Object Detection for Underwater Port Security

Transcription

1 Object Detection for Underwater Port Security Dr. Lloyd Huff LCHUFF CONSULTANCY,LLC Mr. John Thomas TRITON IMAGING,INC Shallow Survey 2012 February 22, 2012

2 INTRODUCTION I am glad to be here today to make my seventh presentation at a Shallow Survey Conference. I am especially pleased to be able to talk, today, about a topic which harkens back to the First Shallow Survey Conference in Sydney, where I was the conference co-organizer, with Roger Neil, and the Conference Keynote Speaker. My concluding keynote remarks in 1999, were that I did not know what advances would occur in high-resolution shallow water surveying over the next decade, but I was confident the advances would depend on three things: Location, Location and Location

3 PREFACE The advances made in multi-beam sonar technology and in postprocessing of GPS observables, since 1999, have made it possible for me to present this topic to you today The enablers have been: (1) Better spatial resolution bathymetry and imagery, and (2) Improvements in techniques for positioning the pixels, with (3) A good measure of computer wiz thrown in. However the mantra remains : Location, Location and Location

4 Maritime Improvised Explosive Devices (MIEDs) are one, among many, reasons to conduct Change Detection Surveys MIEDs are cheap, easy to acquire or build MIEDs can be deployed from small craft, commercial vessels, swimmers, etc. MIEDs are difficult to detect, identify and counter

5 Two Examples of the Challenge Today s modern ports and harbors: Complex layouts of channels, berths, harbors, and basins Varying water depths, bottom types, currents, turbidity Commercial disruption from an event could be disastrous Port of Los Angeles 43 miles of channels 270 berths 17 recreational marinas Extensive deep water approach 5,800 ship transits per year Wellington Harbour Area of 85 km² 14,000 transits per year Seaview tanker terminal Langdon container terminal Cross-bay and inter-island ferries Mix of recreational and commercial uses Port of Los Angeles Wellington Harbour

6 The Solution Apply MBE state-of-the-art hydrographic survey technologies to the port security problem Institute a program of repeat surveying using hi-resolution MBE bathymetry/imagery and precise geo-location, as the basis for detecting changes in the bottom. Maintain a database of port seabed bathymetry/imagery to support change detection and environmental awareness Employ a semi-automated change detection toolset that is precise, accurate and easy to use

7 Essence of Change Detection Collect baseline MBE bathymetry/ imagery of port seabed Periodically, or on-demand, re-survey areas of interest Process soundings into a hi-resolution bathymetric surface and compare with baseline bathymetric surface to check for changes Mosaic recent imagery and compare with baseline mosaic to check for changes Scan for large changes using visual comparison tools Locate potential objects using change detection tool Examine detections with analysis toolset Mark, report and store objects of interest in a contact database, as well as, store the new survey data in a historical survey database Update baseline databases with recent bathymetry/imagery

8 Port of LA Change Detection Program The LA Port Authority has a historical survey database with 10 years of bathymetry/imagery data available on-line and can instantly access any subset of the on-line database by location, date/time, and acquisition sensor. The on-line database is continually updated by periodic and on-demand repeated surveys

9 Case Study #1 Port of Los Angeles Main Channel Detection Exercise

10 Change Detection Blind Test Seven shapes placed in main channel SENSITIVE SENSITIVE SENSITIVE Images courtesy U.S. Navy and John Hopkins University

11 MBE Baseline Survey 7 lines, 50 m spacing, 15km total length 3 hours to collect 2.5GB raw data at 5 knots

12 MBE Baseline Data Bathymetric DTM and sidescan image layers 10GB processed data

13 Towed Sidescan Baseline Data Two 150m range passes in the main channel

14 Difference MBE Baseline & Resurvey Bathy Layers Detections in red displayed on bathy swath

15 Zoom on change detections.candidate MBE Change Detection

16 Multi-Sensor Verification: MBE Sidescan Change detection displayed on MBE imagery layer

17 Change detection in Towed Sidescan Data Mosaic image, target snippet, and raw data waterfall display of object

18 Multi-Sensor Change Detection Baseline Resurvey Bathy Object Towed Sidescan MBE Imagery

19 Multi-Sensor Change Detection & Target ID SENSITIVE MIED photo and target snippet images displayed on bathy layer with change detection

20 Future Development Improved change detection algorithms Potential enhancements Region-based processing using higher-order statistics Temporal dimension expansion Data fusion: MBE bathy, MBE sidescan, SSS sidescan CUBE Alternative Hypothesis method Soundings from baseline survey (green) and resurvey (red) CUBE surface from only baseline soundings CUBE surface from combined soundings. Alternative hypotheses indicate change

21 Case Study #2 Change Detection in the Shallow Survey Common Dataset

22 Change Detection in the Shallow Dataset A good opportunity to test the flexibility of the change detection tools All areas in the common dataset were investigated by processing and comparing data from the different sonars Date Jun, 2010 Dec, 2010 Mar, 2011 May, 2011 Jun, 2011 Manufacturer Sonar Geo GeoAcoustics Swath Applied SBP Acoustics Kongsberg 2040 R2Sonic 2022 Reson 7125

23 Change Detection in the Shallow Survey Common Dataset No changes were detected in most areas Seeps Area showed some indications of change that were investigated as shown on the following slides Date Jun, 2010 Dec, 2010 Mar, 2011 May, 2011 Jun, 2011 Manufacturer Sonar Geo GeoAcoustics Swath Applied SBP Acoustics Kongsberg 2040 R2Sonic 2022 Reson 7125

24 Shallow Seeps Area HUTT RIVER MOUTH SEAVIEW PENNINSULA The dominant artesian aquifer of the Hutt Valley groundwater system discharges into the harbour via submarine springs. Ted Taylor, Greater Wellington Regional Council

25 Large Scale Change Detection Blink Comparator No large changes from March (Kongsberg) to May (R2Sonic) Both data sets show a strong agreement with ENC representation of the bathymetry

26 Small Scale Change Detection and Analysis Change = Depth(A) Depth(B) Δ ± Tide offset

27 Small Scale Change Detection Detections Differences (A-B) cue operator to points of interest

28 Small Scale Change Detection Detections Zoom in to examine detection points

29 Small Scale Change Detection Detections Close examination shows seabed depressions at detection points

30 Small Scale Change Detection Bathy Profile PROFILE VIEW Bathy profile of detection point shows relative depths of layers Note accommodation for tide offset

31 Small Scale Change Detection 2D Swath Viewer R2SONIC 2D SWATH VIEW (MAY) KONGSBERG 2D SWATH VIEW (MARCH) PROFILE VIEW PROFILE VIEW Swath display of raw sounding data for both A and B layers Available toolset supports rapid investigation of detections

32 Change Detection in the Shallow Seeps Area PAPER CHART SPOIL GROUND S57 CHART 3D AREA VIEW (MARCH) CONCLUSION: detections are associated with the moorings for the buoys that are deployed to delineate the spoil ground.

33 High Frequency Change Detection Further Analysis of Shallow Survey Seeps Area Two lines collected with Kongsberg 2040 Line 15 collected at 2:57pm Line 16 collected at 3:16pm Change detection was used to compare line 15 with line 16 Comparison shows large change 2D swath views: LINE 15 LINE 16 3D area views: Conclusions: Objects on the far edge of sonar detection capability (e.g. chains) may cause false indications of change Some seabed changes (e.g. fresh water vents) can be rapid, continual, and episodic LINE 15 LINE 16

34 Summary and Key Points Presently available hydrographic survey technologies and methods can be effectively applied to object and change detection in ports and harbors. Positional accuracies, which are readily achievable, when using a hiresolution multi-beam sonar, a high accuracy MRU and GPS postprocessing, support the use of manual and semi-automated change detection techniques. The Shallow dataset example has shown, that the Triton Imaging System for change detection, is capable of exploiting available and diverse datasets to detect changes. Detection of rapid changes on the seabed may require a standing capability for adequate monitoring.

35 Final Points The change detection tool that I have presented provides a robust means for detecting temporal changes in a area of monitored seabed. However, the user must determine the causal mechanism for the changes, that are detected, and establish their importance, given the context of the situation.