Edge Intelligence: The Central Cloud is Dead – Long Live the Edge Cloud!
(2) Machine intelligence
can take place at the cloud or edge,
whichever approach is optimal for the
specific scenario.
False alarms can occur as the processing
power of many edge devices does allow full
analysis to discern between normal
movement, work activity, and an exceptional
situation such as a fall. Nor is the processing
sufficient to allow intelligent power
management of the peripheral devices.
Required edge services include the following
services:
(3) Data transport cost
The cost of current data transport
mechanisms is not justified by the level of
data being transported.
Physical Access Control – Tailgating
Detection/ Fire Detection via
Surveillance Cameras
Besides requiring a greater local compute
power, as described for the indoor location
tracking, this use case would need
additionally:
(1) Containerization
The ability to install the camera modules
without hands-on access and with a high
degree of modularity and security in existing
environments is essential. In nearly all cases
such applications will be add-ons to existing
camera
systems
and
fire/security
infrastructures.
N EEDED C APABILITIES
In order to ensure data privacy and prohibit
any data or system tampering, IoT edge
computing solutions are expected to be
securely integrated with the cloud. Edge
solutions also need to be centrally managed
to minimize costs and to optimize lifecycle
management across a wide range of edge
devices. Data management and processing
- 28 -
Persistence – to store IoT data on IoT
gateways. IoT administrators can
configure which data should be stored
locally and set a data aging policy.
Streaming – to analyze IoT data streams.
IoT administrators can define conditions
with adjustable time windows to identify
patterns in the incoming IoT data as a
basis for automated events. For example,
the vibration, sound and other
continuous data stream from a variety of
sensors deployed in the machines, which
can only be received in accordance with
the sliding window order, should be
analysed and compared to the existing
rules just-in-time so as to detect the
abnormal and initial subsequent
transactions and notification of
appropriate parties.
Business transaction – to execute
business transactions at the edge to
provide continuity for critical business
functions, even when the edge is
disconnected from the cloud.
Predictive analytics – to use predictive
models for analyzing the IoT data. The
predictive algorithm is constantly “being
trained” and improved in the cloud
based on all available data. The resulting
predictive model is then sent to the edge
and applied there.
Machine learning – to apply machine
learning algorithms at the edge
specifically for image and video analysis.
September 2017