»
The diversity of neural networks is made
accessible to users through a wide range of open
source frameworks, high-level software and
services. A multitude of published neural network
architectures already cover different requirements for
complexity, accuracy or inference times. Automation
and monitoring of industrial plants provides more
and more suitable image data for the training of
these architectures.
“Image to information on-camera”
The interpreter application “IDS NXT ferry”, specially
developed for compatible network architectures,
converts existing, already trained neural networks
for use on the embedded vision platform. Users
can thus conveniently provide their own neural
networks to the IDS NXT inference camera for
various tasks, which in addition to generating
image data also directly takes care of the analysis
and evaluation of this information. Determining
the information content and thus a data reduction
then already takes place decentrally in the camera,
which avoids bandwidth bottlenecks in the data
transmission. With the distribution and networking
of such “cyberphysical components”, direct process
data will become available according to the internet
of things approach, which will sustainably boost
the automation and processing speed of industrial
manufacturing processes.
The FPGA-based acceleration of artificial intelligence
allows inference times of only a few milliseconds
with common neural network architectures. Cameras
based on the IDS NXT platform can thus keep pace
with modern desktop CPUs in terms of accuracy
and speed of results - with significantly less space
and energy consumption at the same time. The re-
programmability of the neural network accelerator
offers additional advantages in terms of future
security, low recurring costs and time-to-market.
AI technology is advancing so rapidly that new
frameworks and architectures are being added every
month. These can be integrated by the manufacturer
via software without changing the hardware platform
and users do not have to purchase new devices. The
fast reconfiguration of the dedicated processor also
12 TUBE NEWS July 2019
Decentralized image analyses generate process-
relevant information directly in the camera.
allows switching between several loaded neural
networks within a few milliseconds during runtime.
This enables the sequential execution of different
classifications with the same image data within a
vision app.
Forecast
Artificial neural networks have already proven their
additional value for the modern machine vision
world. Machine object recognition and classification
are two of the most important new capabilities
that industry automation brings forward, as well
as many applications in other markets. The flexible
customization of the IDS NXT platform simplifies its
integration into an existing system and adaptation
to different markets. Equipped with this AI-based
embedded system, the user can conveniently deploy
his own neural networks in the inference camera for
different tasks. For completely autonomous, PC-
independent operation in industrial environments,
model variants of the industrial cameras with
industrial protocols such as PROFINET or OPC-
UA will also be available. Starting in the second
quarter of 2019, IDS will show an easy controllable
way to quickly and easily bring AI-supported image
processing to the machine as a complete embedded
vision system with IDS NXT industrial cameras.
.
www.ids-imaging.com