AI at the edge

(Unfortunately, this presentation is currently only available in German.)

In naher Zukunft wird KI zu unserem Alltag gehören. So werden selbstoptimierende Maschinen, Spracherkennung sowie diagnostizierende Medizingeräte uns täglich unterstützen. Die zugrunde liegenden Algorithmen sind extrem leistungsfähig aber auch leistungshungrig. Was steckt wirklich hinter der KI? Wie kann diese «at the edge», also eingebettet im Endgerät laufen, ohne Daten in der Cloud zu rechnen? In diesem Talk erklären David Gschwend und Florentin Marty, wie Ingenieure von SCS für Ihre Kunden erfolgreich KI in eingebettete Systeme packen.

Dieser Vortrag wurde am 9. Juni 2020 im Rahmen der Embedded Computing Conference 2020 mit dem “Audience Choice Award” ausgezeichnet.

You have an idea, a project or a problem you’d like to discuss? Get in touch with me now!

    Contact David Gschwend

    Protected by reCAPTCHA. The Google Privacy Policy and Terms of Service apply.

    Around 3 years after the first ideas, we were able to realize the rc_visard, a 3D-stereo camera for robots, on behalf of the project partners KUKA and Roboception within just 9 months. For this project SCS could rely on their extensive expertise in the electrical development and camera systems. This camera with the on board NVIDIA Tegra K1 chip allows realtime 3D measurements and positioning and enables a multitude of further developments in robot automation. For the development of the sensor, Roboception was focusing on intuitive usability and seamless integration with standard interfaces. At the Hannovermesse in April 2017, Roboception presented the solution to the general public and has worked since then on the ramp up of serial production. From end of September 2017, the rc_visard solution will be delivered to customers across the world. We congratulate for the successful product launch.

    Further information:

    Dr. Uwe Franke from Daimler introduced together with Felix Eberli the potential of deep learning for autonomous driving during this year’s first tech event. The presentation covered the impressive technical developments, driven by Convolutional Neural Networks and increased computation power. These methods are even nowadays superior to the human driver in particular situations such as night vision or mirrored images due to rain fall. Over the next years, we can expect significant progress in the field driven by the technical capabilities.

    Wir demonstrieren ein Entwicklungssystem für Stereo-Kamera-Fahrerassistenzsysteme auf der Basis des Zynq All Programmable SoC. Optimiert für die Entwicklung von Bildanalysealgorithmen demonstriert dieses System sowohl eine Bildentzerrung als auch die Berechnung des SGM Stereo um die Lage und den Abstand von Objekten abzubilden.

    Um mehr zu erfahren, besuchen Sie uns an der Embedded World in Nürnberg.
    Wir sind zu Gast am Xilinx-Stand 1-205.

    SGM Stereo on SCS Zynq Box


    Modern cars are using more and more cameras in order to provide a better and more comprehensive display of their surroundings. Camera manufacturers need reliable measurement technology to design such systems. The SCS measurement technology box makes it possible to record and play back up to six LVDS cameras. The ‘Recorder System’ developed by SCS has been adopted by automobile OEMs and tier 1 companies for the development of cameras and surround systems.

    Filters for EB Assist ADTF for recording and playing back are available, as well as a stand-alone GUI developed by SCS.

    The system consists of the Spartan 6 FPGA card developed by SCS and a customised adapter card for the corresponding camera system. This allows precise hardware timestamps to be recorded and multiple cameras to be synchronised with each other. Furthermore, algorithms can be calculated directly on the FPGA in order to relieve the measurement technology computer. The data is transferred to the measurement technology PC via Ethernet connections.

    Thanks to its system structure, SCS can adapt it very quickly to your Maxim, TI or National Semiconductor serialiser.

    A German OEM commissioned us to make a high-resolution and exact calculation of the 3D structure of a vehicle using video images of two cameras. Vehicles built in the future will be able to assess events in their immediate vicinity with the help of such information, because they will be able to calculate and determine the position of vehicles, pedestrians and all types of obstacles in advance.

    Pentium4-PC is slower by a factor of 50

    The complex image processing software was developed on a Pentium Dual PC by our client.  The objective of the project was to clarify, in terms of the pre-production, whether the algorithm is suitable for implementation on an automotive-compliant FPGA.

    The Intelligent Solution

    SCS opted for a development board with a Xilinx Virtex4 FPGA which exchanges the image data via a PCIe. After analysing the algorithm, the following development steps took place: Design, implementation and testing on the FPGA. Further optimisations were carried out in order to reduce the electricity consumption and to allow for use in an automotive-compliant Xilinx Spartan 3A-DSP FPGA. The computing performance of the FPGA solution exceeds that of the PC by a factor of 50. With the prototype as a PCIe card in the test vehicle, the researchers could verify the practicability of algorithms in vehicles of the future. The FPGA solution will be employed from 2013 onwards in series production vehicles.

    SCS is a member of the Xilinx Alliance Certified Third Party Program and offers SCS products and services with Xilinx FPGAs. This includes a wide range of hardware and firmware developments as well as consulting services. The development cycle for new products is significantly accelerated on account of our long-term cooperation with Xilinx.

    The JPEG IP core for FPGAs, developed by SCS, allows compressed Ethernet packages to be received and then decompressed. The decoder has been optimised for low consumption of resources for a Xilinx Spartan6 or Zynq FPGA and is already in use by an OEM and a tier 1 company.

    The JPEG Decoder has the following properties:

    • Processing rate of up to 140 MSamples/sec on Spartan6 FPGA
    • 12Bit / 8Bit version available
    • Four Huffmann tables (fixed or extracted from header)
    • Up to 8 quantisation tables
    • Support to decode several interleaved image stripes
    • 3 color components
    • Support 1 scan configuration and YUV 4:2:0 (Different format on request)
    • Supports any image size up to 64kx64k
    • Supports restart markers

    The research team of an established automotive company in Germany is developing highly complex image processing algorithms for new driver assistance and safety systems. An example of this is the calculation in real time of several thousand displacement vectors in an image sequence (optical flow) without restrictions to vector length. With the aid of this information, vehicles will be able to assess the situation in their immediate vicinity in the future, distinguish between moving and motionless subjects, and calculate the position of other road users in advance.

    “Automotive”-compliant DSP instead of the PC

    The complicated image processing software has been developed on a Pentium Dual PC. The objective of the project was to clarify, in terms of the pre-production, whether the algorithm is suitable for implementation in an automotive-compliant DSP. Furthermore, an adaptor board was created which enables a camera to be connected to a DSP board.

    The Optimum Solution

    SCS opted for a development board with a Texas Instruments DSP (DM6437) and a plug-in adaptor board for connection with the camera. Porting the software onto the DSP was then followed by elaborate optimisation, whereby computing performance was increased by a factor of 43. With the prototype as PCI card in the test vehicle, the researchers could verify the practicability of the algorithms in vehicles of the future.