Page 32 - EETimes Europe June 2021

32 EE|Times EUROPE

OPINION | EMBEDDED VISION
The most obvious of these are frameworks
Embedded Vision such as TensorFlow or PyTorch and libraries
like OpenCV.
But widely used task-specific neural net-
at the Tipping Point works, such as Yolov4 or Google Inception,
have changed the game. No longer do most
developers design a neural network; rather,
they pick a free off-the-shelf neural network
By Phil Lapsley, BDTI and train it for their task. (Of course, to train
a neural network, you need data. Depending
on your application, this may represent a
A TECHNOLOGY REACHES a tipping point when it hits three challenging data-collection project, although
milestones: First, it becomes technically feasible to accomplish there is an increasing number of open-source
important tasks with it. Second, it becomes cheap enough to use datasets available, as well as techniques to
for those tasks. And third, critically, it becomes sufficiently easy for augment your data or reduce the amount of
non-experts to build products with it. Passing those milestones is a data you need.)
great indicator that a technology is poised to spread like wildfire. At These building-block libraries and tools
this year’s Embedded Vision Summit, held as a virtual event in May, may be chip-vendor–specific. An example is
we saw clear evidence that embedded vision has reached this point. Nvidia’s DeepStream SDK, which simplifies
Embedded vision passed the first two milestones a while back. A the creation of video analytics pipelines.
huge part of putting the technical feasibility milestone in the rearview mirror was the advent of Although DeepStream is tied to Nvidia’s
deep neural networks, which revolutionized the tasks that vision could do. Suddenly, classifying Jetson processors, it’s a great example of
images or detecting objects in messy real-world scenes was possible, in some cases with accuracy a vendor’s providing something closer to
surpassing that of humans. To be sure, it wasn’t easy, but it was doable. a complete solution (as opposed to “just
Moore’s Law, market economics, and domain-specific architectural innovation took care of silicon”). BDTI and Tryolabs recently built a
the second milestone. Today, for US$4.99, you can buy a tiny ESP32-CAM board that has a dual- face-mask–detection smart camera product
core, 240-MHz processor and a 2-MP camera module with an on-board image signal processor using DeepStream and YoloV4. 1
and JPEG encoder; it’s a squeeze to do computer vision on it, but it’s certainly possible, and it’s Second is the availability of tools spe-
tough to beat the price. If you have more money to spend, your options widen significantly. For cifically designed to simplify the process
example, US$99 will get you an Nvidia Jetson Nano Developer Kit with a quad-core 1.4-GHz of creating embedded-vision and edge-AI
CPU, a 128-core Maxwell GPU, and 4 Gbytes of memory — more than enough to do some serious systems. A great example is Edge Impulse,
embedded-vision processing. whose tools ease development of embedded
Best of all, new processors show up monthly and at all price, power, and performance points, machine-learning and -vision systems. For
often with specialized architectures that boost performance on computer-vision and neural- instance, the Edge Impulse platform can be
network inference tasks. Examples include new offerings from Xilinx, Cadence, and Synaptics. used to train and program an image-
It’s that pesky third milestone, ease of use, that’s been the rub. Sure, deep learning radically recognition neural network for that US$4.99
changed what vision systems were capable of, but you needed to be a ninja to be able to design ESP32-CAM mentioned above.
neural networks, gather the data needed, and train them, to say nothing of then having to imple- Similarly, for beefier processors, Intel’s
ment them on a resource-constrained embedded system. But that’s really changed in the last few DevCloud for the Edge and OpenVINO tools
years. Two big shifts have driven that change. aim to make vision far easier to implement at
First is that you don’t have to build embedded-vision systems from scratch anymore, thanks the edge.
to the widespread availability of high-quality, well-supported tools and libraries for vision. Think back to the 1990s, when wireless
communications was the “new new thing.”
To start with, it was expensive magic that
required a team of RF wizards to make
happen. But it reached the tipping point, and
today, anyone can buy RF modules for a few
dollars to enable wireless communications in
an embedded product. In the process, literally
billions of wireless units have been shipped
with correspondingly huge economic impact.
Embedded vision is at a similar tipping
point, and the Embedded Vision Summit is a
great place to watch it happen in real time. ■

REFERENCE
1 A Mask Detection Smart Camera Using the Nvidia
IMAGE: SHUTTERSTOCK Phil Lapsley is a co-founder of consulting
Jetson Nano: System Architecture and Developer
Experience. Embedded Vision Summit, May 25,
2021. bit.ly/2QXncDv

firm BDTI and one of the organizers of the

JUNE 2021 | www.eetimes.eu Embedded Vision Summit (bit.ly/3ege1a3).