At the interface between microprocessors and video

In 1986, Matrox Imaging introduced a device capable of performing 3 x 3 convolutions on a 640 x 480 pixel image in real time using a graphics chip. It was the first PC-based image processing accelerator, and it was “unheard of in our industry at the time,” recalls Sam Lopez, senior vice president of sales and marketing at Matrox Imaging.

The device was born thanks to the experience of the co-founder of Matrox, Lorne Trottier, with graphics cards. He came up with the idea of ​​using a backwards graphics engine to drive memory, and using the chip to run convolutions and histograms and implement image processing algorithms. PC processors were extremely slow in 1986, and here was a vision engine that didn’t rely on the host processor for processing.

“That’s what I love to do, looking at technology and figuring out how to apply it,” Trottier said. Imaging and machine vision Europe. “We were already a bit into image processing, so we knew the image processing needs. I knew about this particular graphics chip, so I figured out that I could use it for image processing.

“That’s the kind of stuff our creatives do all the time at Matrox,” he added. “That’s the fun thing about it, there’s always new technology coming out and harder problems to solve, and figuring out how to use that technology in a smart way to solve those problems is what we thrive on here at Matrox. .”

The MVP-AT accelerator – which was used by NASA in the early 1990s, for example, in a satellite data analysis software package called PC-Seapak – is just one example of the pioneering work being done at Matrox.

Trottier and his co-founder started the company in 1976, in the early days of the microprocessor. The idea was to build an interface between microprocessors and video. “Bi-directional microprocessor input and output was the central idea that started the company and has been a theme throughout its history,” Trottier said.

Matrox’s first product was called Video RAM, a controller for a microprocessor to display computer-generated alphanumeric data. It was called video RAM because ACSII text, the binary code for characters, was mapped to memory.

“You didn’t need to know anything about the video,” Trottier explained. “If you knew RAM, you could plug RAM into your microprocessor, and as soon as you used ACSII text, it would show up automatically.”

Years later, another video RAM technology was developed for graphics cards. At one point, Trottier received a call from the president of Micron asking about Matrox’s video RAM to see if Trottier could help with a patent dispute Micron had with Texas Instruments over VRAM. “I told him the only similarity between our VRAM and the one you’re talking about is the name,” Trottier said.

Matrox started life as a personal project between the two founders. The company’s first telephone line was installed in the Trottier family home, with Lorne’s mother acting as receptionist. Two months after Matrox was founded, Trottier visited a computer show in Atlantic City in July 1976, collecting data sheets from a number of small start-ups in the field of early personal computers, among them a data sheet from the ‘Apple I, which he still has in his archives.

Matrox’s first product, Video RAM, in 1976. Credit: Matrox

His pioneering spirit led Matrox to develop some of the first frame grabbers, a foundation that is evident today – Matrox still manufactures frame grabbers. ‘Exactly when image processing started [for Matrox] is hard to say,” Trottier said. The first capture card designed by Matrox dates back to the late 1970s, before the release of the PC, for Intel’s multi-bus card-level computer. “Hardly anyone had done digitized video because it was high bandwidth at the time,” he recalls. “We used an A/D converter that cost $600 a chip – back in the 1970s – to digitize the video to our capture card.

The price of these chips quickly dropped, and once the PC was introduced in the early 1980s, Matrox was the first to build a frame grabber card for the PC. “Exactly where the machine vision kicked in along the road is a bit of a blur,” Trottier said. ‘These capture cards were used for all sorts of things at the time; there were all kinds of research projects going on and machine vision and image processing was one of them.

“Back then, image processing systems were mini-computers and cost a fortune,” he continued. “We were one of those pioneers that brought the prices down and made it much more accessible.”

In the early 1990s, the company split into three divisions: Matrox Graphics, providing graphics solutions; Matrox Video, for the broadcast industry and digital video editing; and Matrox Imaging, which focuses on component-level solutions for machine vision applications. The unified thread that underlies the Matrox model remained the original notion of interfacing between microprocessors and video.

In the early 1990s, Matrox Imaging decided to develop a library of imaging software, launching the Matrox Imaging Library (MIL) in 1993. At that time, the company had many frame grabbers different, and the software to support the MVP-AT accelerator was getting hectic, Trottier said, so MIL was created as a uniform library that worked across all Matrox products. “That was key and it was one of the reasons we kept our clients for so long,” Trottier added.

Today, Matrox Imaging’s portfolio includes frame grabbers, smart cameras and software libraries. Credit: Matrox

More recently, Matrox Imaging released Matrox Design Assistant software, which eliminates the need for programming by using flowcharts to create a vision application. MIL and Design Assistant now have deep learning functionality, which can solve problems that could not be solved with conventional rule-based image processing tools.

One of Matrox Imaging’s newest cameras is the Matrox AltiZ 3D profile sensor, featuring a dual optical sensor design and data fusion capability. The dual camera sensors reduce occlusions found in more traditional laser profilers to give fuller 3D coverage of a scene.

“Combining 2D and 3D, classic image processing, deep learning, with robotics, allows you to create extremely powerful solutions,” Trottier said. “When you integrate these elements together, you get something surprisingly powerful.”

Trottier said there are new areas that have high demand for computer vision, citing the electrification of automobiles as an example. He said, “The simple assembly of batteries and batteries is a whole new emerging sub-industry,” and that Matrox Imaging is attracting customers in this area.

Lorne Trottier co-founded Matrox and has now acquired full ownership of the company. Credit: Matrox

Lopez noted that reducing the cost and reducing the complexity of implementing machine vision systems has “opened up many new opportunities in industries where they couldn’t afford to put vision in the past.” He said standards, whether video standards but also communication protocols to communicate with other devices like robots or PLCs, reduce the complexity of vision systems. Matrox is active on a number of standards committees, including Coaxpress and GigE Vision, to help define and develop these standards.

Lopez added, “There will be more competition in the future as technology becomes more accessible and easier to use. There are many good ideas that deserve attention. We don’t even neglect start-ups, which could come up with some interesting things in the future.

Trottier said his advice to start-ups is to work with companies to get exposed to real-world imaging needs. “It’s the catalyst,” he said. “If you become aware of needs and aware of new technologies, you will discover new solutions.”

In 2019, Trottier acquired full ownership of Matrox, declaring a renewed commitment to customers, suppliers, business partners and employees.

“For 45 years, to remain viable and profitable, we had to constantly reinvent ourselves,” said Trottier. “It’s one of the things I love about this industry. I’m a techno-geek myself; I started messing around with electronics, building crystal radios when I was child. It’s a sequel of sorts. Most of the engineers we have working for us have the same geeky love of technology, and we like to stay on the cutting edge and find out what the latest technology is and how it works. can be applied to solve real-world problems.

“There has been no shortage of innovation and new technologies in machine vision,” he continued, “and we’re at the forefront of some of these things, including things like 3D …and deep learning – we have customers in many projects applying this [deep learning], which is very exciting.