution sensors possible. Gone are the days when a 16-megapixel camera with a CCD sensor could only take a few frames per second.

Today, sensors with resolutions of 2, 5, 8, 12 or 24 megapixels are used in combination with a C-mount lens mount, where 5 to 10 years ago only megapixel cameras were in use.

Even higher resolution cameras require large lens mounts such as APS/C, M42, F-mount or even M72 mounts due to their large sensor area. High resolution industrial area scan cameras from 30 to 245 megapixels offer impressive image quality and detail that is essential for demanding industrial applications.

These cameras are capable of capturing extremely detailed and clear images, making them ideal for inspecting large objects with very fine structures. Typical applications include print inspection, display inspection, medical applications and aerial imaging.

Resolution, mono or colour, data transfer...

Once you have decided whether you want to shoot in monochrome or colour, the next step is to choose a camera with the right resolution.

The choice of transfer interface is also important. Standard image transfer can be achieved using the GigE or USB3 interface, for example. High-resolution sensors require a more powerful transmission interface. 5 GigE or 10 GigE, CameraLink HS or CoaXPress are good choices.

Example:
Sensor resolution and image details

VGA resolution

5 megapixels

What resolution do I need for my camera?

One of the toughest questions in industrial machine vision! Let's get to the bottom of it... Please read this page to the end!

Important information

  • What is the camera's field of view?
  • What is the smallest feature I want to resolve?
  • How many pixels do I need to resolve this structure?

The required sensor resolution can then be calculated.

An example

  • Component, length 64 mm
  • Sensor 640 pixels.

The sensor therefore has an X resolution of 10 pixels per millimetre. But can structures as small as 0.1 mm be captured accurately?

What would happen to a feature that is exactly one pixel in size?

For example, the small object feature could fall exactly in the centre of the pixels on the sensor. In simple terms, the feature is (theoretically) 100% captured. But what happens if the smallest feature falls between two or four pixel structures on the sensor? Then the ideal 100% signal is also distributed to the surrounding pixel structures and the contrast drops significantly.

Unfortunately, a 1:1 mapping does not work to obtain useful data.

Physical resolution required to detect the smallest features

A spot size of at least 3 to 4 pixels is required for clean optical detection. This means that an optical feature must cover at least this number of pixels on the sensor in order to be clearly and reliably detected.

Smaller spots can also be detected, but only if the contrast is good. If the contrast is poor, the detection accuracy drops significantly. The minimum size of a detectable feature therefore depends on the contrast of the feature.

Rule of thumb

A general rule of thumb is that a camera with 3 times the number of pixels is needed to detect smaller features. This means that the higher the physical resolution of the camera, the more accurately the smallest features can be detected.

Feature with 3-4 pixels in size

Practical example

A clear example illustrates the importance of physical resolution: for a 64 mm wide image with 640 pixels, the physical resolution is 0.1 mm per pixel. This means that clearly visible features are not 0.1 mm in size, but 0.3 mm. This increase in size takes into account the required spot size of at least 3 pixels for reliable detection.

Required sensor resolution when measuring objects

Measurement Algorithms and Accuracy

Modern measurement algorithms have an impressive accuracy of less than 1 pixel. An example of this is the calculation of distances: Instead of measuring 28 or 29 pixels, precise algorithms allow measurements of 28.37 pixels. This decimal accuracy is achieved through 'sub-pixel interpolation', which uses the inflection point of the greyscale function to achieve more accurate results.

Subpixel Interpolation

Sub-pixel interpolation is a technique that can increase measurement accuracy to less than 1 pixel. Under laboratory conditions, an accuracy of up to 1/20 pixel can be achieved. In real-world applications, however, the measurement accuracy is typically between 1/2 and 1/10 pixel, taking into account probing and calibration errors.

Averaging local anomalies

Another important aspect of measurement accuracy is the averaging of local anomalies such as imaging errors, contour variations and sensor noise. With a large measurement range, averaging of these anomalies is critical to improving accuracy and achieving consistent measurement results.

In the example shown, it therefore makes a difference to the achievable accuracy if the measurement tool window is only one pixel wide or many pixels wide!

Edge detection in software

Practical example

A concrete example illustrates the importance of measurement accuracy. If a distance of 64 mm is to be measured and the camera has a resolution of 640 pixels, the measurement accuracy is not limited to whole pixels. By using precise algorithms, the distance can be measured with an accuracy of one decimal place, which means that clearly visible features are not just 0.1 mm in size, but can be measured with an accuracy of around 0.05 to 0.01 mm.

This means that a much lower camera resolution is required than for detecting the smallest structures.

Required sensor resolution depends on software algorithm and application

Another approach to determining the required camera resolution is to consider the software algorithms used.


A few examples

  • For reliable text recognition, characters should be around 30 pixels high to distinguish characters such as a 'G' from a '6', an 'O' from a '0', etc.
  • Barcodes require a minimum of 2 to 3 pixels per smallest bar width.
  • Data Matrix codes can be recognised from a module size of 3 to 4 pixels, depending on the marking process, but code grading requires approximately 10 pixels.

The required camera resolution can also be recalculated if the feature size, software resolution requirements and total image size are known.

Need to buy the right camera?

Vision-Doctor.com is a private, independent, non-commercial homepage project and not a technology provider or system integrator. Suitable technologies and further professional support can be obtained from the companies & partners listed below.

If necessary, I will be happy to provide a quick recommendation, contacts and brief information.

Industrielle Kamera fuer Bildverarbeitung