Sensor and pixel sizes of CCD and CMOS sensors
The advancing technological development of CCD and CMOS sensors allows for the production of finer and finer semiconductor structures. As a general trend, sensor and pixel sizes shrink in order to cut more and more sensors out of one wafer. This is possible because the sensitivity of the pixels increasingly enhances, too, as much as the noise performance of the electronics is being optimised.
As technical limits are reached in this respect, too, it is worthwhile to compare cameras with different sensor and pixel sizes with the same resolution, especially if…
- there is little light
- low-noise images with high dynamic image response are required
- precision measurements are supposed to be conducted
A larger sensor with larger pixels is in almost every case the technically better choice, however, the price is always higher.
Sensor sizes of standard cameras
Classic machine vision cameras have varyingly large sensors, depending on the camera and resolution used. The majority of cameras with smaller sensors are used with so-called C-mount or possibly CS-mount optics. The C-mount thread has an actual diameter of 1 inch, i.e. 25.4 mm and a thread pitch of 1/32 inch.
The sensors used in standard cameras are clearly smaller and range from 4 to 16 mm image diagonal. These sensor sizes, too, are indicated in inches. The 1-inch sensor has a diagonal of 16 mm.
The inch data of the CCD and CMOS sensors only have a historic explanation: pick-up tubes of TV cameras were used up to the mid-1980s and were long superior to CCD or CMOS sensors which were invented in the late 1960s.
The actual image converter of the tube cameras was located in a glass vacuum tube, and the different pick-up tubes were, among other things, classified according to their outer diameter of the glass bulb. The diagonal of the light-sensitive surface within the tube was of course smaller and represented approximately two thirds of the outer diameter. Equivalent CCD sensors which are supposed to replace the cathode-ray tubes had to cover exactly this surface. A CCD the light-sensitive surface of which corresponds to a 1/2-inch tube was therefore called 1/2-inch sensor, even if this does not correspond to the real CCD sensor size.
Industrial cameras usually use 1/3" sensors in case of resolutions of 640 x 480 pixels, cameras with 1280 x 1024 pixels mainly 1/2". The quite popular camera resolution of 1600 x 1200 pixels often uses a somewhat larger sensor with 1/1.8" with the same pixel size.
In general there is the trend that the sensors become smaller and smaller on the mass camera market. If a standard VGA sensor had, in some cases, a size of 2/3" in the late 1980s, it is only 1/3" today. The miniaturisation is a consequence of enhanced production processes which allow for smaller light-sensitive surfaces with a (hopefully) similar performance. It enables the manufacturers to produce a larger number of sensors at a lower price from one wafer. A 1/3" sensor, for example, has only approximately 40% of the surface of a 1/2" sensor and is therefore cheaper.
Important: If you have the choice between a larger and a smaller sensor for the same camera version, please take the larger variant if you…
- conduct precision measurements, for instance, or finest surface inspections with as little camera noise falsifying the result as possible.
- plan light-critical fast applications with a short exposure time.
- use colour cameras which are possibly supposed to replace monochrome cameras and if only little light is available, as they require 3 - 4x more light than a comparable monochrome sensor.
Large-format sensor sizes of area scan cameras or line scan cameras
In case of high-resolution area scan or line scan cameras, significantly larger sensors with a size of several centimetres are used. The dimensions of these sensors are normally not standardised and result from the resolution and pixel sizes of the sensors. Everything is permitted and only limited by the budget.
A line scan camera with 2048 pixels with 10 μm pixel sizes has a line length of 10.48 mm, in case of 14 μm pixel size the sensor is already 28.6 mm long. From 20 mm sensor diagonal on, the C-mount lens connection can no longer be used.
These cameras typically use Nikon bayonet (F-mount) or M42 to M72 as lens connections. Only then high-resolution sensors with large pixels can be used in order to build line scan cameras with up to 12k pixels or area scan cameras with up to 28 million pixels.
Pixel sizes of CCD or CMOS sensors
As a consequence of the miniaturisation of sensors, the pixel sizes grow smaller and smaller. Sensors of consumer cameras (8 to 12 megapixels for 200 euros) have pixel sizes of mostly 1.7 μm today, the light-active surface per pixel is therefore only approximately 3 μm2. This results in an extremely strong sensor noise in case of non-optimal lighting conditions. For quality control using cameras, this is absolutely inacceptable.
Machine vision cameras (C-mount) with resolutions from VGA to 2 megapixels normally have pixels of 4.6 to 6.5 μm with a 10 - 15 times larger light-active surfaces and thus clearly better signal results. If you need images as noise-free as possible and precise measuring results, look for preferably large sensor pixels, even if these cameras are more expensive!
Pixels with an edge length of 14 or 10 μm are preferentially used in line scan cameras. Due to the high line frequency of 18 Hz, for instance, the maximum exposure time is 1000/18000 = 55 μs for one captured image line. The light-active surface of the pixel can never be large enough in this case.
Full well capacity of a pixel
This specification describes how many electrons a pixel element can hold before it is completely saturated. A pixel of 5.5 μm structure size can accumulate approximately 20,000 electrons, a 7.4 μm pixel 40,000 electrons.
The larger the full well capacity, yet the better the maximum signal-noise ratio. Consumer cameras with pixel sizes of 1.7 μm require only about 1,000 photons for the pixel saturation. In case of digitalisation with 8, 10, or even 12 bits, other noise effects (photon noise, digitalisation noise, dark noise) can already assume significant scales, interfere with the signal and thus influence the image in an extremely negative way.
The larger the full well capacity, yet the better the maximum signal-noise ratio. Consumer cameras with pixel sizes of 1.7 μm require only about 1,000 photons for the pixel saturation. In case of digitalisation with 8, 10, or even 12 bits, other noise effects (photon noise, digitalisation noise, dark noise) can already assume significant scales, interfere with the signal and thus influence the image in an extremely negative way.
Important for machine vision
- The smaller the pixels, the more light is required to capture an image. In case of short inspection times the fact that the light is not sufficient can easily be a problem.
- In case of little light, small pixels deliver clearly noisier images than large pixels, the dynamic image response is reduced. Noise is disturbing for the application. Use bright illumination or LED flash controllers in order to have more light available!
- Many megapixels do not necessarily help. Small pixel structures require high-quality optic imaging, that is high-resolution lenses. Otherwise, blurred images with many pixels are created, but no real structure details.
- Small camera pixels furthermore require an extremely precise mechanic alignment of the sensor, as the depth of field significantly decreases. The tilting of a 5 μm sensor in the housing must only be half as large (+/1 15 μm at aperture = 2.8) as in case of pixels with a structure size of 10 μm. Therefore look for suppliers producing high quality, otherwise the best sensor in the camera is in vain.