Efficiently convert massive quantities of sensor data into actionable information for tactical commanders

Transcription

1 Efficiently convert massive quantities of sensor data into actionable information for tactical commanders Mr. Otto Kessler Program Manager Tactical Technology Office

2 Motivation What the Commanders get... Large numbers of partially overlapping sensors s of reports; s of images per minute Unregistered, soda straw sensor observations Very high false alarm rates Signals - based What the commanders want... II I I I I 7 X 7 I Timely situation knowledge Comprehensive coverage (> targets over ~ Km 2 ) Accurate target locations with small Circular Error Probabilities Low burden, geo-referenced database

3 The Problem Sensor data increasing exponentially FIA, Global Hawk, etc. Single source analysts decreasing at high rate No multi-sensor analysts Targeting decision cycle delayed by manual processing Missed opportunities caused by too much un-exploited data Current Process (hours) DTED Level WGS-84 MTI Sensor Analysts Tactical NTM CGS E-8 Dissemination Images Text Messages Intel/Ops Analysts AOC TOC AOC ASAS Commanders (Mixed Views) WGS-84 SIGINT IPF Text Messages CGS ASAS Example - Image centric surveillance - Kosovo Pixel by pixel eyeball change detection Single sensor at a time( stovepipe analysis ) Manual exploitation - hours to days for product No automatic multi-sensor geo-registration Consequence Failure to find / identify targets

4 DDB Solution Common geo-registered database Common grid tied to wide area terrain data (DTED, CIB, FFD) Multi-sensor observations (SAR, EO, IR, GMTI and SIGINT) Fusion across sensors Model based evidence accumulation Track targets and features at object level Wide area coverage, large numbers of targets Dynamic closed loop tasking overcomes missing/ambiguous data Self evaluation of database and task requests to AIM DDB Solution (minutes) DTED MTI SIGINT Common Targeting Grid Automated Georegistration Multi- Sensor Change Detection Commanders (Consistent tailored views) All-Source Track and ID Fusion Sensor requests (AIM) Dynamic Multi-sensor ISR Database

5 DDB Architecture Enables Sensor Data Access and Technology Growth Database architecture overcomes limitations of data ownership Enhances Data sharing, interoperability Technological growth - applications and visualization Leverages COTS thrusts in open systems & object oriented databases Explore military needs which exceed likely commercial interests Sensor Data Store Registration Dynamic Data Services Info Products Info Requests Spatial Temporal Indexing Rapid Search, Retrieval, Purging Space-time queries Probabilistic Queries Query & Data Distribution Database Mediation Selective data push Real-time processing Sensor History Data, Foundation Data DTED Hide XX X SIGINT Hide XX GMTI X NTM Hide Scalable spacetime indexed databases Future XX X Online data purging High I/O rates of multimedia data User Interface Derived Products Low latency push of new data Applications Change Detectors, All source Track Fusion Model Based Classifier, Networks,... Situation History

6 Registration of all Sensors to a Common Targeting Grid DDB APPLICATIONS CHANGE DETECTORS FUSION ALGORITHMS GMTI SIGINT INGEST HPDS Dynamic Reference Fiducial Dynamically Updated DEM Georegistered Imagery, MTI,... Road Networks, Feature Data Reference Metrics DDB Registrar Absolute geospatial accuracy for targeting (~ m.) Relative georegistration for change detection (<m.) DDB Registrar Registration Georegistration Dynamic update of reference data DTED/CIB Reference Data Intensification Conversion Services

7 Normalcy Models Enable Wide Area Change Detection Detectors Detectors Detectors, GMTI, SIGINT Sensing parameters, Terrain, Landclass, Object relationships Object queries Change Analysis Output Filter Object updates Normalcy models provide context for detection thresholds Is this a region of high clutter? Was it there on the last sensing pass? Has it changed state or shape? Is it emitting as expected? Is it moving in a new way, place, or time? Object Normalcy Models Terrain / statistical General vehicle Target model Feedback from Track and ID Fusion Combined multi-sensor data is required to derive normalcy Spatial, temporal, feature based representation of scene content, background, and behavior.

8 () The Ar ty Bty receives orders from the st MR Bn for indirect fire on tar gets on the far bank (235) 2 (4) Bridge company reports MTU-2 retrieval c omplete to the st MR Bn (7327) 3 (2) The MR Regt CP orders the MR Bn s tank company to prepare to conduc t direct fire on targets on the far ban k (24456) 4 (5) MR Re gt CP orders units to confront U N troo ps, but not to p rovo ke hostile action (74634) 5 6 (3) The MR Regt CP acknowled ges rec eipt of report o n readiness to cross an d order s the MR Bn to begin crossing ( 3236 ) 7 (6) MR Re gt CP orders the tan k company s T-62s to take up po sition o n designated highw ay s (8324) 8 SIGINT Change Detection Approach COMINT Intercepts (DF/TDOA/FDOA location measurements, extracted audio ) Target reports Message Handler ELINT Intercepts (DF/TDOA location measurements) COMINT Multi-Platform Association COMINT Radio/Platform Pairing ELINT Multi-Platform Association EOB / Weapon System Pairing COMINT Tracking & ID Features COMINT / ELINT Correlation ELINT Tracking & ID Features Emission Density Alerts for tactically significant events SIGINT Tracks, ID Emission Profile Analysis of spatio-temporal structure of RF activity Emission Density - Exploits spatial features of the battlefield Emission Profile - Exploits periodic activity of the battle space Communication Network - Maps lines of communication between entities and echelon levels Event Recognition Comm Network

9 GMTI Change Detection Approach Multiple Collection Raw GMTI New Roads Normal Flow MPCD Collated Standardized GMTI Tracks MTI Ingestor MP Tracking FlowCD Change Event AMTIR TWS Registered Roads Ingest Common representation Sensor History Database Spatial-temporal indexing Register Registered GMTI Improved position accuracy Improves multi-source fusion Multi-Platform Tracking Improved track continuity, purity, position and velocity accuracy Flow Densities Traffic Flow Wide area movement estimates Indications and warnings of significant movements AMTIR - Absolute MTI Registrar MP - Multiple Platform MPCD - MTI Pattern Change Det. TWS - Trip Wire Sentry FlowCD - Flow Change Det.

10 Change Detection Approach Increasing PID and Decreasing FAR Change Detection Increasing Semantic Specificity Change Fusion HSI EO IR SAR Multi-sensor Imagery Image Object ILCD SAR OLCD EO OLCD Associated Changes OLCF Significant Fused Object Changes Detections Object Locations, Features Sensor history enables modeling of image response under various conditions Image-level Object-level Object Level Change Fusion enables decorrelation of false alarms Change fusion provides high confidence discovery of objects for simultaneous track & ID Multi-look -- Multi-temporal -- Multi-sensor -- Multi-algorithm Fusion

11 All-Source Track & ID Fusion Approach Fuses Change Detector Outputs Detections, tracks, identifications, features MHT (Multiple Hypothesis Tracker) Support sensor tasking Target Identification Fuse and maintain SIGINT/ identity information Model based classifier - EO and SAR classification of vehicles SIGINT ID features link with moving and stationary cycles Target Track Continuity Through Move-stop-move Cycles Fuse GMTI and SIGINT when moving Fuse and SIGINT when stopped Increase track length relative to MTI alone Recognize force level activity and relationships SIGINT MTI Force Relationships II I 7 I II X I All-Source Track & ID

12 DDB Algorithm Relationship MTI MTI Registrar MTI CD Flow Density MTI Tracker SAR & EO ILCD SIGINT Registrar SAR OLCD EO OLCD OLCF Model-based Classifier Top SIGINT level DDB Architecture Illustrating Connections between ATIF, SIGINT SIGINT CD Registrar Tracker ATIF OBJECTS Location Kinematics ID DTED/CIB/FFD Roads, etc. Registration Shared multi-spectral fiducial Salient feature matching Iterative refinement Data flow Fiducial Emission Density Change Detection Adaptive background modeling (normalcy) Spatial and temporal pattern analysis Tripwires Force Analysis Applications Object Oriented Database Management System Sensor Tasking System Integration All-Source Track & ID Correlates detection features from sensors Accumulate over time Maintain through move/stop/move cycles I LCD Image Level Change Detection ❽ OLCD Object Level Change Detection ❽ OLCF Object Level Change Fusion ❽ ATIF All-Source Track & ID Fusion

13 Data Driven Collection Management Goals ISR Strategy Developer AIM Multi- Asset Synchronizer TASKING Task Valuation Task Generation Track & ID Fusion DDB Multi- Sensor Change Detection Automated Registration DATA Goal Close the Loop Between Collection Management and Situation Estimation Now Manual coupling - Forces tasking focus to be on static targets - Limits use of uncertainty info in task valuation Future Automate process by developing - Task generation from ambiguities in fused product - Task valuation based on uncertainty estimate of the current situation

14 DDB Accomplishments Demonstrated Benefit of Fusion Gain Wide Area Change Detection (WACD) - Statistical normalcy models to recognize change and reject false alarms - EO/SAR fusion reduced FAR ~4x in experiments MTI - Increased track continuity by 27% - Reduced error in position by 37% SIGINT - New emitter mapping and profiling capability - 5% reduction in emitter location error through multi-platform fusion All-source track and ID Fusion (ATIF) Developed new capability to maintain track through move - stop - move cycles Association of mover/sitter tracks extends track continuity ~x4 Dynamic Data Services Common targeting grid Registered sensor histories over space and time. Foundation technologies to produce a timely, accurate estimate of the ground situation (kinematics, locations and IDs of objects).8.6 Pd.4.2 FA / KM 2 Stationary Tracks SAR+EO Fused Changes SAR+EO Fused Change SAR OLCD EO ILCD SAR OLCD SAR