Software is playing a major role in the growth of physical security
"Various predictions forecasts that the vol-
ume of data captured by IoT connected
devices will grow nearly six-fold and top over
2,000 exabytes (2.0 zettabytes ) in 2021."
Security News Every Day
www. securityworldmarket.com
Security News Every Day
www. securityworldmarket.com
TOMORROW
In the last two years we have witnessed some significant changes in the
structure and product performance of the physical security software
business and much of this has come through merger and acquisition. In
particular PIAM, Access Control Software and VMS have been in the
top five places when benchmarked on number of acquisitions and this
year five deals involved the acquisition of Artificial Intelligence Software
companies.
While deep learning is rapidly becoming the new star of video analyt-
ics actual commercial implementations are rare and there are only a few
commercial products with deep learning that are available today in the
field of video surveillance.
Machine learning encompasses a range of algorithms to enable a trend
or pattern recognition over time. It can be supervised, i.e. with expert
training, or unsupervised, with no inputs from humans. Deep learning is
a subset of machine learning and has applications in today's world, from
speech recognition to image recognition to even biomedical informat-
ics. In general, one can think of it as a cascade, many layers of nonlinear
processing units for feature extraction and transformation. Each succes-
sive layer, uses the output from the previous layer as input. Deep learning
when perfected should be much more accurate tool than traditional
video analytics.
Neural networks are loosely designed based on the biology of our brain.
It tries to simulate how humans think. All the interconnects between the
neurons, where the neurons can connect to any other neurons within a
certain physical distance. So artificial neural networks have several layers
of connections and directions of data propagations. Their applications
include pattern analysis, classification etc. which are all based on the
learning of multiple layers of features or representations of data. The
higher level features are divided from the lower layer features to form
hierarchical representations.
Nvidia and Neurala are relatively new starts chip manufacturers develop-
ing deep neural network frameworks designed for use in low-power
GPU's. Their aim was to give devices like drones, security cameras, and
consumer electronics some decision-making capabilities previously una-
vailable in these categories due to size, cost, or power restraints.
Motorola is among Neurala's investors and has invested an undisclosed
sum in Neurala's $14M Series A as part of the partnership between the
two companies. Motorola's Si500 Advanced Bodycam was designed to
take advantage of AI/DNN technologies, enabling Neurala's software to
run within their bodycam.
While Neurala also offers edge-based analytics capabilities for object clas-
sification, a number of manufacturers such as Avigilon, Axis Commu-
nications, Hanwha, Hikvision, and others, have already begun releasing
products with edge-based analytics, making Neurala's offering in this
regard of lower interest for the surveillance market.
The analytics market today is processed in the cloud but it will move to
the edge of the network as advances in compute capabilities and lower
cost of communications hardware make it cost effective to apply analyt-
ics closer to where data is generated or collected. Various predictions
forecasts that the volume of data captured by IoT connected devices
will grow nearly six-fold and top over 2,000 exabytes (2.0 zettabytes ) in
2021.
Having previously had its fingers burned, the security industry in general
is a little skeptical that new deep learning products are as practical and
robust as manufacturers say they are. However analytics in general is be-
coming an important driver and deep learning will add more momentum
to its importance.
James McHale,
Managing Director, Memoori Research
Note:
This article is based on The Physical Security Business 2017 2022
which is available to buy for only $1,500 USD for a single user license through
www.armedia.se
voice of the
security industry
d e t e k t o r i n t e r n a t i o n a l · 9
Analytics in general is becoming an important driver and deep learning will add more momentum to its importance, states James McHale in this artical.
Det INT 2_18_v2.indd 9
2018-04-09 13:53