OMNI AI S7200 Real-Time Anomaly Detection on Video and SCADA. OMNI - - PowerPoint PPT Presentation

OMNI AI

S7200 – Real-Time Anomaly Detection on Video and SCADA.

OMNI AI – Who We Are

• Start-up, funded technology company
• Acquired BRS Labs Technology & Patents
• Historic Customers – Video Analytics
• Transit
• Oil & Gas
• Municipalities
• Headquartered in Dallas, Texas.
• R&D staff authored 150+ artificial intelligence patents
• Focused solely on artificial intelligence machine learning
• Evolving to custom AI solutions

Video Analytic Problem

“After only 20 minutes of watching and evaluating monitor screens, the attention of most individuals degenerates to well below acceptable levels.”

OMNI AI –Methodology

• Methodology
• Observe, Learn, Analyse, Rank, Report
• Alert Always/Never
• Visualization
• Software Platform that
• Easy to setup
• Learn by itself
• Adaptive to change
• Control Alert Volume
• Refine Alert

Known-Known

Handcrafted Knowledge

Known-Known

Handcrafted Knowledge

Known-Unknown

Statistical Learning

Known-Known

Handcrafted Knowledge

Known-Unknown

Statistical Learning

Unknown-Unknown

Cognitive Learning

OMNI AI – The Platform

Online Unsupervised Learning

• Neuro Linguistic Engine
• GPU Accelerated
• Meta Data from sensors
• Video, Image, SCADA, Audio
• Multi Source Sensor Fusion
• Identifies Feature-Symbols (Alphas).
• Builds Feature-Words (Betas)
• Collection of Alphas
• Constructs Feature-Syntax (Gammas)
• Collection of Betas
• Build custom abstract language to describe what it

has learned.

a a a a a a b b b b g g Numeric Symbol Abstract Symbol Graph

Foundation Video Meta Data

Video Case Study – Public Safety

• Goal
• Real Time Video Analysis System
• Deploy 500+ cameras
• Improve Public Safety
• Challenge
• 500 Different Field of View
• Pan-Tilt-Zoom Camera(s)
• Alert Volume
• Security personal become tire and/or bore
• Un-expected Anomaly

KNOWN-UNKNOWN Behavioral Analytics KNOWN-KNOWN Rule Based

UNKNOWN- UNKNOWN Cognitive Analytics UNKNOWN-KNOWN

Undefined

Video Tracking

• GPU Accelerated
• Real-time Object Tracking
• Online update
• Complex scene
• Indoor
• Outdoor
• Pan-Tilt-Zoom
• Video Stabilization
• Minimum Calibration
• Generate Meta Data
• 200 1-CIF Video Sensor on 2

Quadro M6000

Linguistic Representation (Video)

• Linguistic Representation is

Unique for each camera

• In this video, hundreds of

words is Learn to describe the behavior.

• Four most Relevant Feature

Words is shown

• Two major classes
• Vehicle
• 2-4 Feature-symbol
• Pedestrian
• 3 Feature-Symbol
• 3 sub-classes in Vehicle with

common Attribute such as Shape and Texture

Omni Video Real World Alert Clips

• System can describe millions of behaviors

• Goal
• Build a Real-Time, Intelligent, Adaptive system

for

• Predict unsafe condition – Over-pressure
• Improve efficiency – Under-pressure
• Detect performance degradation in regulator

stations

• Efficiently use their equipment and assets
• Challenge
• 2500 Distribution regulator stations
• 3.5 million alarms in 2.5 years
• Volatile time-series with changing “arithmetic”

means

• Difference sensor has difference behavior
• Valve failure, Debris contamination, Sulfur build-up

and Liquids contamination

SCADA Case Study – Natural Gas Regulator

Foundation SCADA Meta Data

Gas Pipeline – Real Time Anomaly Detection

Normal Operation

Anomaly Detection Autonomously

Gas Pipeline – Real Time Anomaly Detection

Online update

Normal Operation

Anomaly Detection Autonomously

Gas Pipeline – Real Time Anomaly Detection

Gas Pipeline – Real Time Anomaly Detection

100X+ reduction in Alarm

Omni AI – Predictive Maintenance

Omni AI Performance – Gas Pipeline

Total Alert (2.5 Years) HH and LL Alert (2.5 Years) Alarm Per Hour Baseline 3,507,107 16,497 0.75 Omni AI 6105 529 0.02

40x reduction in Total Alerts

Future Directions

• Custom solutions
• Certified apps
• Partner opportunities
• http://www.omniaicorp.com/OmniAIFoundationWhitePaper.pdf