Search
Close this search box.
Mark Williamson Vdma

Machine Vision: An Enabler Driven by Megatrends

Mark Williamson is Chair of the board of the VDMA' machine vision group. In the following article, he analyses the megatrends that have shaped and will shape the future of the machine vision industry.

MACHINE VISION

Mark Williamson
Chair of VDMA’ machine vision group

As the world embarks on the next megatrend “Artificial Intelligence” which is set to change the way we all work this article looks back on how megatrends have driven the development of the machine vision market and looks forward on what we can expect in the up and coming years and decades.

When starting out in the machine vision industry in 1989, machine vision was expensive and complex with limited viable real world applications. In fact, one of the first production systems I worked on was in the Nuclear Fuels industry where manual inspection was impossible thus justifying the high cost.

Early PC’s were having an impact in working practices however they just did not have the memory, processing power or bus speeds to acquire image data in real time requiring dedicated rack based pipeline processors costing many tens of thousands of Euros for rather basic functionality.

“The most profound step in the evolution in machine vision and robotics will take place later in the 2020’s where AI based vision system design assistants will truly ease the deployment of robust vision solutions”

MEGATREND 1: Advancements in computing power 1990's onwards

The megatrend that first brought machine vision in to the mainstream was the advancements in computing power seen in the 1990’s onwards. With increasing processing power and faster bus speeds introduced with the PCI bus, PC’s were able to acquire images in real time into PC memory and do more real processing which meant the external pipeline processors were replaced within PC cards, often with accelerators for processing intensive tasks.

We also saw the emergence of mainstream CCD sensors developed specifically for machine vision rather than broadcast or CCTV cameras use, giving the ability to trigger an image exposure at the exact time needed. This megatrend has driven the evolution of machine vision, allowing more complex but more robust algorithms to be developed and with more available memory opened up the possibility of 3D machine vision in real world applications from the early 2000’s.

The millennium also saw the introduction of the first standardised camera interface for machine vision “Camera link”. Before then, each camera and interface card required a custom cable, be it an analogue or digital connection.

MEGATREND 2: Networking and interoperability - 2000's onwards

While networking was expanding prior the new millennium the data rates were not suitable for video data until the introduction of Gigabit Ethernet. The American Imaging Association (AIS) jumped at the chance to leverage IT technology and introduced its second machine vision interface standard after Cameralink, GigE Vision, which allowed easy and low cost connection to cameras using standard networking technology.

The beauty of this technology is that the standard follows the continual improvements in networking speeds. While peripheral interfaces for computers such as USB2 were able to handle images in real time, no standard was available for plug and play machine vision camera control. Products were launched which required a proprietary driver for the manufacture enabling lower cost solutions but were tied to a specific manufacturer. The exception was Firewire, which offered a window into the possibilities of plug and play but were generally more expensive and had limited functionality.

During this time, the European Machine Vision Association (EMVA) was formed and collaborated with the AIA to create an interface independent camera control standard “GenICam” (Generic Interface for Cameras) with the goal of making camera interfaces fully plug and play irrespective of the physical connection. Today this camera abstraction layer is used across all current interfaces including the original “Cameralink” and its higher speed successor Cameralink HS, Coaxpress , USB3 Vision and now for embedded applications MIPI-CSI-2.

“The adoption of network based PLC control also gained adoption with the use of protocols such as Ethernet IP which removed the need for dedicated hardwired I/O and gave a wider integration in the factory environment”

MEGATREND 3: Cloud, IoT and the birth of industry 4.0 2010's onwards

Supported by the two previous megatrends, the emergence of vision systems integrated directly into cameras started to become mainstream. Some companies heralded these smart cameras would be the end of the PC based vision system. Typically these cameras contained ARM based processors and Linux operating systems with Web based interface which could be deployed with little machine vision experience.

While they expanded the use of vision and reduced the total cost of ownership, the need for knowledge to create a successful solution was still key and deployment without the knowledge often created an unreliable system giving machine vision a bad name. However, while the use of these small vision systems grew rapidly the demand for complex vision systems also grew and PC based vision systems with multiple cameras and high speed inspection of continuous materials continued and expanded also into 3D.

The adoption of network based PLC control also gained adoption with the use of protocols such as Ethernet IP which removed the need for dedicated hardwired I/O and gave a wider integration in the factory environment. Robot controllers were part of this networking of automation devices which enabled easy connection of the vision system and robot controller introducing Vision based robot guidance, a market that continues to expand enabling robotic pick and place of random objects.

With the price of memory and processor power still falling, resolution and 3D data models continued to enable more applications to be realised with ease. The cloud started to become a buzz word and architectures with distributed processing became a reality. This distributed whole factory communication labelled “Industry 4.0” required further interoperability standard and OPC foundation was formed creating the OPC-UA standard allowing all information levels to communicate such as ERP systems, manufacturing planning, automated build and product inspection to be completed with whole company visibility from cloud to real time factory processes.

Both the Machine Vision and Robotics industries adopted this platform introducing companion specifications to enable their domains to interact with other systems easily bridging the divide from factory control to cloud based planning and reporting.

“When starting out in the machine vision industry in 1989, machine vision was expensive and complex with limited viable real world applications”

MEGATREND 4: Artificial intelligence, the disruption is just starting

In recent years, deep learning techniques (convolutional neural networks) have developed significantly as processing power of the first megatrend continue to advance, in particular the power of GPU accelerators driven by the gaming industry.

For the first time, these techniques have enabled inspection, grading and classification or varying organic products, previously not possible using traditional rule based algorithms.
This has opened up many new market possibilities for example in food and agriculture, where products are never the same while the surroundings are uncontrolled. Combining deep learning machine vision with autonomous farm machinery new possibilities are emerging to optimise and improve efficiency and quality in farming.

The challenge with these techniques is the high number of training images required to create the classifiers, especially where your need to train on a defect that only happens occasionally. Research in this area is developing quickly, not only use of pre-trained classifiers to reduce training but also with the concepts such as anomaly detection which the classifier is only trained on good product and highlights when it sees a variation that’s not been seen before. We are also starting to see in line retraining where items getting close to a decision boundary are identified for human judgement and then adding to the training set.

However, all this is just looking at the evolution of technology in machine vision and this latest AI megatrend will have a far bigger impact. Within the last year or so technology such as ChatGPT and the recently announced Microsoft Co-Pilot for Office will revolutionise all aspects of work removing repetitive tasks and speeding up others such as researching and writing texts, analysing spreadsheets and suggesting responses to emails. Probably a bigger change to working life than the introduction of the personal computer. The biggest cost and resource in developing reliable machine vision solutions remains knowledge. Novice users don’t know what they don’t know.

The most profound step in the evolution in machine vision and robotics will take place later in the 2020’s where AI based vision system design assistants will truly ease the deployment of robust vision solutions opening up even more possibilities for our industry.

SUSCRÍBETE A NUESTRA NEWSLETTER

<strong>Introduce tu e-mail y recibe mensualmente las últimas noticias del sector. Consulta tu bandeja de entrada, y revisa la de SPAM; debes confirmar tu suscripción en el correo recibido para poder empezar a recibir nuestra NEWSLETTER </strong>

    X
    NEWSLETTER