VISION 2018 – Machine vision at the edge of a new era
These were the undisputed buzz words of the show and were also heavily represented at the Allied Vision booth.
If you had any doubts that embedded vision is coming, a visit at the VISION show should have discarded them for good. Many exhibitors showcased their imaging solutions for embedded systems. So did we at Allied Vision with “Rethink embedded vision” as a booth motto and a strong focus on our new Alvium Series, a camera platform designed for easy integration into embedded systems.
Looking for the right way to embedded vision
You could observe different approaches from one booth to the other, though. Some exhibitors insisted on the compatibility of their traditional USB camera models with embedded systems. Others presented bare board modules with USB or MIPI CSI-2 compatible interfaces and stripped-down functionality for low-cost applications.
In the first case, the approach was to integrate a machine vision camera into an embedded system with an ARM processor, a solution that delivers all the performance benefits known from machine vision cameras in terms of durability, image quality and on-board image correction, yet failing to fulfill the specific requirements of embedded systems in terms of size, weight, power consumption and cost.
At the other extreme, simple bare board modules offer the benefits of industrial durability and high quality sensors known from the PC-based machine vision world. However, on-board image corrections and preprocessing got lost on the way in an attempt to meet the price expectations of embedded system developers used to low-cost sensor modules from the consumer electronics industry.
Building bridges between industrial machine vision and embedded systems
With the Alvium camera series, Allied Vision introduced a completely novel approach with a camera platform designed from the beginning to combine the performance of a machine vision camera with the specific requirements of an embedded camera module for size, weight, power consumption and cost. The only way for us to achieve this was to develop a new processor technology that can accommodate on-board image processing without the drawbacks of an FPGA in terms of size, power consumption and cost. The ALVIUM® Technology is a proprietary system on chip (SoC) designed by Allied Vision for exactly that purpose. The first camera models using that technology, the Alvium 1500 and Alvium 1800 Series, were showcased at our booth in several live demos.
Artificial intelligence and deep learning, the new frontier of digital image processing
Two of the live demos showcasing the new Alvium cameras were applications relying on deep learning algorithms, a technology that facilitates the programming of vision systems.
The first demo, developed in partnership with Au-Zone (Canada) and Toradex (Switzerland), showed a miniature system using an Alvium MIPI CSI-2 camera and a Toradex Apalis System on Module with NXP i.MX8 QuadMax SoC. The Au-Zone software DeepViewRT runs an inference engine to reliably perform the characterization of different types of pasta by comparing them with a few reference images taken with a mobile phone and fed into the system.
The second deep learning demo was designed in co-operation with Antmicro, a Polish embedded hardware and software company. It featured the all-new NVIDIA Jetson AGX Xavier processor connected to an Alvium camera via a MIPI CSI-2 interface. The demo was running an embedded deep learning neural network application developed by Antmicro detecting people and objects in real time in the live stream of the camera. More information on this project can be found on the Antmicro blog.
With such demos, the VISION show – and particularly the Allied Vision booth – made it very clear that the future of machine vision lies in embedded systems combining embedded-optimized camera modules with powerful systems on module (SoM) and neural network applications. Together, these new technologies open nearly endless possibilities to implement vision in areas where it was not possible so far for cost, space, or power consumption reasons.