1 iWatch: BIG Data M anagement and Analytics for Intelligent - PowerPoint PPT Presentation

1

iWatch: BIG Data M anagement and Analytics for Intelligent Surveillance Farnoush Banaei-Kashani, Ph.D. Research Associate, Computer Science Department Associate Director, IM SC Viterbi School of Engineering University of Southern California Los Angeles, CA 90089-0781 banaeika@usc.edu 2

Outline • An Overview of the iWatch Project • Vertical Cuts: Application-Specific Prototypes – iWatch for Safety and Security (i4S) – iWatch for Health (i4H) – iWatch for Energy (i4E) 3

Project Objectives • A M ulti-purpose S ystem for Intelligent Geoimmersive Surveillance • An End-to-End S ystem! • A Research Showcase • A Technology Showcase 4

GeoImmersive Surveillance • Detect Events • Sense • Act    Forensic analysis M ulti-modal sensing Visualization    Real-time monitoring Deep sensing Recommendation    Prediction of potential Active sensing Actuation expected (and unexpected!) events 5

iWatch System Architecture Data Acquisition Sense Act Active Sensing Incident Detection Event Detection Event Detection Real-time Event Incident Efficient Incident Stream Detection and Retrieval Prediction from Each Incident is an object with Incident Stream spatial, temporal, and textual attributes Incident 6 Archive

Research Showcase 4. Scale-up Data Acquisition Sense Act 1. Active Sensing Active Sensing Incident Detection Event Detection Event Detection Real-time Event Incident Efficient Incident Stream 3. Dynamic Integration 2. Inferred Archival Detection and Retrieval Prediction from Each Incident is an object with Incident Stream spatial, temporal, and textual attributes Incident 7 Archive

Technology Showcase • VideoIQ (through DPS): Smart PTZ Cameras • Qualcomm/ HTC: Evo 3D Smartphones Data Acquisition • Qualcomm/ HTC: Evo 3D Smartphones • Verizon (through AIL)? Sense Act • USC: KNOWM E Network (BAN) • Samsung? • OSIsoft (through Chevron): SCADA/ PI • ESRI? • Verizon (through AIL)? Active Sensing • Intel? Incident Detection Event Detection Event Detection • M icrosoft: StreamInsight CEP Engine Real-time Event Incident • IBM : IBM InfoSphere Streams Efficient Incident Stream Detection and • Oracle: DBM S 11g Retrieval Prediction from Each Incident is • Lockheed M artin/ Rocket Software: AeorText (?) an object with Incident Stream • NEC? spatial, temporal, and textual • HP? attributes Incident 8 Archive

Project Timeline J anuary 2011 Applications Sponsors Technologies Funding Team Safety and Security IM SC Oracle 11g ~20K IM SC Researchers (4) IM SC Students (2) 9

Project Timeline (cont’d) J anuary 2012 Applications Sponsors Technologies Funding Team Safety and Security IM SC Oracle 11g 1.2M + USC Public Safety Public Health NIJ IBM USC Doctors (3) Energy NGC M icrosoft CHLA Doctors (1) CREATE (DHS) Qualcomm Reservoir Engineers USC (DPS) Sponsors’ PM s CIA? IM SC Researchers (4) NGA? IM SC Postdocs (4) CTSI? IM SC Students (15) NIH? Industry Partners LA County? USC AM I Oracle? CiSoft (Chevron) 10

Outline • An Overview of the iWatch Project • Vertical Cuts: Application-Specific Prototypes – iWatch for Safety and Security (i4S) – iWatch for Health (i4H) – iWatch for Energy (i4E) 11

iWatch for Safety and Security (i4S) • Purpose: Forensic and Real-time Criminal Activity Detection from M ulti-Source M ulti-M odal Data • Sponsors: NIJ, NGC, CREATE, DPS • Team: – Law Enforcement and Security/ Intelligence Experts: M ark Greene (NIJ), Ed Tse (NGC), Carol Hayes (DPS) – Risk Analysis: Isaac M aya (CREATE) – Incident Detection from Video: Ram Nevatia (Tracking), Gerard M edioni (Face detection) – Geo-keyword Incident Indexing: Cyrus Shahabi – Spatiotemporal Event Detection: Farnoush Banaei-Kashani – M obile Video Search: Seon Ho Kim 12

Approach • M otivation: M ulti-modal integration enables more effective surveillance systems for criminal activity detection • Challenge: Data overload in detecting events in large environments over long time intervals • Proposed Approach: Utilize state-of-the-art content analysis techniques to extract incidents from input data streams, while integrating the incidents in the spatiotemporal domain (rather than content domain) to detect events 13

Approach (cont’d) • Advantage: Sensors Allows for event detection in large Content Analysis M odules spatial and temporal scales Incidents Spatiotemporal Cross-Referencing C Events 14

Last Y ear’s Demonstration • M ode: Forensic Analysis • Input: Video feed from 25 PTZ cameras 15

i4S Prototype: System Architecture 72 PTZ Cameras, Police Reports, Crowdsourced Tweets M obile Video 45 LPRs 1100 ACR (GB/ day) (TB/ day) (KB/ day) (KB/ day) Data Acquisition Sense Act Active Sensing for Face Capture Incident Detection from Video, Text, Sensor Event Detection Event Detection Real-time Event Incident Efficient Incident Stream Detection from Retrieval Incident Stream Incident 16 Archive

i4S Prototype: Research Showcase 72 PTZ Cameras, Police Reports, Crowdsourced Tweets M obile Video 45 LPRs 1100 ACR (GB/ day) (TB/ day) (KB/ day) (KB/ day) Data Acquisition Sense Act Active Face Tracking (NIJ : M edioni) Active Sensing for Face Capture Dynamic Risk Analysis (CREATE: M aya) Incident Detection from Video, Text, Sensor Event Detection Event Detection Efficient Incident Search Online Tracking (NIJ : Nevatia) (NIJ& NGC: Shahabi, Kim, Real-time Event Incident Banaei-Kashani) Efficient Incident Stream Detection from Retrieval Dynamic Event Detection (NIJ Incident Stream : Banaei-Kashani) Event Detection Supporting Data Uncertainty Incident 17 (NGC: M edioni, Shahabi, Banaei-Kashani) Archive

i4S Prototype: Technology Showcase 72 PTZ Cameras, Police Reports, Crowdsourced Tweets M obile Video 45 LPRs 1100 ACR (GB/ day) (TB/ day) (KB/ day) (KB/ day) Qualcomm/ HTC Smartphones Data Acquisition Sense Act VideoIQ PTZ Cameras Active Sensing for Face Capture Incident Detection from Video, Text, Sensor Event Detection Event Detection IBM Streams Real-time Event • VideoIQ PTZ Cameras Efficient Incident Incident Detection from Stream • IBM Streams (Text Retrieval Incident Stream Analytics Toolkit, Video Oracle 11g Analytics Toolkit (iM ARS, Incident OpenCV 2.0) 18 Archive • AeorText

Sample Demonstration M obile Client Server-side User Interface Create Update M onitor Search for “Geofence Demo” on Y outube 19

Outline • An Overview of the iWatch Project • Vertical Cuts: Application-Specific Prototypes – iWatch for Safety and Security (i4S) – iWatch for Health (i4H) – iWatch for Energy (i4E) 20

iWatch for Health (i4H) • Special-Purpose Prototypes – Prototype I: Contact Investigation – Prototype II: Understanding Geography of Diabetes – Prototype III: Point-of-Care M obility M onitoring 21

i4H-Prototype I: Contact Investigation • Purpose: Retrospective and Real-time Contact Investigation • Sponsor: NIH? • Team: – Contact Investigation: Dr. Brenda Jones (TB), Dr. Pia Pannaraj (Flu) – Tracking and Face Detection: Gerard M edioni – Spatiotemporal Contact Analysis: Farnoush Banaei- Kashani, Cyrus Shahabi 22

Step I: Data Collection 23

Step II: Reachability Analysis Input A graph G which is: • Large scale (Huge number of edges and vertices) • T emporal (Edges are added and deleted over time) • Geospatial (Nodes are moving in space) Queries Within a time interval [a,b], find: • Whether u is reachable from v ? • The individuals reachable from v ? • The individuals that can reach v ? 24

i4H-Prototype I: System Architecture Data Acquisition Sense Act Active Sensing for Face Capture Incident Detection from Video Event Detection Event Detection Incident Efficient Incident Stream Retrieval Incident 25 Archive

i4H-Prototype I: Research Showcase Data Acquisition Sense Act Active Face Tracking (M edioni) Active Sensing for Face Capture Efficient Reachability Analysis Incident Detection from Video Event Detection Event Detection (Banaei-Kashani, Shahabi) Online Tracking (M edioni) Incident Efficient Incident Stream Retrieval Incident 26 Archive

i4H-Prototype I: T echnology Showcase Data Acquisition Sense Act PTZ Cameras Active Sensing for Face Capture Incident Detection from Video Event Detection Event Detection Incident Efficient Incident Stream Retrieval Oracle 11g Incident 27 Archive

i4H-Prototype II: Understanding Geography of Diabetes • Purpose: Spatial Analysis and M ining of Diabetes Patient Data to Understand Spatial Causative pathways, Processes, and Patterns • Sponsor: Verizon?, Oracle? • Team: – Diabetes: Dr. Andy Lee – Use-case and M arket Analysis: Nathalie Gosset (AM I) – Body Area Sensor Network: M urali Annavaram – Spatial Data Analysis and M ining: Farnoush Banaei- Kashani, Cyrus Shahabi 28

Vision Care-Providers Analytics Patient Patient Analytics (PA) Expert Analytics (EA) 29