.::APHESA SPRL - imaging, electronics and software consulting::.

Why do you need a specialized partner to develop a custom camera

Why do you need a specialized partner like Aphesa to develop your custom camera? Many companies have electronic development capabilities and because today's CMOS image sensors are fully digital and easy to use, many of them attempt to develop their own cameras. Even if it is indeed a lot easier today to develop a camera based on a CMOS device than it used to be during the CCD era, there are plenty of pitfalls that explain why cameras are still developed by specialized companies. We will present some of them based on our experience.

Camera interfaces

Industrial cameras, but also cameras in some other applications, use special computer interfaces defined in standards like GigE-Vision, USB3-Vision, CoaXpress or CameraLink. Some of these standards require a lot of development in hardware, firmware and software, as well as certification tests. The development cost and learning curve varies depending on the selected interface type. Most camera design companies already have proven designs using these interfaces, this reduces the development time, the development cost and the risks.

Lenses also have specific mounts, and there are international standards to select a mount for a given image sensor. Some of these mounts require experience in order to design them right.

Image quality

A CMOS image sensor is a digital device, controlled by some registers and pulses and providing an image over a digital interface. Most of these communication and readout interfaces are usually very easy to use as they are similar to memory interfaces or are parallel high speed LVDS buses with or without clock.

However, what comes out of the sensor is not a nice image. It is affected by defects, noise and artifacts. The most common of them are incorrect colors, fixed pattern noise and defect pixels.

Getting the color rights requires algorithms to linearize the pixels, compensate the channel gains, correct for channel crosstalk and apply the color blending matrix. This process also requires color calibration. Some sensors do not even provide a color reconstructed image as they only give access to the raw pixel data and the reconstruction of the color channels from the raw pixels does not work well if only the basic Bayer algorithm is used. Camera design companies have libraries of algorithms for computer or for FPGA to perform all these tasks. The calibration equipment for the color reconstruction process is usually already available in the lab.

Fixed pattern noise is one of the main differences between CMOS and CCD devices as CMOS devices still have more FPN than CCD because of the complexity of the pixel design. Therefore DSNU (Dark Signal Non-Uniformity) and PRNU (Pixel Response Non-Uniformity) must be calibrated and corrected. The designer will also make sure that the device will not reach a mode where the SNR response is dominated by fixed pattern noise, especially the PSNU at low gains (Pixel Saturation Non-Uniformity).

Defect pixels are an extremely wide and complex field of research. With pixel sizes and feature sizes getting smaller and smaller, the probability of having at least one defect pixel in an image sensor is close to 100%, especially at higher resolutions. There are so many types of defect pixels that they are not necessary white or black spots in the image, they may change behavior, exhibit extreme noise or extremely low linearity or may work only at low temperatures. Algorithms that dynamically adjust to the behavior of the pixels are required.

Noise

Image sensors only reach their specified noise performance if they are well operated and correctly mechanically assembled. This means: low noise and highly stable power supplies over time and temperature, precise power supplies, low crosstalk between the high speed data lines and the power lines, well designed filters on the reference voltages and decoupling pins, stable temperature over the whole image sensor, good dissipation of heat or even active cooling.

The mechanical assembly is also very important to avoid some other forms of optical noises like reflections on the lens mount, blurring due to tilt, misalignment and other mechanical issues. The camera in the example image shows a proper design that includes thermal management and mechanical alignment.

EMVA1288

The EMVA1288 standard is now widely adopted and becomes a mandatory requirement in some markets. Most cameras and many image sensors today have an EMVA1288 compliant datasheet.

EMVA1288 can be used to select the right image sensor, but it is not enough. There is more to image sensor selection than only comparing numbers and experienced experts may have good reasons to point you towards another image sensor.

Knowledge of the design trade-offs and design possibilities

We had one bad experience in a custom project because the customer has specified a Bayer color sensor is his design and made us design the camera based on this specification. After we finally learned the details of the confidential application, we found out that a Bayer pattern color sensor was not the right solution. The customer was unaware that other solutions existed for color imaging.

In general a custom camera design company is also familiar with lighting, lenses, interfaces, image processing and other system integration tasks.

But i still want to use my resources to design the camera

Aphesa also provides consulting services for image sensor integration and camera development as well as IP cores for FPGA and training.


 
 

About us

Aphesa develops custom cameras and custom electronics including FPGA code and embedded software. We also provide EMVA1288 test equipment and test services as well as consulting and training in machine vision and imaging technologies. Aphesa works in several markets including industrial, medical, oil&gas and security.