congatec, Basler and NXP Semiconductors have developed a function toolkit for deep learning applications in retail. The platform is a proof-of-concept, using artificial intelligence (AI) to fully automate the retail checkout process

computing resources needed for deep learning inference algorithms. Intel, on the other hand, offers a distribution of the OpenVINO toolkit that optimises deep learning interference while also supporting many calls to traditional computer vision algorithms implemented in OpenCV – in other words, it provides a total integrated package. Ultimately, by supporting FPGAs and

the Intel Movidius Neural Compute Stick, Intel aims to use not only the expensive GPUs from AMD or Nvidia, but also to present other in-house alternatives for the inference systems. NXP offers answers for the use of AI as


he toolkit demonstrates the possibilities of vision technologies in

embedded applications and how they can simplify our daily lives. The kit is application ready, providing everything necessary for training automated checkout systems. Goods have already been trained for

automatic video recognition without the use of bar or QR codes. This way, goods such as fruit or vegetables, which cannot be identified by a code, can now also be checked out and a symbolic invoice total can also be created. This illustrates that modular embedded systems offer all the basic features needed for integration into existing checkout systems that also map all payment functions. Such vision-based systems open up new perspectives for retail applications. Retailers benefit from lower labor costs

and a significantly improved shopping experience through instant checkout, shorter queues and 100% checkout. However, providing such solutions requires preparatory work that OEMs serving the retail market cannot accomplish from a standing start. This is why they need embedded

partners, who collaboratively provide them with application-ready platforms for the integration of camera and AI technologies at the embedded level. The effort this involves is not

significantly different from the effort 32 NOVEMBER 2020 | ELECTRONICS

required for the integration of other peripheral components. So it shouldn’t really present any major challenges – if it wasn’t for the need to integrate additional AI technologies that don’t require costly and lengthy training in server farms, but that get by with just a few images and can even be trained in the embedded system itself. There is also invariably some effort associated with developing application-ready platforms on the basis of ARM technologies, since these must be adapted to the application-specific requirements. So regardless of which processor

technology is used, there is always a need for OEMs to bring the sum of the individual parts to series maturity as smoothly as possible. Ideally, they find a supplier who can provide them with specific solution platforms that already offer more than the sum of the individual components, as this allows them to fully concentrate on new applications. The challenges begin with the

integration of MIPI-CSI based camera technologies, for example. While they are standard for ARM-based technologies, x86 platforms require special integration effort. AMD and Intel also have quite different software support strategies for AI technologies. As with OpenCX/CV, AMD relies on open source solutions such as ROCm and TensorFlow to support the heterogeneous use of embedded

Figure 1:

The congatec, Basler & NXP Vision-Platform

well, with the eIQ Machine Learning Software Development Environment. Next to the automotive segment, this also targets the industrial environment. It includes inference engines, neuronal network compilers, vision and sensor solutions, and hardware abstraction layers, providing all key components required to deploy a wide range of machine learning algorithms . Based on popular open source frameworks that are also integrated into the NXP development environments for MCUXpresso and Yocto, eIQ is available in the early access release for i.MX RT and i.MX. As these three different AI approaches

of the semiconductor manufacturers clearly indicate, OEMs have different implementation requirements for their applications depending on the chosen solution path. But in any case, the embedded computing hardware must be prepared for whichever software solution is used, and this requires careful selection of the individual hardware components, which is why cooperation between semiconductor manufacturers and embedded computing providers is crucial. OEMs using such application-ready

solution platforms benefit from significantly reduced development effort, since many functionalities have already been tested and the interoperability of the individual components has been validated. / ELECTRONICS

Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46