provided by: 
There are many different industrial video cameras for machine vision available. The perfect camera does not exist for all applications because optimizing every parameter is not possible (or profitable). As a result, every camera manufacturer makes choices to optimize the camera performance for certain applications. Camera specifications can be reported differently making it difficult to decipher the real differences between different manufacturers. The following sections explain some key parameters to consider in machine vision and how these can influence different applications.
CAMERA SPEED
Camera speed is defined as the number of frames a camera can deliver each second. The speed of the camera can be reported for a portion of the image, such as using partial scanning or binning. This value will be greater than the number of "full" frames per second, which is the number of frames of the full resolution of the camera that are output per second.
Throughput is a key performance indicator in almost all of the machine vision applications. When the camera speed is the determining factor for throughput in a system, a higher camera speed results in a higher throughput for the system. Even one additional frame per second can have a significant effect.
A high camera speed can also allow you to improve the accuracy of a system by obtaining multiple images of one object very quickly to avoid errors that could arise from having just one image of the object.
In order to obtain the fastest cameras possible, the manufacturer must understand the sensor and maximize the output of the sensor without taking risks. One of the commonly known tools to increase the camera speed is the use of multiple tap sensors, but there can be some side effects.
A dual tap sensor allows for the image to be split in half so that the charge transfer can be completed much more quickly. One of the side effects of using a dual tap camera is a channel mismatch, which is visible as a difference in offset, linearity or sensitivity of the two image halves.
Another effect that can be present with a dual tap sensor is a pixel shift. A pixel shift is a very small apparent spatial shift between the two image halves. Note that this is a very small shift that cannot be seen with the human eye and is much smaller than a pixel.
These side effects of using a multiple tap sensor can influence the accuracy of a measurement. For example, if items that need to be measured lay exactly on the seam of these image halves, both channel mismatch and pixel shift can cause small position and size deviations.
IMAGE ARTIFACTS
Other image artifacts that can affect the quality of the image are smear and blooming. Smear is defined as light leakage into the transfer part of the CCD causing unwanted vertical lines in the image.
Blooming is an overflow of photodiode charge into neighboring pixels. It can occur when trying to image a very bright spot. Instead of an image with circular spot, it is deformed on the bottom.
Either of these artifacts results in a distorted image but can be avoided by using correct lighting; therefore, these artifacts almost never appear in machine vision applications where the lighting is under control.
UNIFORMITY
The uniformity of an image is influenced by the fixed pattern noise present in the image. There are two types of fixed pattern noise: fixed pattern noise in a black image and fixed pattern noise in a white image.
The fixed pattern noise in a black image can further be divided into two types. The first is a difference in the average dark current of the photodiodes. This can differ from pixel to pixel and is visible as a granular effect. This effect does not change from image to image. The second is an excessive difference in dark current of a single photodiode with respect to other photodiodes. This can be seen as white spots and are often referred to as hot pixels.
Both types of fixed pattern noise in a black image depend on gain, integration time, and temperature. For example, the dark current in a pixel doubles with every 6 to 8 degrees Celsius temperature increase.
Fixed pattern noise in white is caused by differences in sensitivities of the pixels and is independent of camera settings. This can observed as dark spots on a white image.
Hot pixels and pixels with extreme deviations in sensitivity are called defect pixels. These defect pixels can be interpreted by the system as contamination on the product causing false rejects. Besides that, defect pixels can influence the measurement algorithms which will have an effect on the measurement accuracy of the system. Defect pixel correction within the camera is often available to avoid these unwanted defects.
NOISE PERFORMANCE / DYNAMIC RANGE
The noise performance of a camera depends on two parameters: the full well capacity of the sensor (how many electrons a photodiode can handle), and the noise floor of the camera.
Dynamic range is the ratio of the useful full well capacity and the noise floor. The higher the full well capacity and the lower the noise, the larger the dynamic range of the camera. Dynamic range is a universal specification where signal to noise ratio can be reported under different conditions.
The effective noise in an image depends somewhat on the scene. If you want to image a darker scene, then the noise in the image is mainly dominated by the readout noise of the sensor. When you want to image a lighter scene, then the noise is mainly dominated by shot noise. Shot noise is unavoidable since it is a fundamental property of the quantum nature of light and arises from random fluctuations in the number of photons emitted from an object.
A higher dynamic range is needed when you want to image a scene with a large amount of contrast with information in both the light and the dark part of the image.
ASYNCHRONOUS CONTROL
Asynchronous control refers to the start and stop of an exposure (and therefore the frame period and integration time of the camera) when controlled by an external signal. This is a camera functionality that can be used when the exposure of the camera has to be locked to an external signal, such as a strobe lighting signal or a conveyor belt move.
This parameter is often important to machine vision users since the camera acts as a slave to the system and the camera needs to act without any delay to the external signal.
OTHER FEATURES
Another camera feature that can be important is the sensor mounting accuracy and the sensor mounting stability in the camera. An accurate and reproducible sensor position eliminates the need for adjustments, maximizing production time, minimizing down time. Besides that, it is important that the sensor position remain stable over time and temperature, even when the camera suffers vibration in the system. The best way to achieve accurate and reproducible sensor mounting is for the sensor to be aligned by the camera manufacturer with a high accuracy in X, Y, Z, rotation and perpendicularity.
The ease to be able to upgrade the system, for example to a camera with a higher resolution, is another important factor to consider. Camera manufacturers that offer a complete line of cameras with similar mechanical, electrical, and software interface will minimize the development time and the risks for a system upgrade.
Since there is no perfect camera for every application, it is important to consider the tradeoffs made by different camera manufacturers to determine the best fit for your system. Data sheets cannot provide all of the information required and some specifications can be reported differently. Since quantification of other important features such as image quality is very difficult, it is often important to test several cameras in your system. Camera manufacturers employ application engineers and have evaluation programs to help you fully analyze different options.
author: By Miriam van Baalen and Gretchen Alper, Adimec