8 Core ARM Processors Including Interface Variety Capture the Image Processing World
There is a new top favorite on the pole position in the race for the highest image processing power in machines and devices: 8 cores, the newest CPU architecture ARM A72, fanless, with all necessary interfaces. The latest VisionBox was therefore christened “LE MANS”, referring to the traditional 24-hour race.
What are the details?
There are market trends which create some challenges for manufacturers of image processing systems. Let’s begin with the market trends in the context of “embedded vision”: Nowadays, vision sensors and smart cameras use ARM-based processors with the Linux operating system. This follows the general trend of processor manufacturers, who develop their individual functionality around an ARM core. An example for this is the AM5728 from Texas Instruments, built into IMAGO’s VisionCam XM.
Nowadays, multicore CPUs are set, whether they are x86, Keystone (Floating Point DSP) or ARM. From using smartphones or tablets, everyone knows, that the ARM architecture provides high computing power with little energy consumption. Not so well-known are today’s supercomputers, which also use ARM processors in a scalable way.
Thus, it seems natural to apply high-performance ARM CPUs in demanding image-processing applications as well.
Series machines create some general challenges: Even embedded x86 CPUs aren’t “forever” deliverable. In this field, long-term availability means 5 … 7 years at the most, which is often not long enough for machine builders. In demanding applications, FPGAs, GPUs or DSPs are applied as accelerators – but this requires more know-how, a second development team, and specialists. OpenCL support could be helpful. Is there a future for CUDA? What do newest DSP cores look like? Time is short, new image-processing systems need to be developed soon. The solution: Focusing on CPUs and finding out whether these computers could be used in a scalable way. The advantage: Own C++ code, OpenCV or Embedded Halcon both work, the development environment in the PC-world is familiar: Visual Studio, as well as the debugging methods. Young generations of engineers are familiar with the Linux operating system, the keyword is Raspberry PI, as well as experiences with IoT.
An image-processing computer does not only compute – nowadays, it also controls more and more interfaces: Cameras do not only send their data through the interface, but can also be triggered accurately to the micro second and provided with power through the same cable (Trigger over Ethernet / Power over Ethernet / Camera Link). The lighting is current-controlled and synchronized to the camera, ideally for every flash individually. Rotary encoder ports, digital in- and outputs and fieldbusses communicate with other machine components, partly even without central PLC. Nowadays, all of these functions are elements of computers in the sense of an Embedded Industrial Vision & Automation Computer.
“LE MANS” – Scalable Embedded Vision & Automation Multicore ARM Computer
What a tongue twister! Still, IMAGO’s development team has succeeded in combining their nearly 20 years of experience in embedded vision with the newest processor trends and interfaces: The newest flagship of the processor developer NXP (formerly Freescale) is called into action. A72 is the newest architecture from ARM, it is made for 24/7 operation as an 8 core processor with 2 GHz. With 10 years of delivery capacity, long-term available systems can be realized. Equipped with two 10 Gbit/s interfaces, it does not only provide the connection of the newest high-speed cameras (AIT exposure, Emergent, …), but also the computer-to-computer communication. Frame grabber, I/O interfaces and fieldbus can be connected through the PCIe bus. In the case of higher requirements to the computing power, a second VisionBox LE MANS can be connected via 10 Gbit/s Ethernet, for example as a compact version without PCIe slots. Here, the principles from supercomputers are realized, where mostly only scalable CPU computing power is cross-linked. For multi-camera systems, there are 4 ports per PCIe card for GigE vision cameras or 2 ports per PCIe card for camera link.
The Linux OS has a lot of advantages in the field of embedded vision. For instance, it is possible to apply real-time Linux – with better runtime performance for higher real-time requirements. Additionally, in contrast to Windows Embedded, for Linux only the software, drivers and add-on programs necessary are installed one by one. In this way, it is easier to keep track. Windows Embedded can be reduced in its combined functionality, although over the years it has become clear that nobody actually demands this. The typical Windows OS risks remain and need to be mastered by the application developers.
Another interesting aspect is that Linux also is set in the cameras, especially concerning preprocessing. For example, the 10 GigE line scan camera “exposure” from AIT runs internally with Linux – functionalities can be shifted between camera and embedded computer.
New applications need an image processing library, unless it is available for the own company. Already 1,5 decades ago, the idea came up to “embed” Halcon functions into DSP-based VisionBoxes. Meanwhile established on the market, Halcon Embedded runs on the ARM A72 and partly uses the NEON SIMD accelerators. Demanding applications can thus be realized in a time-saving way.
Concerning the keyword industry 4.0 – but also the necessity for machines to become cleverer through image processing – faster running processes – higher resolutions, more images per second – intelligent, meaning more complex algorithms: the VisionBox LE MANS offers the next generation of supercomputers for machine vision. In this way, IMAGO’s tradition of bringing scalable embedded computing power into machines continues. Whether in the control cabinet or next to line scan cameras directly on a crossbar, nowadays IMAGO’s computers are already in use worldwide.
You can see the new flagship live at the VISION 2016 at the IMAGO booth B41 and at the AIT Austrian Institute of Technology booth D82.