Ship Detection and Motion Parameter Estimation with TanDEM-X in Large Along-Track Baseline Configuration

Transcription

1 Ship Detection and Motion Parameter Estimation with TanDEM-X in Large Along-Track Baseline Configuration SEASAR 2012 Workshop, Stefan V. Baumgartner, Gerhard Krieger Microwaves and Radar Institute, German Aerospace Center (DLR)

2 TerraSAR-X-Add-on for Digital Elevation Measurements Launched on 21 st June 2010 Mission Goals Global DEM Level-3 standard Local DEMs Level-4 like quality Demonstr. of innovative techniques applications (formation flying, bistatic acquisiton, Pol-InSAR, GMTI,...) Slide 2

3 TanDEM-X Data Acquisition Modes Pursuit Monostatic Bistatic Alternating Bistatic both satellites transmit and receive independently temporal decorrelation & atmospheric disturbances backup solution one satellite transmits and both satellites receive simultaneously dual use of signal energy requires synchronisation transmitter alternates between PRF pulses provides two baselines enables synchronisation, calibration & verification Slide 3

4 Synthetic Aperture Radar and Moving Targets Single-channel TerraSAR-X images. Left: vessel in the Strait of Gibraltar; right: train near Wolgograd Moving targets in conventionally processed SAR images appear... Displaced from their true positions Large azimuth displacement (slow ship up to 600 m; fast car up to 2 km and more) Small range displacement ( 5 m for line-of-sight velocites 100 km/h) Reason: Doppler shift due to across-track velocity Blurred in azimuth direction Reason: Along-track velocity and across-track acceleration Slide 4

5 Ground Moving Target Indication (GMTI) x r 1 true beam center platform t = t bc,1 r 1 TRUE POS. Objectives of GMTI Detection (at wrong position!) Parameter estimation - true position - absolute velocity - moving direction v p,1 v t,1 x 1 displaced position in SAR image 1 r 1 WRONG POS. Spaceborne GMTI so far.. Single platform Along-track interferometry ( small baseline) A priori knowledge-based GMTI ( Road database) Slide 5

6 Classical Along-Track Interferometry (ATI) Principle t baseline v p v p x (azimuth) r 1 t 1 v r0 The RX antennas are separated in flight or along-track (azimuth) direction, respectively. Slide 6

7 Classical Along-Track Interferometry (ATI) Principle t baseline v p r 1 v p x (azimuth) Range Difference r r r v t 1 2 r0 Interferometric Phase 4 4 r v t r0 r t 1 r 2 Azimuth displacement x v r0 r0 v p v r0 t 2 t temporal baseline v r0 line-of-sight velocity The RX antennas are separated in flight or along-track (azimuth) direction, respectively. ATI phase v r0 x Problems: Noise + Clutter Re-displacement errors (10s to 100s of meters) Slide 7

8 Along-Track Baseline small t large t v p Image 2 Image 1 Image 2 Image 1 moving target Interferogram Interferogram Small baseline t ms Moving target keeps in res. cell Only sensitve to v r0 (1D) Classical ATI Large baseline t s Moving target leaves res. cell 2D velocity estimation Large-Along Track Baseline GMTI Slide 8

9 Along-Track Baseline small t large t v p Image 2 Image 1 Image 2 Image 1 moving target Interferogram Interferogram Small baseline t ms Moving target keeps in res. cell Only sensitve to v r0 (1D) Classical ATI Large baseline t s Moving target leaves res. cell 2D velocity estimation Large-Along Track Baseline GMTI Slide 9

10 Large Along-Track Baseline GMTI Principle Large along-track baseline 20 km time lag 2.5 s Monostatic pursuit mode Moving target displaced in both SAR images Displacement difference true geographical position, velocity, heading (and acceleration) No a priori knowledge required! Detection of vehicles moving on open land and open sea! S. V. Baumgartner et al, A Large Along-Track Baseline Approach for Ground Moving Target Indication Using TanDEM-X, in International Radar Symposium (IRS), Cologne, Germany, September Slide 10

11 Large Along-Track Baseline GMTI Algorithm (II) Ocean Land ( no clutter suppression) Preprocessing Image 1 Image 2 Image 1 Image 2 Reference Patch Clutter Suppression Detection + Clustering 2D Correlation DPCA + - 2D Correlator Parameter Estimation Reference Patch Without Reference Patch Result Georeferencing Visualization 2D Correlator Result Slide 11

12 Large Along-Track Baseline GMTI Algorithm Ocean ( no clutter suppression) Land ( clutter suppression necessary) RGB RGB Superposition of single-channel images Superposition of single-channel images red: TerraSAR-X image (fore) green: TanDEM-X image (aft) DPCA Clutter-suppressed DPCA image Slide 12

13 Experiments and First Results Slide 13

14 Overview TanDEM-X Commissioning Phase TDX Launch Cycle LEOP GS Checkout LEOP Ground stations TDX ASM Offline Products Timeline Commanding Instrument & Hot/Cold Screener TMSP Monostatic CP Start: GS-Maintenance (3 days) Baseline Product Geo Cal TDX mono -static CP 20 km Formation Pursuit Monostatic Mode Pursuit-Monostatic / Bi-static Timelines Bi-static Commanding Antenna Pointing Antenna Model Radiometric Calibration SAR System Performance TMSP / SAR Product Verification. ITP Checkout / Pursuit Monostatic Sync Link Performance Exclusion Zone Tests Baseline Offset Determination FFR / Release of the close formation TSX ASM MPS Operationalisation Bistatic CP Start: GS- Maintenance Formation adjustment Station Checkout Bi-Static Timeline GS Updates Bi-static CP m Close Formation Bi-static Timelines Adjust. Commanding Fine Adjust. Radiometric Verif. Bi-static Performance DEM Calibration Tests FQR Nominal Close Formation time Monostatic TDX ITP Fine commissioning Adjustment phase Error Model Verification Large along-track Sync Horn baseline Selection 20 km Baseline 2.5 s Offset time Determination lag Global DEM Acquisition of 20 GMTI data takes Slide 14

15 Test Sites for Traffic Monitoring Interstate 15 NE of Las Vegas Monitoring of road vehicles Low clutter contribution Strait of Gibraltar Monitoring of ships Always traffic AIS data available Slide 15

16 Verification Using AIS Data as Reference Principle Automatic Identification System Standardized VHF transceiver Ships with GT 300 tonns ID, position, speed, moving dir., Available via internet ( real-time) AIS: Position GPS antenna / AIS time t AIS time t SAR Different acquisition times t AIS t SAR (sec. min) Extrapolation of the AIS data Const. velocity & head. assumed Position Difference SAR: Area centroid geocoded position acquired at t SAR Slide 16

17 First Results: Vessel Monitoring in the Strait of Gibraltar Coordinates: N, E Velocity: 16.7 kn Coordinates: N, E Velocity: 9.5 kn Rotation has to be considered Coordinates: N, E Velocity: 12.7 kn Coordinates: N, E Velocity: 8.9 kn Coordinates: N, E Velocity: 8.8 kn Coordinates: N, E Velocity: 9.8 kn What we estimate

18 Ludwig Axt MarineTraffic.com Slide 18 Ship Detection and Motion Parameter Estimation with TanDEM-X in Large Along-Track Baseline Configuration >

19 Verification Using AIS Data as Reference First Results (I) Northing Position Difference N Position Difference [m] -25 m W S E bad correlation ( RCS change) large acquisition time diff. (7 min 20 s) Vessels have moved mainly in range direction Northing pos. difference ~ azimuth re-positioning error True azimuth position is more difficult to estimate than range position! Slide 19

20 Verification Using AIS Data as Reference First Results (II) Velocity Difference Velocity Difference [kn] bad correlation ( RCS change) 0.15 m/s 0.54 km/h Heading Difference [ ] bad corr. Moving Direction Difference Slide 20

21 Verification Using Road Database as Reference Principle Position Error Absolute Geographical Position Error Estimated target position closest road point Target Moving Direction Compared with known road direction Detection / False Alarm Rate Unfortunately no ground truth available! Slide 21

22 Verification Using Road Database as Reference Results Absolute Position Difference MEAN : m STDDEV: m Occurence Moving Direction Difference MEAN : 0.55 STDDEV: 3.82 Position Difference [m] Occurence Simple detector fixed threshold False detections skipped manually 31 remaining true targets (SCNR 10 to 24 db) Heading Difference [ ] Slide 22

23 Conclusions and Outlook Large Along-Track Baseline Algorithm Proof of concept provided Targets: point-like (small cars) extended (ships) other movers / non-movers? Excellent parameter estimation performance You get both with one pass: High resolution SAR images + GMTI data Reference Data Data Fusion Traffic Monitoring assignment to certain roads (automatically) Ship Monitoring comparison / fusion with AIS data (automatically) Outlook Not all acquired data takes evaluated so far; no robust (CFAR) detector implemented Principally applicable for future platforms... Slide 23

24 State-of-the-Art and Future? Airborne Spaceborne TerraSAR-X SAR-Lupe F-SAR Global Hawk Differences / selection criterions: Spatial / temporal coverage Endurance / range Applications TanDEM-X Lockheed Martin DARPA High Altitude Airships (HAAs)? Turtle Airship