Feature based passive acoustic detection of underwater threats



Similar documents
Passive acoustic threat detection in estuarine environments

TRAFFIC MONITORING WITH AD-HOC MICROPHONE ARRAY

Spectrum Level and Band Level

Robot Sensors. Outline. The Robot Structure. Robots and Sensors. Henrik I Christensen

Seismic Systems for Unconventional Target Detection and Identification

E190Q Lecture 5 Autonomous Robot Navigation

ANALYZER BASICS WHAT IS AN FFT SPECTRUM ANALYZER? 2-1

Julie Pullen, CSR Director Stevens Institute of Technology

Analysis of Wing Leading Edge Data. Upender K. Kaul Intelligent Systems Division NASA Ames Research Center Moffett Field, CA 94035

This document is downloaded from DR-NTU, Nanyang Technological University Library, Singapore.

Synthetic Aperture Radar: Principles and Applications of AI in Automatic Target Recognition

Application Note Noise Frequently Asked Questions

Adaptive feature selection for rolling bearing condition monitoring

Bandwidth-dependent transformation of noise data from frequency into time domain and vice versa

Underwater Acoustic Communications Performance Modeling in Support of Ad Hoc Network Design

MUSIC-like Processing of Pulsed Continuous Wave Signals in Active Sonar Experiments

An Energy-Based Vehicle Tracking System using Principal Component Analysis and Unsupervised ART Network

Timing Errors and Jitter

Environmental Effects On Phase Coherent Underwater Acoustic Communications: A Perspective From Several Experimental Measurements

telemetry Rene A.J. Chave, David D. Lemon, Jan Buermans ASL Environmental Sciences Inc. Victoria BC Canada I.

MATRIX TECHNICAL NOTES

The Effect of Network Cabling on Bit Error Rate Performance. By Paul Kish NORDX/CDT

Robot Perception Continued

Manual Analysis Software AFD 1201

WATER BODY EXTRACTION FROM MULTI SPECTRAL IMAGE BY SPECTRAL PATTERN ANALYSIS

VCO Phase noise. Characterizing Phase Noise

Building Design for Advanced Technology Instruments Sensitive to Acoustical Noise

The ArubaOS Spectrum Analyzer Module

How To Control Noise In A C7 Data Center Server Room

Anomaly Detection in Predictive Maintenance

RFSPACE CLOUD-IQ #CONNECTED SOFTWARE DEFINED RADIO

Testing of Partial Discharges in Generator Coil Bars with the Help of Calibrated Acoustic Emission Method

Establishing the Uniqueness of the Human Voice for Security Applications

Optimizing IP3 and ACPR Measurements

Environmental Remote Sensing GEOG 2021

SIGNAL PROCESSING FOR EFFECTIVE VIBRATION ANALYSIS

Audio Engineering Society. Convention Paper. Presented at the 129th Convention 2010 November 4 7 San Francisco, CA, USA

A Sound Analysis and Synthesis System for Generating an Instrumental Piri Song

APPLICATION NOTE. RF System Architecture Considerations ATAN0014. Description

A PHOTOGRAMMETRIC APPRAOCH FOR AUTOMATIC TRAFFIC ASSESSMENT USING CONVENTIONAL CCTV CAMERA

Multisensor Data Fusion and Applications

Implementation of a Gabor Transform Data Quality-Control Algorithm for UHF Wind Profiling Radars

Workshop Perceptual Effects of Filtering and Masking Introduction to Filtering and Masking

Application Note: Spread Spectrum Oscillators Reduce EMI for High Speed Digital Systems

The accurate calibration of all detectors is crucial for the subsequent data

Lecture 1-10: Spectrograms


A Microphone Array for Hearing Aids

Nuisance alarm suppression techniques for fibre-optic intrusion detection systems

3D Vision An enabling Technology for Advanced Driver Assistance and Autonomous Offroad Driving

System Design in Wireless Communication. Ali Khawaja

Making Accurate Voltage Noise and Current Noise Measurements on Operational Amplifiers Down to 0.1Hz

VEHICLE TRACKING USING ACOUSTIC AND VIDEO SENSORS

WHITE PAPER Professional Series Detectors Sensor Data Fusion February 2008

Module 13 : Measurements on Fiber Optic Systems

Harmonics and Noise in Photovoltaic (PV) Inverter and the Mitigation Strategies

PHYS 331: Junior Physics Laboratory I Notes on Noise Reduction

Research Article ISSN Copyright by the authors - Licensee IJACIT- Under Creative Commons license 3.0

Figure1. Acoustic feedback in packet based video conferencing system

Introduction to the EXAFS data analysis

Robotics. Lecture 3: Sensors. See course website for up to date information.

The Periodic Moving Average Filter for Removing Motion Artifacts from PPG Signals

Broadband over Power Line (BPL) Test Procedures and Equipment Authorization

AN Application Note: FCC Regulations for ISM Band Devices: MHz. FCC Regulations for ISM Band Devices: MHz

AN-007 APPLICATION NOTE MEASURING MAXIMUM SUBWOOFER OUTPUT ACCORDING ANSI/CEA-2010 STANDARD INTRODUCTION CEA-2010 (ANSI) TEST PROCEDURE

WAVELET ANALYSIS BASED ULTRASONIC NONDESTRUCTIVE TESTING OF POLYMER BONDED EXPLOSIVE

Study of RF Spectrum Emissions in High Pressure Sodium and Metal Halide Lamps. Lawrence P. Glaister VE7IT, Automation Engineer.

Detection of Magnetic Anomaly Using Total Field Magnetometer

Galaxy Morphological Classification

A Review of Anomaly Detection Techniques in Network Intrusion Detection System

Command-induced Tracking Jitter Study I D. Clark November 24, 2009

Instrumentation for Monitoring around Marine Renewable Energy Devices

Fundamentals of modern UV-visible spectroscopy. Presentation Materials

High Resolution Spatial Electroluminescence Imaging of Photovoltaic Modules

Sampling Theorem Notes. Recall: That a time sampled signal is like taking a snap shot or picture of signal periodically.

PRAKlA SEISMDS "V V PRAKLA-SEISMOS AG

Experiment 7: Familiarization with the Network Analyzer

MICROPHONE SPECIFICATIONS EXPLAINED

Online Filtering for Radar Detection of Meteors

Airborne Sound Insulation

Exhaust noise control case study for 2800 class locomotive

Use Data Budgets to Manage Large Acoustic Datasets

Technology in the littorals Below the surface and in-shore. Linus Fast R&D SwAF HQ, Naval Dpt linus.fast@foi.se

FFT Spectrum Analyzers

Quarterly Progress and Status Report. Measuring inharmonicity through pitch extraction

Development and optimization of a hybrid passive/active liner for flow duct applications

Title : Analog Circuit for Sound Localization Applications

Lastest Development in Partial Discharge Testing Koh Yong Kwee James, Leong Weng Hoe Hoestar Group

AN-837 APPLICATION NOTE

Supervised Classification workflow in ENVI 4.8 using WorldView-2 imagery

High Resolution RF Analysis: The Benefits of Lidar Terrain & Clutter Datasets

Measurement of Adjacent Channel Leakage Power on 3GPP W-CDMA Signals with the FSP

Monitoring of Internet traffic and applications

Preview of Period 3: Electromagnetic Waves Radiant Energy II

CONDUCTED EMISSION MEASUREMENT OF A CELL PHONE PROCESSOR MODULE

Transcription:

Feature based passive acoustic detection of underwater threats Rustam Stolkin*, Alexander Sutin, Sreeram Radhakrishnan, Michael Bruno, Brian Fullerton, Alexander Ekimov, Michael Raftery Center for Maritime Systems, Stevens Institute of Technology, Hoboken, NJ, 07030, USA ABSTRACT Stevens Institute of Technology is performing research aimed at determining the acoustical parameters that are necessary for detecting and classifying underwater threats. This paper specifically addresses the problems of passive acoustic detection of small targets in noisy urban river and harbor environments. We describe experiments to determine the acoustic signatures of these threats and the background acoustic noise. Based on these measurements, we present an algorithm for robustly discriminating threat presence from severe acoustic background noise. Measurements of the target s acoustic radiation signal were conducted in the Hudson River. The acoustic noise in the Hudson River was also recorded for various environmental conditions. A useful discriminating feature can be extracted from the acoustic signal of the threat, calculated by detecting packets of multi-spectral high frequency sound which occur repetitively at low frequency intervals. We use experimental data to show how the feature varies with range between the sensor and the detected underwater threat. We also estimate the effective detection range by evaluating this feature for hydrophone signals, recorded in the river both with and without threat presence. Keywords: Underwater, acoustics, detection, recognition, diver, harbor security 1. INTRODUCTION One of the most challenging aspects of port security is providing the means to protect against threats from under the surface of the water 1. In particular, it is felt that a significant terrorist threat might be posed to domestic harbors in the form of an explosive device delivered underwater by a diver using SCUBA apparatus. It is believed by the Departments of Defense and Homeland Security that terrorist organizations have been training individuals in SCUBA diving techniques 2. Several systems exist which can detect and track underwater moving targets. The most practical systems use active sonars 3. Other systems employ underwater bioluminescence 4, and lasers 5. However, while such systems can track moving objects, the problem of recognizing which, if any, moving entities are human divers is less well understood. This recognition problem lends itself to a passive detection approach, since these techniques can make use of prior knowledge of the specific sounds generated by a diver. Additionally, existing techniques which rely on active sonar devices, may be prohibited in many domestic harbors due to their environmental effects (e.g. disturbance of marine mammals). This paper addresses the problem of recognizing the presence of a diver using passive acoustic methods. We describe experiments to measure the acoustic radiation signals from divers using SCUBA systems which have led to the identification of the primary source of diver generated sound. Based on these characteristics, we then discuss methods for automating the diver detection process and present an algorithm which can detect the presence of a diver in passive acoustic signals and which is robust against the severe background noise inherent in urban harbor environments. Other work in the area of passive methods of diver detection includes fiber-optic hydrophone array techniques 6, 7, 8. *RStolkin@stevens.edu, ASutin@stevens.edu, SRadhakr@stevens.edu Photonics for Port and Harbor Security II, edited by Michael James DeWeert, Theodore T. Saito, Harry L. Guthmuller, Proc. of SPIE Vol. 6204, 620408, (2006) 0277-786X/06/$15 doi: 10.1117/12.663651 Proc. of SPIE Vol. 6204 620408-1

2. MEASURING THE ACOUSTIC SIGNATURE OF A DIVER A number of experiments have been carried out to investigate the acoustic signature of a diver and to identify important characteristics of this signature which might be used for the automated recognition of diver presence. The diver tests were conducted in the Hudson River near the Stevens Institute of Technology. Fig.1 shows the location of the tests. Study area Figure 1. Stevens Institute of Technology, Hoboken campus and Hudson River where the diver detection tests were conducted. Left Channe I -900 -ceo E -1000 40.OIe 48838 sx Time (ID eecldiv) -108.0 329 Ole Figure 2. Spectrogram in the frequency band below 100 KHz (Y axis) versus time (X scale 10 sec/div). The entire record is approximately 160 sec and the diver moved 120ft during this time. The periodic signal of the diver breathing is clearly visible at any measured distance (up to 60ft from the hydrophone). Proc. of SPIE Vol. 6204 620408-2

The depth in the area of the test was between 6 and 10ft. An omni-directional hydrophone was placed on the river bottom and the diver swam along several paths at different distances from the hydrophone. The diver swam in the middle of the water column at a height 3-5ft above the bottom. The diver swam along straight line paths of approximately 120 ft length, passing the hydrophone at ranges between 3ft and 20ft. Since the diver paths are known, and the diver was instructed to swim at a constant speed, it is possible to approximately estimate the range from the diver to the hydrophone, for any given portion of the recorded signal. Fig.2 presents the spectrogram of the recorded signal. Bright strips indicate the acoustic signal produced by the diver breathing. The amplitude of the signal increased when the diver swam toward the hydrophone (left half of the figure) and decreased when the diver moved away from the hydrophone (right half of the figure). The periodic signal is clearly visible at all measured ranges (up to 60ft) from the hydrophone. Fig. 3 shows the spectrum of the recorded signal for a diver s breathing sound and the difference between this and the spectrum for an example of background river noise with no diver present (Signal Noise Ratio, SNR). The time window of the spectral analysis was 1 second. The difference signal reveals a frequency band that provides a high signal/noise ratio for passive acoustic diver detection. 30 20 Diver signal Diver signal - minus river noise river noise (difference) (SNR) Spectral density, db 10 0-10 -20-30 -40-50 0 10 20 30 40 50 Frequency 60 70 for 80 90 best SNR Frequency, khz Figure 3. The spectra of the recorded signals for the breathing sound of a diver and SNR in the Hudson River. 3. ACOUSTIC NOISE IN HUDSON RIVER The diver detection range depends on the level of the background acoustic noise. This noise was recorded with the same single hydrophone placed near the bottom of the Hudson River. The spectra of the recorded signals were recalculated to the acoustic spectral density in db re 1 µ Pa Hz taking into account the frequency dependence of the hydrophone sensitivity. The examples of the Hudson River acoustic noise spectra are presented in Fig. 4. These measurements reveal the range of noise variation in the frequency band of interest that looks preferable for passive diver detection. Proc. of SPIE Vol. 6204 620408-3

Acoustic Noise db re 1 µpa/ Hz 90 80 70 60 50 Noise level 4 Noise level 1 Noise level 3 Noise level 2 40 0 10 20 30 40 50 60 70 Frequency (khz) Figure 4. Acoustic noise in Hudson River for various environmental conditions. Noise level 1: River noise with low traffic levels, at nighttime. Noise level 2: River with ferry and helicopter noise. Noise level 3: Rough surface conditions, and two helicopters present. Noise level 4: Severe background noise sources including airplane and helicopter traffic, speed boat and ferry. 3. FILTERING TO IMPROVE SIGNAL TO NOISE RATIO Examining the spectra of diver breaths and background noise samples (figure 3) indicates that a component of diver sound in a particular frequency range offers the highest signal to noise ratio (SNR). For the purposes of detecting diver presence, it is therefore sensible to filter all hydrophone signals at this frequency, which we will refer to as the prominent diver frequency. This prominent frequency component can be thought of as discriminatory, since amplitudes are relatively high during diver presence and relatively low for background noise. Therefore a simplistic approach to detecting diver presence might be to compare the amplitude of the prominent diver frequency signal component against a threshold value that is determined by the levels of acoustic noise in the same frequency band. If the threshold is exceeded, then we accept a hypothesis of diver presence. For such a method to succeed, the threshold must be set high enough that it exceeds background values for the vast majority of instances, in order to avoid false positive errors (indicating diver presence when no diver is present). This condition leads to an estimate of detection range. Figure 5 shows a sample of diver sound that has been narrow band-pass filtered at the prominent diver frequency. The resulting signal has then been smoothed, giving a signal envelope. The values of this signal are plotted against approximate range between the diver and the hydrophone. Also plotted are the equivalent values calculated for samples of background river noise. It is apparent that this detection method can only work at low levels of noise or at very short ranges (range 1 on figure 5). The method is also vulnerable to any noise source which also emits sound in the frequency band around the prominent diver frequency. In the following section we propose an alternative detection algorithm, which is robust against noise sources in this critical frequency band, and enables reliable detection against severe noise conditions and over extended ranges. Proc. of SPIE Vol. 6204 620408-4

3.50E-04 3.00E-04 Swimmer Noise level 3 (mean) Amplitude (Volts) (Volts) 2.50E-04 2.00E-04 1.50E-04 1.00E-04 Noise level 3 (max) Noise level 1 (mean) Noise level 1 (max) 5.00E-05 0.00E+00 0.0 7.5 15.0 22.5 30.0 37.5 45.0 52.5 60.0 67.5 75.0 Range 1 Range 2 Distance Range (ft) from hydrophone to diver Figure 5. Variation in diver signal, filtered at prominent diver frequency, with range. Comparison with different levels of background noise in urban environments. Noise level 1: River noise with low traffic levels, at nighttime. Noise level 3: Rough surface conditions, large waves and two helicopters present. Even with low noise levels, only intermittent detection is possible at moderate ranges (range 2). Error free detection does not appear possible beyond very short ranges (range 1). 4. FEATURE BASED DETECTION SCHEMES More robust detection may be achieved by trying to extract features from the hydrophone signal. A feature is a number which is some function of the raw sensory data (i.e. the time varying hydrophone signal). The feature should be tightly coupled to the event which we are trying to recognize, i.e. exhibit a typical range of values when the diver is present which is statistically significantly distinct from the typical range of values when the diver is not present. Features can be used to discriminate between two events using an appropriate discriminating function which is often learned from training data, i.e. historical sets of feature values with known ground-truth (diver present or not). Generally, appropriate selection of features is a difficult problem. It is often desirable to use multiple features of different kinds to achieve good classification. We have exploited a single feature, which can be used with a simple discriminatory thresholding function to determine diver presence. This feature exploits two key observations: 1) The primary source of diver sound relates to the diver s breathing. 2) The sound has a wide frequency spectrum, but is most distinct from background noise at around the prominent diver frequency. Our feature (which we term the swimmer number ) thus attempts to evaluate to what extent an object is present which emits packets of noise in the prominent diver frequency region at regular intervals, such that these intervals fall within a typical range of human breathing rates. Proc. of SPIE Vol. 6204 620408-5

4.1 Computing the swimmer number The procedure for computing the swimmer number feature is as follows. Firstly the raw hydrophone signal is band-pass filtered in the prominent diver frequency range, in order to improve SNR (see section 3). Figure 6 shows an example of a hydrophone signal, recorded for a diver in the Hudson River, after band-pass filtering. io3 0.5 1ALA 0, 1rt i! 0 10 15 20 Time, Sec Figure 6. Signal for a diver following band-pass filtering at prominent diver frequency. Next, an envelope is fitted to the signal. Negative values are removed and consecutive peaks are connected. The resulting signal is then smoothed by low-pass filtering (figure 7). 6 5 1 0 10 15 20 Time, Sec Figure 7. Diver signal envelope. This envelope is now Fourier transformed. Figure 8 shows the spectrum of the envelope for an example of a diver in the Hudson River, whereas figure 9 shows the spectrum of the envelope for an example of typical Hudson River background noise. Proc. of SPIE Vol. 6204 620408-6

3x108 x io6 (ID N : 1.5 CO6O81 Frequency, Hz Figure 8. River with diver present. 0.2 0.4 0.6 0.8 Frequency, Hz Figure 9. River with no diver. In signals recorded with a diver present, there is clearly a cluster of energy around the diver s breathing frequency (around 0.3 Hz or three breaths per second) which is not present in background river noise. This gives rise to a useful discriminating feature. We can now integrate over a likely range of human breathing rates (see figure10) to give a single number, the swimmer number. Integrating over a range of frequencies is useful since it enables generality, i.e. the algorithm can cope with divers who breathe at a variety of different rates. It should be noted that generality comes at a cost. By integrating over a range of possible breathing rates, we sacrifice optimal detection performance for any specific breathing rate. In terms of detection errors, we are trading off false positives (claiming that there is a diver present when there isn t) and false negatives (failing to detect a diver when one really is present). This trade off can be adjusted by adjusting the range of integration. Due to the low resolution of the spectrogram, in cases where the integration range is chosen to be small, and where the diver breathes consistently at one breathing rate, the procedure can become informationally equivalent to simply identifying the peak value in the integration range. In this paper, for proof of principle, we have chosen to integrate over a relatively small range, between 0.2 and 0.4 Hz. Future work should use additional diver data to explore optimum integration ranges in different conditions. Figure 10. Integration over range of possible human breathing rates to give a single number, the swimmer number. Proc. of SPIE Vol. 6204 620408-7

4.2 Variation of swimmer number with range and noise level The swimmer number, calculated by integrating the spectrogram of the hydrophone signal envelope, is useful as a discriminating feature, in that it takes large values when a swimmer is present and small values when no swimmer is present. Thus, diver presence can be detected using a simple thresholding function. Swimmer Numbers above the threshold are classified as indicating diver presence, those below are classified as indicating no diver presence. Care must be taken when choosing this threshold value. Threshold choice, background noise levels and maximum detection range are closely related. Swimmer Number values were calculated for samples of hydrophone data featuring a diver in the Hudson River. It has been possible to estimate the range from the diver to the hydrophone for each sample (see section 2). We can thus estimate the fall off in Swimmer Number with range (figure 11). It is convenient to work with the logarithm of the Swimmer Number values, giving a log(swimmer Number) plot, expressed in db scale. Swimmer Number, (db) 70 60 50 Hudson with Swimmer Noise level 3 Noise level 2 Noise level 1 Noise level 4 Linear (Hudson with Swimmer) 40 30 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 Range 1 Range 2 Range 3 Range from Hydrophone to swimmer (ft) Figure 11. Drop off in log Swimmer Number value with range. Comparison with Swimmer Number calculated for various ambient noise conditions. Noise level 1: River noise with low traffic levels, at nighttime. Noise level 2: River with ferry and helicopter noise. Noise level 3: Rough surface conditions, large waves and two helicopters present. Noise level 4: Severe background noise sources including airplane and helicopter traffic, speed boat and ferry. Ranges 1 and 2 are marked for comparison with figure 5. Superimposed on this plot are the log(swimmer Number) values for various samples of background noise for which no diver was present. Extrapolating the plots reveals a theoretical upper limit for the range at which a swimmer can just be detected. In theory, if a discriminating threshold is set to a value just above the Swimmer Number value for Noise level 1, we might expect to detect a diver as distant as range 3 (compare with range 1 obtainable with a simple filtering technique, see figure 5, section 3). However, in practice it is necessary to use a more conservative threshold to ensure robustness to noise levels, which vary considerably in an urban harbor environment. Again, there must be a design tradeoff between extending the Proc. of SPIE Vol. 6204 620408-8

maximum detection range and achieving robustness of detection decisions at lesser ranges. To investigate variation of detection ranges and appropriate threshold levels with noise levels, Swimmer Numbers were calculated for various kinds of extreme background noise which are present intermittently in the Hudson River (see figure 11). During occasional episodes of extreme background noise (e.g. Airplane, two ferries, speedboat and helicopter ) the possible detection range is considerably reduced. However, it should be noted that these levels of noise are so extreme as to prohibit conversation between two personnel standing together on a boat during these conditions. 4.3 Discussion of results Intermittent episodes of extreme noise are problematic. We can either settle for a conservative (high valued) discrimination threshold (severely reducing the detection range), or we must expect occasional false positive detection errors (i.e. noise levels trigger the detection system when no swimmer is present). One approach to this difficult problem would be an adaptive threshold, i.e. an algorithm which continually adjusts the threshold in response to varying noise levels. This is discussed as a possible extension to this work (section 5). A limitation of the suggested approach is that detecting diver presence is contingent on a relatively long segment of hydrophone signal. Since our algorithm attempts to identify packets of sound which occur at the diver s breathing rate (approximately one breath every three seconds), each Swimmer Number value must be derived from at least 6s of sensor signal. The results described in this paper were derived using 10s portions of signal. This poses a problem of localization, i.e. the diver may change his position during the detection process. However, the focus of this work is addressing the problem of detecting diver presence (distinct from the problem of estimating diver position). This work might conceivably be combined with additional techniques in order to also track a diver s trajectory. Additionally, our expert divers have reported the need to move very slowly in the turbid and cluttered river environment, with typical speeds of around 1ft /s, causing perhaps ±5ft error on position measurements derived from 10s of hydrophone signal, a reasonable and realistic level of accuracy for a difficult, noisy environment. Additionally, it might be possible to use the feature based Swimmer Number approach in combination with the simplistic approach, described in section 3, of utilizing the amplitude of the prominent diver frequency filtered signal itself as an indication of diver presence. This might provide for an early warning of potential diver presence, followed a few seconds later by a more informative Swimmer Number evaluation. The Swimmer Number can also be continuously evaluated for a moving window of the past few seconds of hydrophone signal. It is worth noting that far greater detection ranges might be possible in quieter waters. The noise problems addressed in this paper are extreme and could be viewed as a worst case scenario. 5. FUTURE WORK There are several ways in which detection range might be extended. Firstly, we are planning experiments which will use highly directional hydrophone arrays to improve SNR. Secondly, the application of various signal processing techniques may improve detection range. These include incorporating the use of matched filtering techniques and also noise suppression techniques based on understanding of the spectra, directivity and correlation properties of common noise sources. Future work should more fully explore the effects of various parameters including different diving apparatus, different depths and different surface and water conditions. In order to ensure the generality of the algorithm, future experiments should record divers breathing at different rates and explore the effects of performing the frequency space integration over different ranges of breathing rate (see section 4.1). Future work may also explore algorithms for intelligently varying a discriminating threshold value (see section 4.2) in response to changing background noise conditions. This paper has addressed the problem of detecting the presence of a diver. The main advantage of this passive technique over active sonar methods, is that it enables a source of sound to be classified rather than only tracked. Localization of the diver presents an additional research problem. Future work may examine probabilistic methods for estimating diver position and trajectory, given multiple signals due to a diver traveling across a distributed array of sensors. A directional Proc. of SPIE Vol. 6204 620408-9

hydrophone array may also provide a means of reasoning about diver location. Additionally it might be useful to combine the discriminatory capabilities of the passive methods, with the localization capabilities of active methods. 6. CONCLUSIONS Divers radiate an acoustic signal in a wide frequency band that can be passively detected. Diver presence can be characterized by regularly repeating packets of ultrasound which occur within a range of human breathing rates. These diver characteristics have been derived from a series of experiments, measuring the acoustic radiation from divers in the Hudson River. Exploiting these characteristics, we have presented an algorithm which can robustly detect the presence of a diver from a single, passive hydrophone signal, even under conditions of extreme background noise. This algorithm has been validated using real acoustic noise data, recorded in a noisy urban river. Using this system, it is possible to detect divers at ranges which are useful, albeit limited. Further refinements of our signal processing techniques may extend this range. We also plan to extend this range with experiments using a highly directional hydrophone array. REFERENCES 1. R. Hansen, Underwater Port Security Signal Processing Challenges, A book of abstracts for the IEEE Underwater Acoustic Signal Processing Workshop, page 9, 2005. 2. R. Manstan, Measurements of the Target Strength and Radiated Noise of Divers Wearing SCUBA Equipment, A book of abstracts for the IEEE Underwater Acoustic Signal Processing Workshop, page 13, 2005. 3. K. Shaw, R. Scott., G. Holdanowicz, Sonar sentinels on guard for submerged swimmers, Jane's Navy International, September 2005. 4. D. Lapota, Night time surveillance of harbors and coastal areas using bioluminescence camera and buoy systems, Proceedings of SPIE, Photonics for Port and Harbor Security, 5780, Pages 128-137, 2005. 5. Weidemann, G. Fournier, L. Forand, P. Mathieu. In harbor underwater threat detection/identification using active imaging, Proceedings of SPIE, Photonics for Port and Harbor Security, 5780, pages 59-70, 2005. 6. Anti-terrorist swimmer detection system to be evaluated by US Navy. http:// www. qinetiq. com/ home/ newsroom/ news_releases_homepage/2004/2nd_quarter/cerberus.html. 7. S. Stanic, C.K. Kirkendall, A.B. Tveten, and T. Barock. Passive Swimmer Detection, NRL review, 2004. http://www.nrl.navy.mil/content.php?p=04acousticsreview 8. D.Hill, P. Nash. Fibre-optic hydrophone array for acoustic surveillance in the littoral, Proceedings of SPIE, Photonics for Port and Harbor Security, 5780, pages 1-10, 2005. Proc. of SPIE Vol. 6204 620408-10