For machine vision, the global joint standardization efforts of the last 15 years have been the key to the industry's success. Since 2009, international machine vision associations have been working together under the umbrella of the so-called G3 initiative to drive forward existing standardization initiatives and open up new fields of standardization both within machine vision and in cooperation with adjacent vertical industries. The European Machine Vision Association (EMVA) is one of the five G3 partners. The EMVA currently hosts four machine vision standards, including GenICam (Generic Interface for Cameras). GenICam is considered the backbone of machine vision, as most CameraLink and all GigEVision, USB3Vision and CoaXPress cameras today use GenICam as their register and operating interface.

Megatrend embedded vision

One of the technical megatrends in the image processing industry is the ongoing miniaturization of hardware and processors, which even enables system-on-chip (SoC) solutions. By combining a processing board with a small camera, it is possible to design such a very compact image processing system. These so-called embedded vision systems are of great interest for industrial image processing - as well as for many other industries. Most embedded vision applications run on Linux with MIPI cameras and APIs such as V4L2, GStreamer and further developments such as libCamera. However, a major drawback of such user interfaces for machine vision applications in mass markets is that they do not provide sufficient support to holistically address problems in the organically grown machine vision environment. In addition, the variety of camera hardware interfaces, camera APIs and possible extensions leads to the problem that a separate driver must be programmed for each camera module connected to an SoC. This effort is added to the number of MIPI modules most commonly used in embedded vision systems and multiplied by the number of SoC providers on the market - an almost unmanageable amount of programming resources.

A large part of these programming costs could be avoided through joint efforts for a generic user interface to the SoC camera interfaces; this would also result in lower development costs for manufacturers and customers in the embedded sector. Therefore, EMVA and GenICam have joined forces for the EMVA standard 'emVision' to meet the requirements of the established machine vision industry for high-volume products from other machine vision markets.

Bridging the gap between mass market and industry

While working on a potential embedded vision standard, it became clear that it would be beneficial to build a bridge between the industrial world, its application requirements and areas of new vision tech applications - such as those used in cell phones and tablets, cameras in cars or imaging linked to AI.

In 2020, the Khronos Group therefore approached EMVA with the idea of working together on an embedded vision interface standard. Khronos worked on an OpenKCAM standard from 2013 to 2015, which was not published, but whose ideas live on in the 'Libargus Camera API', which is widely used in image capture for AI applications. The Khronos Group also offers established standards such as OpenCL, OpenGL and OpenVX. They are used in imaging applications that are developed and implemented at chip level for mass market products by Khronos members such as Google, ARM, Amazon, Intel and AMD.

With this in mind, Khronos and EMVA partnered in February 2021 to establish an Embedded Camera API Exploratory Group open to all interested participants free of charge. Its purpose was to explore the industry's interest in creating open API standards for the control of embedded cameras and sensors. In this group, 60 companies with 145 representatives discussed use cases of embedded vision and requirements for new interoperability standards in many industries. The aim was to increase the acceptance of embedded vision in many markets by reducing development costs and time-to-market. Sensor and camera manufacturers, silicon providers and software developers involved in image processing and sensor technology took part in the initiative. At the end of November 2021, the exploratory group reached consensus on the development and introduction of a new standard hosted by Khronos and managed under the joint logos of EMVA and Khronos. The 'Call for Participation' to the industry was sent out in December 2021 and the first meeting of the new Khronos Camera Working Group took place in March 2022.

The author: Werner Feith is Standard Manager at EMVA.

© EMVA

The collaboration between EMVA and Khronos will continue in the upcoming development phase as an official joint initiative between the two organizations. This means that EMVA can directly participate in the working group, influence the development of the design and contribute to the new standard. This entire process of collaboration between leading associations and their members marks a new direction in embedded vision standardization. The EMVA has paved the way for its members to participate in the standardization of embedded vision in a clearly defined way. The EMVA standardization working groups GenICam and emVision are an important part of the Khronos Camera working group. As a result, the Khronos working group receives feedback from the world of industrial image processing and its already established standards. In turn, the EMVA working groups are able to optimize the GenICam standard based on the information received as feedback from the collaboration with the Khronos working group.