Panel Discussion: How to choose a camera?

Isn’t the full well capacity more important for a high SNR than the pixel size?

Contine: The maximum SNR value depends on the pixel size. But if you are looking for the maximum SNR and you have a lot of light you can also use small pixel, because noise is negligible in low light condition. In general, the read noise is more relevant. In some way the absolute threshold sensitivity tells you how much photons you need to overcome the noise. The pixel size is not that relevant if you run with good lighting condition. But as mentioned before, the read noise can easily be different between one or two electrons, meaning that you have a factor of two. But increasing the pixel size by factor two, then it’s a very different sensor because at the same resolution you have a much bigger image sensor. Therefore, the read noise is more relevant than the pixel size. The pixel size matters when you really need high dynamic range, if you have both a dark and a bright situation in the same image.

Von Fintel: It’s sometimes a philosophic question of big or small pixels. We had seen a trend toward smaller pixel, especially for backside illumination sensors. It stopped around 2.5 micron. But the trend towards higher resolutions were only enabled by smaller pixel.

In the end, does it always come down to the price when choosing a camera or can other criteria outweigh the lowest price?

Von Fintel: There are always criteria which can outweigh the price. Of course, you have to meet a certain price range because each manufacturer has their own bill of material, and there’s a certain cost that they will spend for a camera. Nevertheless, at the end, in the overall system it matters whether you can reach a certain performance. You’re willing to spend more if you have specific needs on temperature ranges or the software side. There must be a certain prince range but it’s certainly not 100% of the decision.

Cocquelin: Price is always a question, but the real question is price for value. When you have an embedded system, you have less constraints than at a full camera system. In the end, it is the same question for the end customers using machine vision: what value does it bring to me and is it worth it? Sometimes you can use a camera and sometimes it’s not possible because the value is not really there. It’s not a pure price issue, but a value proposition depending on the application.

What about the usability of cameras and what role does software / SDK play for choosing a camera?

Cocquelin: SDKs plays an important role when you want to discover a new technology but also facilitate the integration afterwards in your application. It has always been the role of machine builders to start developing your own interface and software to be able to adapt to different kinds of cameras. But today we see more and more interest of having already built in SDKs to increase the time to market or increase the integration of new cameras. It’s important to have a good software/SDK especially for new technologies with a lot of new functionalities like polarized cameras. Before that you used to use maybe one image and now you have a lot of more information that you need to process and there are pre-functionalities already made in some SDKs e.g. for multi spectral cameras.

Von Fintel: Basically, the integration part is important. It’s quite useful because then you can pass a test, especially with custom made features which are not standardized. There are very sophisticated advanced customers like Tiama but there are also customers who have less knowledge. For them it’s quite interesting to have other software features and easy viewing tools or even graphical programming, which we had introduced just recently to our SDK. The users want image analysis functions, maybe even integrated into the camera. Backwards compatibility is an important issue: an SDK should support the latest products, but also support all relevant older products backwards.

Which new camera feature is most demanded from your customers or which new camera feature do you need developed by camera manufacturers the most?

Contine: The biggest wishes from many (embedded) users is to have an off the shelf camera. Most of the time, the transition is happening, but it’s often forgotten that a camera is a battery included system that they plug and should work immediately. For embedded systems, users have to deal with functions like debayering, color correction, tone mapping …. This is the gap that is going to be filled. In the future, System on Module (SoC) with ISP functionalities and any kind of digital signal processing in the middle between the image sensor and the system will help to have image sensors not only in machine vision, but almost everywhere.

Von Fintel: I see three trends from customers: umber one is an ongoing trend towards higher resolution and higher bandwidth. With that going along some system constraints like how do I deal with CPU load and how can I offload some of this CPU load to a frame grabber or a FPGA preprocessing. Number two are system related things which are not only to be solved by the camera, but maybe by a combination of camera and other components like autofocus or the integration of smart light functions, where in-the-system-illumination and camera can communicate with each other. The last topic is a trend towards different spectral ranges outside the visuals or SWIR range. This becomes more and more affordable and is getting into classical machine vision cameras.

Cocquelin: From a technology point of view, we are moving into a good direction like increasing resolutions, increasing speed, increasing high dynamic range and the possibilities to use multiple wavelengths at the same time. What we should talk about is to improve troubleshooting because this is a pain that can be huge for customers. Our systems are reliable and are working well, but from time to time it fails and this happens to everyone and in that case we need to be good not only in operations, but also good in the diagnostic and failure analysis.

Seiten: 1 2Auf einer Seite lesen