How do you select the right camera for your application and what are the most important criteria for you?
Benjamin Cocquelin (Tiama Inspection): Tiama is using a range of different cameras to inspect mainly glass bottles, but is also focusing on the packaging industry. It’s not just about choosing a camera. When it’s coming to an application the first question is: what do you want to see with the camera, what’s the application and what’s the best wavelengths? Then we choose the right lighting and wavelengths. Thereafter we come back to more traditional questions like resolution, speed… What is really important for the final customer is the easiness of integration and usage.
All camera manufacturers are using the same image sensors, so why are the products different?
René von Fintel (Basler): The sensor itself is just a sensor delivering images with some technical specifications and then there are other technical specifications you might need from the camera or the system which comes on top. Even so today’s CMOS sensors are quite comparable, there are still differences left in terms of what kind of housing you need, what software you’re using, how easy it is to integrate… There are other components which has to play together with the camera e.g. for high bandwidth you need a frame grabber. But how does a frame grabber work together with the camera? Furthermore, there are several criteria which operate just the sensor inside the camera like image quality, stability and how easy it is to handle inside the software API. Also, for camera manufacturers who have quite some experience in integrating sensors, it’s still not that easy. Things like power consumption, stability, temperature ranges are other criteria which are definitely different between camera manufacturers.
Giuseppe Contini (Framos): We see from the interface side that integration is very important. Quality of high level interfaces like Ethernet, USB but also lower level like MIPI or LVDS is very important. The manufacturing itself can also have a big impact on the quality of the image. EMVA1288 reports can give you some information about that. For example, you can see the linearity of the camera which depends on the implementation of the sensor or you see pattern noise or raw noise, that might depend on how the camera is designed or manufactured. Another important point is the reliability of the supplier, their ability to maintain very stable products in the market and their supply chain. Relation matters!
Cocquelin: You need to have a good relationship with your camera/OEM manufacturer to help you with the implementation if you need some added functionalities to program it in specific modes to adapt it to your camera. Compliance to standards is important because then it is easier to compare and you are talking the same language. Standards bring a lot of value to be more effective in the way you use the sensors. There are a lot of aspects regarding supply chain. Tiama is a global player. We need to be able to supply cameras all over the world in a fast way because our industry is requiring most of the time 24/7 inspection and machine vision systems. For this you need a strong partnership to be able to provide the right level of support to the end customer at every time. Of course the price is another important aspect, but that’s not the only one.
How do you or your customers compare camera performance and which role does EMVA1288 plays there?
Contine: Because we distribute image sensor, we are really at the basis of the selection. Most of the time the customers are looking for traditional approaches. Big pixel or very high quantum efficiency (QE) are the most asked question, but some manufacturers do not provide such information. This is always a kind of conversation that we have with the customer to address them on questions like how to select the sensor or the camera itself. What we see is that the EMVA1288 can help. The main driver is the signal to noise ratio (SNR) value. Most of the time, unless you have a very high quantity of light, you run in low light condition. What really matters in those situations is the SNR. We have to discuss this with customers and explain that QE is not always the key feature. Many times you see that the sensors or cameras have a very similar QE. You can have 10% of variation but that could mean that SNR doubles just because you have one electron less of noise. To explain this in detail you need EMVA1288 reports to compare 1:1. Unfortunately, not all camera manufactures have EMVA reports. Users want to understand a bit more about these EMVA reports. It is our job to explain to them how to read the data.
Are you using the EMVA1288 data sheets?
Cocquelin: Yes, but mainly for internal reasons. It really helps us to compare between different cameras. Because we have such a variety of applications within our industry, it will take a lot of time to go through any test or application. Coming back to standards and having some pre-qualification based data on the same measurement brings value regarding our own customers, because they don’t look at all this data. What they’re interested in is the result of the image and if it is working. Everyone has a smartphone and images are all around. Today’s feedback is more linked to pure feelings if it is a good image quality or not. They compare between different camera images and think this one is better than the other. As long as they have a sharp and clear image that you can display they think that they have a good camera. For them, the performance comes from processing and the end result you get from the camera.
Von Fintel: EMVA1288 is a certain quality level which the camera manufacturer should have on their homepages. It makes it easier to make a first choice, e.g. Basler has more than 700 different material numbers alone for cameras. If you’re searching for information in terms of QE or SNR the fixed pattern and spatial noise data gives you a feeling about the quality of the camera. At the moment, EMVA1288 is getting a little bit less important than it used to be. Now there are more important things like how easy it is to integrate, the software (functionalities) you get, or your processing load within the system. We are trying to give more understanding of the overall system performance and how these things work together instead of only saying EMVA1288 data is the one and only. Nevertheless, if you compare cameras to each other, it’s still a good choice for having some indications. Internally we use EMVA1288 in production to see any quality differences we get with the cameras and we use it to evaluate new sensors for new cameras.
Contine: It is also important to consider the lens. The lens and the camera are two separate objects but it’s very important to consider them as a whole an select the most appropriate lens for the given sensor. Depending on the SNR of the image sensor you must select a different kind of lens or, thanks to a proper shading on the pixel, you can have a bigger aperture of the lens. You can easily get twice the light that you have on a different sensor. It’s not only QE, sensor or camera, you have to consider the whole vision system as well, because with a simple trick you can double the light and so you can really double the SNR and get much better images.
What is a better approach: to select a camera manufacturer for all use cases or to identify the best fitting camera per use case from different suppliers?
Cocquelin: All camera manufacturers are trying their best to cover all applications and all needs, but that’s almost an impossible task. Of course, we are working with preferred manufacturers, building long term partnership but it’s really difficult to find one camera manufacturer for all our use cases. That’s why we are having standards: to move in that direction and to be able to build the best system for the application.
What do you think of small and large pixel and how can we use them?
Contine: If you have a lot of light you can use small pixels. The SNR depends on the pixel area and the read noise. If you have a larger pixel you increase the SNR or you have a smaller pixel with a much better read noise. In this case you have the same SNR. So it depends on the noise. The SNR linearly increase with pixel size. The SNR linearly decrease with the read noise. So if you have two electrons instead of one, it means that you have to compensate with twice the pixel size to have similar performance. Most of the time, bigger pixels are more expensive because it’s like renting a house. The base price is per square meter. Users should focus on read noise instead of pixel size unless you really have a lot of light or you need dynamic range. In those cases you need both: lower noise and a high full well capacity that comes with big pixels.
