Paper 050328R received Apr. 26, 2005; revised manuscript received Jun. 28,2005; accepted for publication Jul. 5, 2005; published online Feb. 27, 2006.
NL-2628 CD Delft, The Netherlands
sst
Titt
Dtn
oww
1 Introduction
Overall situational awareness is of the utmost importancefor a command team aboard of a naval vessel. Picture com-pilation of the surroundings of the ship is done by all theavailable sensors aboard, with the objectives of knowingwhere all “players” are and to acquire a positive identifica-tion of those units around the naval vessel. By using activesensor systems the naval vessel gives away its position.Consequently, the use of passive systems is necessary.Aboard the Royal Netherlands Navy �RNLN� Air DefenseCommand Frigates �ADCF� are several passive sensor sys-tems are available, an one of them will be a dual-band IRsearch and track �IRST� system. The IRST system con-stantly scans the environment of the ship and automaticallygenerates 2-D tracks of target objects. The missing param-eter is the distance estimation to the target. By applyingdifferent passive ranging methods it should be possible topassively obtain 3-D data of a target, which increases thesituational and tactical awareness of the command teamaboard the naval vessel and enables them �when the accu-racy of the distance estimate is high enough� to use a pas-
Downloaded from SPIE Digital Library on 21 Jan 2010
ive sensor as a fire control sensor. There is currently noystem available that acquires 3-D information without ac-ively transmitting energy.
This research focuses on a dual-band IRST system.here are several reasons why the dual-band IRST system
s placed aboard the ADCF. The IRST system contributes tohe ship multisensor database, and its contribution is par-icularly vital in the following conditions:
1. during ship emission restrictions2. when active sensor�s� performance is degraded by
electronic countermeasures3. when active sensor�s� performance is degraded by
multipath interference4. when the environment favors infrared propagation
over radar propagation
espite all these contributions, there is one disadvantage:he current passive sensor to be placed aboard the ADCF isot able to obtain 3-D tracks that include range.
In this paper, we present our study of the feasibility ofbtaining range information from passive IR sensors only,ith sufficiently high accuracy, in particular in the anti-air-arfare �AAW� scenario. The study consists of the applica-
February 2006/Vol. 45�2�
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
TesF
Istst�sa
rftfpWi
2
2Booasd
de Visser et al.: Passive ranging using an infrared search and track sensor
tion, comparison, and combination of different ranging al-gorithms. We limit ourselves to the following fourboundary conditions:
1. The output of the sensor for the distance estimationconsists of 2-D track information �bearing and eleva-tion� and bearing and elevation rate.
2. The images belonging to the track information of theIR cameras are available.
3. Real-time aspects of processing, uncertainty not di-rectly linked to the distance estimation and design
Fig. 1 Dual-band stabilized sensor head of the long-range IRSTsystem. The system rotates at 1 Hz and has elevation coverageoptimized for sea skimming missiles and surface contacts. The plat-form is stabilized over three axes. The sensor at the left side, asseen from the back, is the long-wavelength �LW� �8.0= to 10.0-�m� sensor and the medium-wavelength �MW� �3.0 to 5.0-�m�sensor is at the right.
Table 1 Compa
Author Sensor M
Blackman and Popoli2 IRST SA
Van Sweeden et al.3 IRST S
de Groot4 IRST I
Kemp5 IRST A
Schwering et al.6 Staring IRST A
Aytac and Barshan7 Infrared I
Sasaki et al.8 CCD camera I
Ono and Komatsu9 PCC I
Ruben and Michalowicz10 IRST T
An N/A denotes that the information cannot be dstereo imaging; IDD, intensity distance diagram;sis; AS, apparent surface of the target; APM,
counting camera.
Optical Engineering 026402-2
Downloaded from SPIE Digital Library on 21 Jan 2010
aspects of the used IRST system, fall beyond thescope of this paper.
4. Meteorological and scenario data is available-in realtime.
he IRST �Ref. 1� consists of two IR sensors, processingquipment, and several control and supply units. Both sen-ors are placed in a stabilized sensor head that is shown inig. 1.
The sensor will be placed aboard the Netherlands ADCF.ts main purpose is to detect and track incoming sub- andupersonic sea-skimming missiles. Before actual integra-ion of the sensor in the combat management system of thehip, a preproduction model is designed by the manufac-urer and placed and evaluated at a land-based test siteLBTS� and a sea-based test site �SBTS�. The system isubmitted to cold water trials �CWTs� and warm water tri-ls �WWTs� at, respectively, the LBTS and the SBTS.
The paper is organized as follows. Section 2 summarizeselated work and discusses the most appropriate methodsor passive ranging with the IRST system. Section 3 con-inues with the combined method for passive ranging usedor evaluating the CWT scenario. Section 4 gives the ex-erimental results. Results of a dynamic scenario duringWTs are also presented. Section 5 is the discussion and
n Sec. 6, the conclusions are formulated.
Passive Ranging
.1 Related Workefore discussing our research in the next sections, anverview of related papers is presented here. Related workn passive ranging is reported by a considerable number ofuthors. These references describe a variety of possible pas-ive ranging techniques, applications and combinations ofifferent sensor systems. A summary is shown in Table 1.
f related work.
Fusion Experiments
TMA, MS, Yes Real and simulated
APM, AS Yes None
ive TMA No Real and simulated
No N/A
No None
y, 2 sensors Yes Real and simulated
y, 2 sensors No Real
y No Real and simulated
No Simulated
form the reference. Abbreviations used are S,target motion analysis; MS, multispectral analy-pheric propagation model; and PCC, photon-
rison o
ethod
, IDD,S
, IDD,
nteract
S
PM
ntensit
ntensit
ntensit
MA
educedTMA,atmos
February 2006/Vol. 45�2�
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
ftfmfittu
I
wt��seLpdai
I
wmic
2Pilcwf
ewagdtgtf
cc
A
wc
de Visser et al.: Passive ranging using an infrared search and track sensor
Techniques mentioned for passive ranging are stereo im-aging �S�, use of intensity-distance diagrams �IDD�, targetmotion analysis �TMA�, multispectral analysis �MS�, calcu-lating the apparent surface of a target �AS�, using atmo-spheric propagation models �APMs�, combining a prioriknowledge from an active sensor such as a laser rangefinder, and using a dual-platform combination for sensorfusion. An important conclusion from the references is thatlittle research is done for single-operating long-range IRsensors. An actual implementation and evaluation for anactual existing IRST system with live target data has notbeen found. Furthermore, the combination of different pas-sive ranging techniques is, to the best of our knowledge,not practically implemented in an IRST system.
Based on the methods described in the related work ofTable 1 �S, IDD, TMA, MS, AS, APM� and the output ofthe IRST sensor, TMA, APM, and AS are further investi-gated. Stereo imaging �S� is not a good candidate becausethe instantaneous field of view of the sensors limits the useof this method for passive ranging to a distance of about2 km. The IDD and MS methods are not taken into accountbecause of unavailable information.
2.2 Methods for Distance EstimationThe output of the IRST sensor consists of bearing and bear-ing rate, elevation and elevation rate of the target, and theimages of both camera systems. This information is deliv-ered in successive time steps. The three selected methodsare described in the next subsections. A comparison of themethods concludes in Sec. 2.2.
2.2.1 APMsBased on the altitude of the target and more scenario infor-mation such as meteorological data, it is possible to make adistance estimate when the target is observed for the firsttime. There are several distortions to deal with at IR fre-quencies when calculating the distance to the horizon or anobject at the horizon.11–14 There are effects that cause ab-sorption and the effective earth radius will change becauseof refraction. Furthermore scintillation �rapid fluctuationsin signal amplitude �fading rate per second� due to atmo-spheric effects� also affects the propagation of IR rays.These effects are not always a disadvantage for the distanceperformance of IR sensors. Certain atmospheric conditionsenlarge the distance of first detection of an object in the IRspectrum.
Various models are available to estimate the propagationof IR rays in the trajectory of interest. Meteorological inputdata is needed and of great importance for the accuracy ofthe used model.14 Some models are
1. MODTRAN, a computer model for prediction of theoptical properties of the atmosphere
2. IRTool, a model for prediction of the effect of refrac-tion
3. ARTEAM, a ray-tracing model for electro-optic �EO�applications
4. IRBLEM, an IR boundary layer effects model15
Since IRBLEM is available for the RNLN and the model isan improved version of MODTRAN for usage in a mari-
time environment, we use this model. e
Optical Engineering 026402-3
Downloaded from SPIE Digital Library on 21 Jan 2010
The IRBLEM estimates the influence of absorption, re-raction, and scintillation on the propagation of IR waves inhe so-called “marine boundary layer” above the sea sur-ace. Previous research on this subject11–14 shows that thisethod provides realistic first-distance estimates. Besides arst-distance estimate, the model IRBLEM gives the effec-
ive transmission for the propagation path. Furthermore, thearget intensity is approximated from the IR images by these of a calibration equation:
=� �Ipeak�target − Ibackground���
�backgroundNEILW�band, �1�
here I is the intensity, Ipeak�target is the peak intensity of thearget, Ibackground is the mean intensity of the background,� is the solid space angle of the irradiance field of viewIFOV�, �background is the standard deviation of the emptyky background, and NEILW�band is the estimated noisequivalent irradiance of the LW camera. During CWTs theW band dominated the MW band, which complicates theossible combination of both bands. We choose to use theominant spectral band. Because an object emits radiationll around and we receive the intensity at a distance R, thentensity �I� is16
= W�
R2 , �2�
here W is the target radiant intensity, and � is the trans-ission to the distance R. The radiant intensity of the target
s assumed known, but estimated here from the TMA pro-ess.
.2.2 AS of the target
rocessing and interpretation of the recorded images addsnformation to the solution of the distance estimation prob-em. The optical AS of the target changes when distance ishanging. The region of interest �i.e., the incoming target�ill grow while approaching the ship �larger apparent sur-
ace�. Figure 2 shows some images taken by the sensor.The AS method requires an extended target �a target that
xtends over multiple pixels� imaged at different distances,ith the distance difference known.3 The first detection of
n incoming target will display a tiny spot in the image. Aray-value-intensity-based threshold procedure is used toetermine the size of the target. The IRST sensor first de-ects the target. Second, examination of the boundary re-ions in the image around the target of interest is performedo obtain a threshold value. Finally, computation of the sur-ace of the target of interest is done.
The area of an extended target may also change with ahanging viewing aspect, but if we assume a straight in-oming target, then
=c
R2 , �3�
here A is the number of pixels of the target, c is theonstant of proportionality, and 1/R2 represents the influ-
nce of distance on the target area. From this it follows that
February 2006/Vol. 45�2�
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
osma
twhsf
pperritrEa
B
de Visser et al.: Passive ranging using an infrared search and track sensor
dA
dR= − 2
c
R3 . �4�
Combining Eqs. �3� and �4� yields
R = − 2A1
�dA/dR�. �5�
When dA /dR is approximated by �A /�R, representing theratio of the measured changes in target area and distancerespectively, an estimate of the distance R is obtained. Alimitation of this method is that an initial reference distanceand the changes in distance are required. Another limitationis that this method is only useful for extended targets.
2.2.3 TMAA distance estimation method based on angle measure-ments is TMA. Using target tracking based on bearing in-formation created by own-ship movement, the distance isestimated. Because of the very accurate bearing measure-ments of IR sensor systems, TMA looks very suitable touse for distance estimation with an IRST sensor. There areseveral variations of TMA; single-leg Ekelund17 for targetswith bearing rate, cross measurements for targets with no orsmall own velocity, or the usage of TMA-based filters18 fordynamic targets in operational scenarios. The TMA prin-ciples used during our tests are described in the following.
TMA used principle for dynamic targets. The TMAprinciple during WWTs is based on batch-type filters.Those types of filters have been used in passive sonar track-ing systems. This batch-type filter is very suitable in a pas-sive tracking system, especially in an optronic system,where angle measurements are very accurate. Due to thelack of range measurements in a passive system, the com-plete state vector of the target cannot be resolved. However,if the sensor is moving with a nonconstant course or speed,the complete target state vector is automatically resolved.The requirement is that the sensor position with respect toan inert reference point is known at any time with sufficient
Fig. 2 IR images from an airplane �left is LWIRmean has been removed for each line.
accuracy. The batch filter is very suitable for adding a pri-
Optical Engineering 026402-4
Downloaded from SPIE Digital Library on 21 Jan 2010
ri knowledge about the target dynamic constraints or otherensor measurements to the batch. Also passive measure-ents from other sensors of other platforms could be
dded.To be able to estimate the state of the target, an assump-
ion must be made about the target dynamic model. Nexte illustrate a constant velocity target model. This modelas 6 degrees of freedom. Other target models, such as con-tant height and fixed position, reduce the degrees ofreedom.
Figure 3 visualizes the geometry of the dynamic TMArocess is visualized. Here psensor�t� is the 3-D Cartesianosition of the IRST sensor with respect to a fixed refer-nce point O, pt�t� is the position of the target at time t withespect to the fixed reference point O, pS�t� is the sensorelative �measured� position of the target at time t, and vt�t�s the speed vector of the target. Suppose our sensor posi-ion is known at every time t with respect to an inertialeference point O. The measured bearing B�t� and elevation�t� �not given in 2-D Fig. 3, this is the measured elevationngle of the target� are given by
�t� = arctan�xs�t�/ys�t�� , �6�
is MWIR�; the axes are pixel numbers; and the
, right
Fig. 3 Visualization of dynamic TMA vectors and geometry.
February 2006/Vol. 45�2�
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
¯
TmFcit
x
a
y
wnbb
y
A
y
wttt
R
E
R
awm
de Visser et al.: Passive ranging using an infrared search and track sensor
E�t� = arcsin� zs�t��ps�t��
� , �7�
where
ps�t� = xs�t�ys�t�zs�t�
= pt�t� − psensor�t� , �8�
and xs�t�, ys�t�, and zs�t� are the components of state vectorof the target. The model equation at t0 looks like:
H = 1 0 0 �t − t0� 0 0
0 1 0 0 �t − t0� 0
0 0 1 0 0 �t − t0� , �9�
where H is the state model and t and t0 are the time and thetime at t=0. Then,
H�pt�t0�vt�t0�
� = psensor�t� + ps�t� = psensor�t� + R�t�
� sin B�t�cos E�t�cos B�t�cos E�t�
sin E�t� , �10�
where R�t� is the range at time t. Eliminating the rangeequation, by using the following transformation:
Lt = � cos B�t� − sin B�t� 0
− sin B�t�sin E�t� − cos B�t�sin E�t� cos E�t� ��11�
results in
LtH�pt�t0�vt�t0�
� = Ltpsensor�t� . �12�
This is the basic equation for each observation. For everymeasurement B, E of a track at t= ti we can add the basicequation to an overdetermined set of equations, which canbe resolved by using singular value decomposition �SVD�:
Ax̄ = b̄ ,
USVTx̄ = b̄ ,
x = VS−1UT b̄ . �13�
If the set is nonsingular the state vector x̄ can be resolveddirectly, otherwise the singular diagonal elements in S−1
can be replaced by variables. This will result in a solutionspace, where the real solution can be written as a linearcombination of range and null space vectors. With a singu-lar set of equations some observable states can be derivedsuch as time to go �TTG�, time, and direction of the closest
point of approach �CPA�. i
Optical Engineering 026402-5
Downloaded from SPIE Digital Library on 21 Jan 2010
MA used principle for stationary targets. The geo-etrical layout of the static TMA process is visualized inig. 4. A limitation is that the ship must move over time toollect different bearing information. Movement of the ships indicated by the thick black arrow to the north ���t�. Forhe bearing at time t=0 the following equations apply:
= R sin�B1� , �14�
nd
= R cos�B1� , �15�
here x is the x coordinate of the target, y is the y coordi-ate of the target, R is the range of the target, and B1 is theearing from the own ship to the target at time t=0. Com-ining Eqs. �14� and �15� gives
= �1/tan�B1��x . �16�
t time t= t, Eq. �16� becomes
= v�t + �1/tan�B2��x , �17�
here v is the speed of the own ship, �t is the time be-ween the bearing measurements and B2 is the bearing atime t= t. Combining Eqs. �16� and �17� gives the intersec-ion point �x ,y�. The range equation is
= �x2 + y2�1/2�� v�t
1/tan�B1� − 1/tan�B2��2
+ ��1/tan�B1��v�t
1/tan�B1� − 1/tan�B2� 2�1/2
. �18�
quation �18� is simplified to
=− v�t sin�B2�sin�B1 − B2�
. �19�
The most relevant assumptions with this TMA analysisre �1� the target is nonmoving or travels in a straight lineith a constant velocity and �2� the availability of a mini-um of two sets of track data of the target.A limitation of this method is that the greater the veloc-
Fig. 4 Geometrical visualization of the static TMA bearings.
ty of the target in relation to the velocity of the own plat-
February 2006/Vol. 45�2�
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
abo
tieimde
3
Uodia
3Uar
de Visser et al.: Passive ranging using an infrared search and track sensor
form, the worse will be the distance estimation. With thisTMA method it is possible to estimate the distance whenthe target comes straight toward the ship.
2.2.4 Comparison of methodsA comparison between the three methods is given in Table2. The distance estimates of the methods APM, AS of thetarget, and TMA �stationary� are implemented separately aswell as combined in the CWT sections. The TMA methodhas a direct numerical output for the distance estimation,and the APM and AS methods require an initial reference.This reference distance estimation is given by the TMAmethod.
3 Combined Method Used in CWT ScenarioUnder the assumption that all methods described in Sec. 2provide useful, but not ideal, range data, we consider thepossibility of combining the different range estimates toimprove the range result. The distance estimation of thespecific methods is then input for a fusion process. Thefusion itself is based at decision levels, so decision levelfusion is chosen �in Ref. 19 a distinction is made betweendata-, feature-, and decision-level fusion�. As methods forfusion we choose “best method,” “voting,” and “weightedaverage.”
3.1 Fusion SchemeAn important factor for information fusion is that the out-put of each method must give a confidence level that de-scribes the �un�certainty. A confidence level expresses aconfidence or belief in the uncertainty of a distance esti-mate at a certain time, which is expressed as a numberbetween 0 and 1. The confidence values are used to indicate
Table 2 Advantages an
Method Advantage Disadvanta
APMAdditional scenarioinformation available fromatmospheric model
MeteorologMany inpuTarget radIndicativeneeded
AS Moving target indicationavailable
Initial referChanges iUseful for
TMA �stationary�Only bearing and ownship information needed
InaccurateShip move
TMA �dynamic� Resolving 3-D positionand velocity of the target;suitable for adding a prioriknowledge and othersensor measurements
Own shipsknowledgerequired; sor surfacerange �les
an order in the uncertainty of the methods. This means that
Optical Engineering 026402-6
Downloaded from SPIE Digital Library on 21 Jan 2010
higher confidence value implies a lower uncertainty num-er for a method. Figure 5 presents the layout for the fusionf the distance estimations.
The decision-level information fusion method adjustshe weight of each method by means of a mapping �weight-ng� function. This mapping function requires one param-ter �ui, dependent on the confidence level� for each methodand adjusts differences between the three methods. Theapped distance estimation methods are combined with a
ecision-level fusion function to acquire a single distancestimation.19,20
.2 Uncertainty in the Methods and ConfidenceLevels
ncertainty in the three chosen methods is present becausef parameter predictions, inaccurate measurements, and/orependency on other methods. We implemented uncertaintyn all input parameters for the different range equations,nd the input parameters are assumed independent.
.2.1 APMncertainty in the APM method is incorporated by adding
n uncertainty percentage to all input parameters in theange Eq. �2�. The uncertainty in the atmospheric input pa-
vantages per method.
Range Equation
formation needed.eters
ensity neededtance estimate and velocity of target
R = �W�
I�
11
istance needed.nce neededed targets
R = − 2A1
dA/dR
city target is larger than own shipequired
R =− v�t sin�B2�sin �B1 − B2�
uvers, or a prioritarget dynamicsfor slow moving airat relative short
10 km�
Dependent on thegeometry of the tracking
problem
d disad
ge
ical int paramiant intfirst dis
ence dn distaextend
if veloment r
maneabout
uitabletargets
s than
Fig. 5 Information fusion flow diagram.
February 2006/Vol. 45�2�
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
tdmr
C
3Uuowmt
3UarEr
Bel
3Tano
3TTmbu
3T
R
w=
de Visser et al.: Passive ranging using an infrared search and track sensor
rameters of the IRBLEM gives an overall uncertainty in thetransmission �. First, the sensitivity of the transmission forthe different input parameters is determined �atmosphericinput parameters like solar radiation, wind speed, etc. havenegligible influence on the total transmission�. Then fourremaining input parameters for determining the transmis-sion are combined in one uncertainty number for the trans-mission:
where � is the transmission with uncertainty; �irblem is thetransmission without uncertainty dependent of the pressure�p�, the air temperature �air�temp�, the relative humidity�rel�hum� and the water temperature �water�temp�; �� isthe uncertainty in the transmission �. The uncertainty iscalculated by:
�� = ����p�2 + ���air�temp�2 + ���rel�hum�2
+ ���water�temp�2�1/2, �21�
where ��p is the uncertainty in the transmission � whenonly the uncertainty in the pressure is incorporated. Weapplied a similarly method for the uncertainties in air tem-perature, relative humidity, and water temperature. The un-certainty in the pressure is calculated by
where �p�min is the transmission calculated with the modelfor the minimum pressure and �p�max is the transmissionwith the maximum pressure. Analogous to Eq. �22�, theuncertainty for the other three atmospheric input param-eters is determined. Analogously to the uncertainty calcula-tions for the transmission, the uncertainty for the intensity Iin Eq. �1� is calculated with the appropriate input param-eters. The uncertainty in the target radiant intensity inputparameter W in Eq. �2� is estimated by calculating this pa-rameter with the information from the TMA method. Hereit is determined from the TMA process to be1460 to 1650 W/sr. This estimate of the parameter W iscalculated with Eq. �2�, where I�=11.31 W/m2� isderived from the available images, an estimate ofR�=7.9 to 8.4 km� is available from the TMA �stationary�process, and � �=0.49� is available from the IRBLEMmodel. An estimate of the parameter W should be knownfor all available targets.
The total uncertainty in the range �R in Eq. �2� is cal-culated for the method APM as follows:
�R = �����2 + ��W�2 + ��I�2�1/2, �23�
where ��, �W, and �I are the impacts in the range uncer-tainty of all the uncertainties in the range estimate due totransmission, intrinsic target radiation, and image intensity,respectively. After calculation of the derivatives of rangeEq. �2�, we get:
�R
R= ����
2��2
+ ��W
2W�2
+ ��I
2I�2�1/2
. �24�
This means that the total uncertainty in the APM method is
dependent on the chosen uncertainty percentages in the i
Optical Engineering 026402-7
Downloaded from SPIE Digital Library on 21 Jan 2010
ransmission, target radiant intensity, and the intensity. Weefine the confidence level �CL� of the method APM as 1inus the uncertainty in range of the method divided by the
ange:
L = 1 −�R
R. �25�
.2.2 ASncertainty in the AS method is incorporated by adding anncertainty percentage to all input parameters in the rangef Eq. �5�. The approach of the uncertainty calculations, asell as for the CL calculations, is the same as for the APMethod. The total uncertainty in the range �R in Eq. �5� for
he AS method and the calculation of the derivatives give:
�R
R= ���A
A�2
+ ��dA
dA�2
+ ��dR
dR�2�1/2
. �26�
.2.3 TMA for stationary targetsncertainty in the TMA method is incorporated by adding
n uncertainty percentage to the input parameters in theange of Eq. �19�. After applying analogous equations �e.g.,qs. �20�–�23�� for the TMA method, the uncertainty in the
ange estimated by the method TMA is given by
�R
R= ��−
�B1
1/2 sin�2B1 − 2B2��2
+ � �B2�− 2�cos�B2 − 2B1� − cos B2���2 sin B2 − sin�3B2 − 2B1� + sin�B2 − 2B1���
2
+ ��vv�2
+ ����t��t
�2�1/2
. �27�
ased on the preceding uncertainty analysis for the differ-nt methods the confidence levels of the methods are estab-ished �according to Eq. �25��.
.3 Methods Used for Fusionhe decision fusion methods “best,” “weighted average,”nd “majority voting” were chosen to use in the CWT sce-ario. This choice was made because these “simple” meth-ds already give satisfying results.
.3.1 Best methodhe best method is probably the simplest form of fusion.he best method just selects the best distance estimationethod available at that time. This method is based on the
est �highest� CL per time step. No mapping function issed. “Selection” is the fusion function used.
.3.2 Weighted average methodhis method is stated as follows:
combined = uAPMRAPM + uASRAS + uTMARTMA, �28�
here ui is the adjusted mapping functions �ui
CLi /CLi�, the average CLi �on the history� per method
s confidence levels, and Ri is the distance estimate for a
February 2006/Vol. 45�2�
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
4Dissrkia
swpns�itp
rprs
4Ft
ate vec
de Visser et al.: Passive ranging using an infrared search and track sensor
specific method �index i�. Its fusion function is “summa-tion.”
3.3.3 Majority voting method
Voting fusion is described by thresholds. A vote is given ifthe method has a confidence level that is higher than athreshold. The votes are summed and a second thresholdselects between one out of three �“or” voting�, 2 out of 3�majority voting�, or 3 out of 3 �“and” voting� votes. Thevoting fusion function and the mapping are given by
F�CLAPM,CLAS,CLTMA� = threshold�CLAPM,u1�
+ threshold�CLAS,u2�
+ threshold�CLTMA,u3� , �29�
where CLAPM, CLAS, and CLTMA are the confidence levels,and u1, u2, and u3 are the mapping parameters. Majorityvoting is implemented by the rule: the best �based on CL� 2methods out of 3 and adjust the mapping function for the 2selected methods to the CL analogue to the adjusted map-ping function in Eq. �28�.
4 Experimental ResultsFor IRST, two distinct relevant operational scenarios exist.These two, the cold and the warm water environments, de-termine the main limitations of the system. We have chosen
Fig. 6 Surface geometry of an example of a WWtarget reference data, and green as the TMA st
to perform experiments in these two scenarios. b
Optical Engineering 026402-8
Downloaded from SPIE Digital Library on 21 Jan 2010
.1 WWTsuring the preliminary WWTs, the objective was to get
nsight into the passive ranging performance of the IRSTensor, since this was the first opportunity for the IRSTensor to be placed on a ship. The dynamical TMA algo-ithm used during the WWTs is described in Sec. 2.2.3. Thenowledge of the absolute own sensor position on a mov-ng platform at any time enabled the use of the full TMAlgorithm.
On September 16, 2003, about 18 runs with typical pas-ive ranging scenarios at different ranges to a target shipere performed. Typical scenarios were simulated as trialreparation, with added Gaussian bearing and elevationoise with a standard deviation of 1 mrad. All results pre-ented are nonconstrained 6 degrees of freedom solutions3-D position and velocity�. This means that no a priorinformation was added to the TMA filter. Figure 6 illus-rates a typical TMA simulation scenario used as trialreparation.
As explained in Sec. 2.2.3, TMA for dynamic targetsequires a nonconstant course of the own ship. As trailreparation, a snake pattern was simulated. The TMA algo-ithm performed as expected, and the state vector solutiontabilized at the simulated target.
.1.1 Results of the WWT scenarioigures 7 and 8 show the result of a 60-deg own ship pat-
ern, starting at a range of 11.5 km.The own ship position in Fig. 7 is determined with glo-
ario with blue as the own ship track, red as thetor.
T scen
al positioning satellite �GPS� data, and the reference target
February 2006/Vol. 45�2�
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
de Visser et al.: Passive ranging using an infrared search and track sensor
Fig. 7 Estimation of the state vector of a surface target; results of 60-deg own ships pattern, startingat a range of 11.5 km.
Fig. 8 Range estimation compared to reference range of the surface target; result of 60-deg own
ships pattern, starting at a range of 11.5 km.
Optical Engineering February 2006/Vol. 45�2�026402-9
Downloaded from SPIE Digital Library on 21 Jan 2010 to 131.180.130.114. Terms of Use: http://spiedl.org/terms
pac
4Tra
Ap2tsEEps
Fa
F
de Visser et al.: Passive ranging using an infrared search and track sensor
track is determined with the calibrated radar system of theown ship. Figure 8 explicitly shows the range solution ofthe state vector solution. Stabilization of the range solutionis acquired after 100 s.
4.1.2 Analysis of the resultsTypical range accuracies of 10% were observed after run-inof the batch filters. Run-in is reached a few seconds afterthe first own ship maneuver. However, due to the requiredown ship maneuver, TMA for dynamic targets gives a gooddistance solution after 100 s. The speed and course solu-tions look less stable, but improvement is possible whenusing a separate low-pass filter for velocity. Several param-eters contribute to an accurate range solution. The short-time stability of the own ship position �i.e., 3-D sensorposition� has a large influence on the range solution. Smallerrors in the IRST sensor position result in larger rangingerrors. The stability of the extended target track is impor-tant; a change in aspect angle with respect to the target canchange the targets center of gravity, leading to biased ob-servations. At short ranges, this effect is stronger than atlong ranges.
4.2 CWTsThis subsection focuses on the scenario for the passiveranging test and the results during CWTs. The methods areimplemented and applied to the CWT scenario data, and theresults are evaluated with the actual ground truth distancesmeasured with a radar system. The dynamical TMA methodused in the WWT scenario performed well. For the CWTwe used the stationary TMA version. The reason for choos-ing a less optimal variant of the TMA was the long timerequired �over 100 s� for the “TMA dynamically variant” tocome to a stable distance solution �see Fig. 8�. Furthermore,the stationary location of the site and presence of head-ontargets support this choice.
To apply stationary TMA, a simple straight-moving ownship is assumed. The input of the target information isbased on the trials with the preproduction model of thedual-band IRST sensor at the LBTS location in Den Helder�NL�. The runs were recorded in January 2003. The basicscenario is put in Cartesian coordinates. The basic 2-D sce-nario is shown in Fig. 9. The airplane began its inbound runat about 25 km. Bearing information for the simulated shipis updated based on the true scenario data including groundtruth range recordings and triangulation calculations:
new � bearing Þcos Bnew
sin Bnew=
y�t � new� − v�t
x�t � new�, �30�
where Bnew is the new bearing, x�t �new� and y�t �new� arethe x and y coordinates at the next time step, v is the speedof the ship, and �t is the difference between time t andt �new.
The airplane trajectory is put in the basic scenario usingradar system recordings. The preceding basic scenarioserves well as a reference model. We use the bearing infor-mation and simulated bearing information from the basicscenario as input for the TMA distance estimation method
and we test the TMA method on the same scenario. This is e
Optical Engineering 026402-1
Downloaded from SPIE Digital Library on 21 Jan 2010
ossible because the true range is removed from Eq. �19�nd has no further impact on the bearing. Furthermore, un-ertainty is added to the bearing.
.2.1 Results of the CWTs scenariohe scenario just described is used to test the three passive
anging methods and the combined fusion methods. Resultsre displayed from the time stamp t=150.
PM. Results of the distance estimation APM method areresented in Fig. 10. The APM method is explained in Sec..2.1. The distance estimate is obtained by combining theransmission estimates from IRBLEM, the calculated inten-ity from Eq. �1�, and the estimated intrinsic radiation inq. �2�. The distance uncertainty is calculated according toq. �23�. The figure shows the distance and errors com-ared to the radar reference data. From the figure, we ob-erve a good comparison between APM range estimates
ig. 9 Basic 2-D model with simulated own ship �green points� andirplane �red points� recordings at LBTS.
ig. 10 APM distance and accuracy estimate compared to the ref-
rence data.
February 2006/Vol. 45�2�0
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
mtr
Fccft0teh
mm
e
Fe
T
1
1
1
1
1
1
2
2
a
de Visser et al.: Passive ranging using an infrared search and track sensor
and the reference data, with some larger deviations at theshort-range side of the graph and hence APM performs welland is stable.
AS. The AS of the target is obtained by counting all pixelsabove the peak value of the maximum of the backgroundnoise:
pixelstarget maxbackground,
where pixelstarget are the pixels belonging to the target andmaxbackground is the maximum pixel value per image of thebackground. This rule is chosen because of stability rea-sons; fluctuations in the background noise are eliminated bythis rule. The results are displayed for the LW band, as thisis the dominating IR band in the cold scenario. Results aredepicted in Fig. 11. The figure shows the distance en errorscompared to the radar reference data. The AS method isexplained in Sec. 2.2.2. The �change in� target area is re-trieved form the available images. The changes in distanceare available from the TMA process. Equation �5� thengives the distance estimate. Uncertainty calculations areperformed analogously to those for the APM method. Fromthe figure we conclude that the performance of this methodfor passive ranging at these ranges is not good.
TMA. Figure 12 shows the TMA results. The figure showsthe distance and errors compared to the radar referencedata. TMA for stationary targets is explained in Sec. 2.2.3.Bearing 1 is retrieved from the available information. Bear-ing 2 is calculated from the basic scenario in Sec. 4.1. Thespeed of the own ship is known. The difference in time isthe time between the successive time steps. TMA distanceestimates are calculated according to Eq. �19�. Uncertaintycalculations are performed analogously to those for the
Fig. 11 AS distance and accuracy estimate compared to the refer-ence data.
APM method. On average, the performance of the TMA
Optical Engineering 026402-1
Downloaded from SPIE Digital Library on 21 Jan 2010
ethod is fair. For the first measurement at t=164, the es-imate is good but the uncertainty is very high. At a closerange the performance degrades.
usion methods. Before applying the fusion methods, theonfidence levels are calculated according to Eq. �25�. Theonfidence levels per time step, which are input for theusion process, for each method are given in Table 3. Theable shows that the TMA confidence levels are on average.04 higher than those for the APM �discarding the firstime stamp�. The AS confidence levels are always the low-st. Only at time stamps 164 and 193 is the confidence leveligher for the APM method.
With the confidence levels from Table 3 and the fusionethods from Sec. 3.3, the distance estimations per fusionethod are calculated.To compare the performance of the separate distance
stimation methods with the fusion methods the errors in
ig. 12 TMA distance and accuracy estimate compared to the ref-rence data.
Table 3 Confidence levels per time step in CWTs.
ime �s� CLAPM CLAS CLTMA
64 0.79a 0 0.50
79 0.85 0 0.97
84 0.86 0.43 0.90
89 0.86 0.67 0.90
93 0.87a 0.71 0.87
98 0.88 0.79 0.90
03 0.88 0.79 0.90
08 0.89 0.83 0.90
A CL value is higher than TMA CL value at that range.
February 2006/Vol. 45�2�1
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
�
wuopTemie
4Ipp
fptsuEe
mrirw
cmtocitw�fti
ccctnmsfnnt
pp
de Visser et al.: Passive ranging using an infrared search and track sensor
distance per method compared to the ground truth werecalculated and are shown in Fig. 13. In Fig. 13 we see thatat a distance of 8 km, the error is within an accuracy ofabout 5% for the fusion method “majority voting” and theAPM method. Results for closer ranges of the target are notwithin this accuracy.
The mean and standard deviation of these errors of thethree separate methods and three fusion methods comparedto ground truth are given in Table 4. The mean error ��error�is calculated according to
�error =1
N�t=1
N
�Rpredicted�t� − Rground�truth�t�� , �31�
where N is the number of measurements, Rpredicted is thepredicted range, and Rground�truth is the ground truth range attime t. The standard deviation of the error ��error� is alsogiven, which is the standard deviation of the mean error.Furthermore, the mean of the uncertainty �� ��R� permethod, with �R described in Sec. 3.2, is given in Table 4according to
Fig. 13 Comparison of ground truth and the range estimates of themethods.
Table 4 Mean, standard deviation of the error, and the mean uncer-tainty comparison between the methods in CWTs.
Method Error APM AS TMA Best Average Maj. Voting
�error 0.96 4.84 1.29 1.25 1.34 0.57
�error 1.11 4.91 1.13 1.12 1.15 0.47
���R 2.24 4.09 1.68 1.36 2.46 1.90
�
Optical Engineering 026402-1
Downloaded from SPIE Digital Library on 21 Jan 2010
� �R =1
N�t=1
N
�R�t� , �32�
here N is the number of measurements, and �R is thencertainty at time t. From Table 4 we note that the averagef our a priori estimate for the distance uncertainties com-ares well with the measured errors for the methods AS,MA, and Best �0.8 to 1.3 times�. However, our a prioristimate apparently was higher than the measured errors forethods APM �2.3 times�, average �1.8�, and majority vot-
ng �3.3�. The majority voting method has the lowest meanrror, 41% less than the APM method.
.2.2 Analysis of the resultsn this section we analyze the results of the CWT by com-aring the results of the different passive ranging ap-roaches.
The APM method performs well and is stable. The per-ormance degradation at closer distances is caused by “clip-ing” �reaching the maximum value in the digital image� ofhe peak value of the intensity of the target. The main rea-on for the large uncertainty range in this method is thencertainty in the estimated target radiant intensity W inq. �2�. Based on Table 4 the APM method has the bestvaluated error of the three methods APM, AS, and TMA.
The AS method performs poorly in these ranges. Thisethod is very sensitive to initial input parameters. The
eason for the poor results of this method is the fluctuationn the values of the target intensity in the images, whichesults in poor target size estimates. Results should improvehen higher resolution images are available.The TMA method gives fair results, and with lower un-
ertainty than the other separate methods. The describedethod for TMA does not work for a so-called weaving
arget, which was the case during the beginning and the endf the track trial. When a target is at a steady course and isoming inbound to the ship, the method gives a distancendication. Furthermore, the time between distance estima-ions is chosen arbitrarily here. This gives more uncertaintyhen the time between two distance estimations is small
for example, for the first measurement at t=164�. The per-ormance degradation at closer ranges is due to the sensi-ivity for the errors in the model �at that moment, the targets no longer stationary�.
The establishment of confidence levels from range un-ertainties seems good. When the target approaches, theonfidence levels are higher. This is correct because atloser ranges more information is gathered and the uncer-ainty in the distance estimate is smaller. There is, however,o influence of the �non-� correctness of the distance esti-ation on the confidence level. This requires further re-
earch. Furthermore, as we have seen that the uncertaintyor the APM method was substantially larger than the fi-ally recorded errors, fusion approaches that are now domi-ated by the TMA, may be improved when the uncertain-ies can be decreased in the other methods.
The fusion method of majority voting shows an im-rovement in the distance estimation error of 41% in com-arison �see Table 4� with the best nonfused method
APM�. Because only one scenario is tested, this result can-
February 2006/Vol. 45�2�2
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
mrpmsvft
dSsdft“nwwoar
roFimo
6WwwIswta
sspsfsdAqmsgpI
ATit
de Visser et al.: Passive ranging using an infrared search and track sensor
not necessarily be extended to more scenarios. Theweighted average fusion method performs worse than APMand TMA because of the influence of the AS method. Thebest method performs worse than APM because of the in-fluence of TMA and its high confidence levels.
5 DiscussionThe APM method performs well and is stable. Improve-ment of the estimation at closer distances would be ob-tained by extrapolating the clipped values of the peak in-tensity of the target to imaginary higher values. Theuncertainty in the distance estimation is reduced if the un-certainty for the input parameter W �target radiant intensity�can be reduced. The fact that the observed error is substan-tially lower than the a priori uncertainty gives some infor-mation that this method may require more optimization. Adisadvantage of this method is that the transmission in thedirection of the target and the target signature are requiredinput data. Besides that, background intensity and standarddeviation of the IRST image must also be calculated atevery time step. Calculation of the TTG is possible withthis method. This is used for priority scaling of targets inthe IRST system.
Typical antisurface warfare scenarios were performedduring WWTs. Bearing and elevation resolution of the im-aging system has a large influence on the performance ofthe TMA algorithm. This is, however, a cost driver and, inmost cases, a given property. Future investigation shouldalso include how other types of information, as discussed inthis paper, could be fused in the TMA algorithm. In thisway, more stressing anti-air warfare scenarios could alsobenefit from passive ranging techniques. The poor perfor-mance of the AS method might well be improved by usingmore sensitive IR equipment with higher spatial resolution.Additionally, image analysis tools, such as, for instance,dynamic superresolution, are essential to apply this tech-nique with good range performance and low uncertainties.
The TMA stationary method should perform better whenit is applied with the use of track filters, as is shown duringthe WWTs. Especially, predictions from a filter for posi-tions of the target in the near future are useful for the dis-tance estimation. During the CWT test scenario the updatedx and y positions of the target are calculated with the ref-erence distance to the target. This was done to demonstratethe principle of the TMA stationary method. A simulatedscenario would give the same information for the TMAstationary variant. The live test scenario mainly added extrainformation for the possible fusion of the APM and ASdistance estimation methods. Furthermore, tests with thepreproduction model and live bearing data placed aboard anADCF have not yet been performed but are a subject forfurther research.
Little information is found in literature about fusion ofpassive ranging methods. Our investigation to combine dif-ferent passive ranging methods looks promising. The fusionmethod majority voting shows a better performance thanthe separate APM and TMA methods. A better performancecan be achieved by enhancing the confidence levels. Theinfluence of the error for each method compared to groundtruth can improve this confidence level. The uncertainty inthe methods �and confidence levels� is probably reduced
when correlation between the input parameters is imple- p
Optical Engineering 026402-1
Downloaded from SPIE Digital Library on 21 Jan 2010
ented in the overall uncertainty analysis. The input pa-ameter sensitivity analysis performed here assumed inde-endent input parameters. Additionally, the APM and ASethods require ranging guidance from some external
ource for the initial positioning. In our case, the TMAalue was used. This also shows that a passive rangingusion system is required, and that a priori uncertainty es-imates may be correlated between the various methods.
Future work should focus on the method “intensity-istance diagrams,” which are only briefly mentioned inec. 2. Using databases with targets and their IR transmis-ions in the MWIR and LWIR bands, will optimize theual-band opportunities of the sensor in question. Practicaluture work is possible when the sensor is placed aboardhe ADCF, the implementation of a software module fordistance indications” provides improved situational aware-ess. An extension for the distance estimation is possiblehen the bearing information of the IRST sensor is fusedith the bearing information from the trainable electro-ptical observation system �TEOOS� that is already placedboard the ADCF �TEOOS is also equipped with a laserange finder�.
Passive ranging is an elegant method to acquire a 3-Decognized air and surface picture of the near surroundingsf a naval ship without giving away the own ship position.urthermore, imaging sensor systems give additional visual
nformation to the operator in the command center, infor-ation that is more important today during brown water
perations.
Conclusionse discussed three separate methods for passive rangingith an IRST sensor and two practical implementationsith real data. The TMA algorithm in combination with the
RST system, used during WWTs, proved to give good re-ults in real antisurface warfare scenarios. Accuracies of 10ere obtained. For anti-air warfare scenarios, where reac-
ion time is a larger issue, other types of information fusionre still required and are a subject for further investigation.
We demonstrated passive ranging capability of an IRensor, with a certain level of accuracy. Based on the CWTcenario and the passive ranging methods atmosphericropagation, AS of the target, stationary TMA, and the fu-ion of these methods, we demonstrated that fusion of in-ormation improves the distance estimation for a realisticcenario. The fusion method majority voting improves theistance estimation with 41% compared to the best separatePM method. This best separate method already performsuite well and is stable. An accuracy of the distance esti-ation of 5% at 8 km was reached for the specific test
cenario. The results of the AS of the target method are ineneral not good. A recommendation for future work is toerform tests at an Air Defense Command Frigate when theRST sensor is placed aboard the ship.
cknowledgmentshe RNLN allowed the main author to spend time conduct-
ng this research besides his normal duties. The manufac-urer of the IRST sensor, Thales Naval Netherlands, has
rovided the reference data.
February 2006/Vol. 45�2�3
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
MarDoaoie
etct
aa
prtc
de Visser et al.: Passive ranging using an infrared search and track sensor
References
1. R. Knepper, “Sirius, a long range infra red search and track system,”in Infrared Technology and Applications, Proc. SPIE 3061, 578–584�1997�.
2. S. S. Blackman and R. Popoli, “Passive sensor tracking methods,”and “Multiple sensor tracking,” Chaps. 5 and 10 in Design andAnalysis of Modern Tracking Systems, pp. 259–324, 661–736, ArtechHouse, Norwood, MA �1999�.
3. R. van Sweeden, H. M. A. Schleijpen, and P. B. W. Schwering, “Pas-sive ranging techniques,” EUCLID RTP 8.2, START WE.256.4F, TheHague, The Netherlands �1996�.
4. J. F. de Groot, “Tracking aspects of SIRIUS �additional� functional-ity,” Internal Document, Thales Naval Netherlands, Hengelo, TheNetherlands �1996�.
5. R. A. W. Kemp, “Point target detection, tracking and passive ranging:a performance analysis study,” Chap. 19, TNO report FEL-99-S230,The Hague, The Netherlands �1999�.
6. P. B. W. Schwering, A. N. de Jong, and J. Winkel, “EindrapportSIRPS �Staring Infrared Panoramic Sensor�,” TNO report FEL-01-A309 �in Dutch�, The Hague, The Netherlands �2004�.
7. T. Aytaç and B. Barshan, “Rule-based target differentiation and posi-tion estimation based on infrared intensity measurements,” Opt. Eng.42�6�, 1766–1771 �2003�.
8. O. Sasaki, K. Sakata, and T. Suzuki, “Optical method for detectingdisplacement of a car in stereo images,” Opt. Eng. 42�7�, 2092–2095�2003�.
9. S. Ono and S. Komatsu, “Simultaneous image restoration and dis-tance estimation of an object from a strongly defocused photon im-age,” Opt. Eng. 42�4�, 1024–1028 �2003�.
10. Ruters and Michalowicz, “Passive location and ranging,” in TheIR&EO Systems Handbook, Vol. 5, S. B. Campana, Ed., Chap. 4.5,pp. 336–341 �1993�.
11. A. N. de Jong and H. Winkel, “Enhanced IR point target detection byatmospheric effects,” in Infrared Technology and Applications XX-VIII, Proc. SPIE 4820, 885–896 �2002�.
12. A. N. de Jong and H. Winkel, “Report on the Baltic 99 IR experi-ments,” TNO report FEL-00-A094, The Hague, The Netherlands�2000�.
13. A. N. de Jong, J. Winkel, M. M. Moerman, and R. de Rooy, “Pre-liminary report on the TG16 POLLEX trial,” TNO-report FEL-01-A222, The Hague, The Netherlands �Sep. 2001�.
14. A. N. de Jong, H. Winkel, M. Moerman, K. Stein, K. Weis-Wrana, L.Forand, G. Potvin, J. Buss, A. Cini, H. Vogel, and A. Stark, “TG16Point target detection experiment POLLEX,” in Infrared Technologyand applications XXVIII, Proc. SPIE 4820, 849–860 �2002�.
15. D. Dion, L. Gardenal, L. Forand, M. Duffy, G. Potvin, and S. Daigle,“IR boundary layer effects model �IRBLEM�—version 4.1,” DRDC-Valcartier, Canada �Dec. 2002�.
16. S. S. Blackman and R. Popoli, “Infrared search and track systems,”Sec. 2.3 in Design and Analysis of Modern Tracking Systems, pp.85–119, Artech House, Norwood, MA �1999�.
17. S. S. Blackman and R. Popoli, “Passive sensor tracking methods,”Chap. 5 in Design and Analysis of Modern Tracking Systems, pp.268–271, Artech House, Norwood, MA �1999�.
18. U.S. Navy, Navy Combat Manuals, “Operations specialist, volume1,” Chap. 10, pp. 10–23, U.S. Navy �1999�.
19. F. Cremer, K. Schutte, J. G. M. Schavemaker, and E. den Breejen, “Acomparison of decision-level sensor-fusion methods for anti-personnel landmine detection,” Inf. Fusion 2, 187–208 �2001�.
20. J. G. M. Schavemaker, E. den Breejen, F. Cremer, K. Schutte, and K.W. Benoist, “Depth fusion for anti-personnel landmine detection,” inDetection and Remediation Technologies for Mines and Minelike Tar-get VI, Proc. SPIE 4394, 1071–1081 �2001�.
Optical Engineering 026402-1
Downloaded from SPIE Digital Library on 21 Jan 2010
aarten de Visser attended the Royal Netherlands Naval Academynd graduated in 1999 in situational awareness and refinement. Heeceived his MSc degree from the electrical engineering faculty ofelft University of Technology in 2004. His research was the subjectf this paper. After several years serving as deputy weapons officerboard different types of frigates, he is currently working in the fieldf project engineering for the Royal Netherlands Navy �RNLN�. His
nterests are in the practical applications and implementations ofnhancing situational awareness aboard frigates.
Piet B. W. Schwering received his MSc de-gree in astronomy in 1983 and his PhD de-gree in mathematical sciences in 1988, bothfrom the University of Leiden, where from1983 to 1987 he was working in the field ofastrophysics, and he studied dust proper-ties and star formation in the Magellanicclouds using Infrared Astronomical Satellite�IRAS� data at IR wavelengths. Since 1987he has been a scientist and project man-ager in the Electro-Optics Group, the Neth-
rlands Organization for Applied Scientific Research �TNO�. His in-erests are the detection of objects in IR surveillance, theharacteristics and reduction of background clutter, and sensor sys-em evaluation.
Johannes F. de Groot received his MScdegree from the electrical engineering fac-ulty of Delft University of Technology in1982 and joined Thales Hengelo, the formerHollandse Signaalapparaten. His main in-volvement was in the development andevaluation of some generations of long-range infrared search and track systems�LRIRST�. His main interest is signal anddata processing, with a special interest inpassive ranging by means of target motion
nalysis �TMA�. He is currently working on algorithm developmentnd prototyping of next-generation staring IRST systems.
Emile A. Hendriks received his MSc andPhD degrees, both in physics, from the Uni-versity of Utrecht in 1983 and 1987, respec-tively, and he became an assistant profes-sor with the electrical engineering faculty ofthe Delft University of Technology. In 1994he became a staff member of the Informa-tion and Communication Theory of this fac-ulty, and since 1997 he has headed thecomputer vision section of this group as anassociate professor. His interests are com-
uter vision, low-level image processing, image segmentation, ste-eoscopic and 3-D imaging, motion and disparity estimation, struc-ure from motion/disparity/silhouette, and real-time algorithms foromputer vision applications.
February 2006/Vol. 45�2�4
to 131.180.130.114. Terms of Use: http://spiedl.org/terms
