Real-time AI Video Surveillance

  • Home
  • Video_Surveillance
Image
Real-time AI Surveillance
Systems
Distributed Heterogeneous AI Infrastructure
mageI
Why a distributed system
It aims to provide storage and computational resources near to
user at the network edge, to minimize latency and response time.
Computing data right up to the network’s edge can give you the
performance, speed and low-latency connectivity you need to
support mission critical task such as surveillance.
Designing distributed system, resilience is achieved by the system
being capable of automatically adapting when adverse situations
occur in order to continue to serve its purpose.
Real‐time AI Systems 2
Image
Real‐time AI Systems 3
Image
Why is Heterogeneous computing infostructure the future
of smart cites
Real‐time AI Systems 4
Heterogeneous computing
refers to systems that use
more than one kind of
processor or cores. (Not
vendor dependent)
These systems gains
performance or energy
efficiency not just by
adding the same type of
processors, but by adding
dissimilar coprocessors.
This state-of-the-art
infrastructure reduces the
government and nations
dependence on a
particular technology or
vendor.
Image
Emerging Computing Platforms for AI
• Heterogeneous Multi‐Core Systems
• Mix of processor cores and specialized hardware accelerators yields more energy‐
efficient computing
• The approach to AI is getting heterogeneous
• DSP, GPU, CPU, FPGA, custom hardware…
• whatever is the best piece of hardware to handle an AI task in the most power efficient way
Samsung Snapdragon 820, 2016
Samsung Snapdragon 855, 2019
Source: Qualcomm
Source: Qualcomm
Real‐time AI Systems 5
Image
EC2
F1
November 2016
Heterogeneous Multi-core System-on-Chip
March 2017
May 2016 (1
st
gen.)
May 2017 (2
nd
gen.)
Ivy Town Xeon + FPGA:
The HARP Program
June 2016
December 2013
Zynq UltraScale+ MPSoC
August 2017
Versal
October 2018
Sources: Intel, Google, Xilinx, Amazon
Real‐time AI Systems 6
Image
What challenges does Heterogeneous system
overcome?
Real‐time AI Systems
7
Vendor Lock: For example,
the USA banning China from
using Nvidia Technology
greatly crippled there AI
infrastructure because of the
dependence on GPU.
Supply Chain: By using this
design base we can overcome
global supply chain issues and
have a distributed, efficient,
resilient, and state of the art
system.
Integrate: We can expand and
integrate any existing systems
and projects into our system,
by being agnostic on Camera
and AI Hardware.
Image
SWOT ANALYSIS
Save cost, bandwidth, and energy in video streaming
and transcoding.
State of the Art, Scalable, Agnostics, Adaptable, and
Resilient.
Edge Computation requires power on the
Edge. Thus, with no power the system will be
offline.
Electro Magnetic Interference Attacks
Physical Attacks on edge hardware
Integrate into existing projects, distribute edge computation
power over the network, centralize all streams from different
ministries into one command center
S
W
O
T
STRENGTHS
WEAKNESSES
THREATS
OPPORTUNITIES
Image
Accelerator
A special‐purpose hardware that is
optimized to perform a dedicated
function(s) as part of a general‐purpose
computational system
While being part of a larger system, it spans a scale from
being closely integrated with a general‐purpose processor
to being attached to a computing system
Increasingly accelerators are migrating into the chip, thus
leading to the rise of heterogeneous multi‐core SoC
architectures”
Implemented as
Application Specific Integrated Circuit (ASIC)
Field‐Programmable Gate Array (FPGA)
Real‐time AI Systems 10
Image
Strong
Partnerships
Real‐time AI Systems 11
Image
Real‐time AI Systems 12
SOLUTION CAPABILITIES
The BriefCam Insights product allowsfor VMS video ingestion, while BriefCam Protect allows VMS
video ingestion as well as le-based video ingestion.
REVIEW
Review hours of video in minutes and rapidly
pinpoint objects of interest. Quickly search
objects and events of interest by metadata
such as men, women, children, vehicles, and
lighting changes with speed and precision,
using 27 classes and attributes, as well as
face recognition, appearance similarity, color,
size, speed, path, direction, and dwell time,
providing an ever increasing and powerful set
of distinct search combinations.
RESPOND
Unlock the ability to trigger a call to action,
bringing relevant events to the forefront with
real- time alerting capabilities. Enable
organizations to proactively respond to
situational changes in their environment,
while e ectively balancing sensitivity,
accuracy, and e ciency.
RESEARCH
Gain operational intelligence from the
extracted and aggregated video metadata
such as men, women, children, vehicles, size,
color, speed, path, direction, and dwell time,
enabling users to quantitatively analyze their
video, derive actionable insights for data-
driven safety.
Image
NVidia Jetson Partnership
Real‐time AI Systems
13
Image
Nvidia Inception Inference Engine
• Utilize USB, CSI-MIPI and IP cameras with our Nvidia Inference
Engine on PCs, workstations, servers and NVIDIA Jetsons and see
inference results in your web browser.
• Experiment with TensorRT-optimized Tensorflow Object Detection API,
Yolo V3, Face Detection and Covid-19 Mask Detection models in 3-
steps.
• Process any local or remote stream from any camera on your PCs,
workstations, servers and NIVIDA Jetsons.
Real‐time AI Systems 14