International Journal of Electrical & Computer Sciences IJECSIJENS Vol:16 No:03
Convergence Analysis of Moving Object Tracking
Algorithm in 3D using modified EKF based on Bearing
and Elevation Measurements for Underwater
Environment
Nagamani Modalavalasa1, G. Sasi Bhushana Rao2 and K. Satya Prasad3
1
Dept.of ECE, SBTET, Andhra Pradesh, India, , e_mail: mani.modalavalasa@gmail.com
2
Dept. of ECE, Andhra University, Visakhapatnam, Andhra Pradesh, India
3
Dept.of ECE, Jawaharlal Nehru Technological University Kakinada, Kakinada, India
Abstract Underwater surveillance is the primary challenge
being faced by the researchers, scientists and marine community,
which include surveillance of harbour, detection of mines,
inspecting pipelines and mapping of ocean bottoms etc. The
underwater surveillance is best undertaken by the Autonomous
Underwater Vehicle (AUV) as it avoids the involvement of
human. It is important to note that the convergence time plays
very important role in aiding the AUV for its safe navigation as
the AUV requires the sufficient time to maneuvering itself for
avoiding the objects that comes into its path. Towards this
various methods / techniques such as Least Squares (LS),
Kalman Filter (KF) and Extended Kalman Filter (EKF) methods
have been explored, however all these methods have their own
drawbacks. In this paper a new method has been developed
wherein tracking algorithm using EKF has been extended to the
Bearing and Elevation only Tracking (BEOT) method. The
performance of this algorithm has been evaluated using Monte
Carlo Method. The convergence issues associated with the new
designed algorithm have also been analysed and it has been
observed that the errors are considerably small and settles down
after the filter learns the existing dynamics.
Index Term Underwater Vehicle, Tracking, Kalman Filter,
BEOT.
I. INTRODUCTION
Tracking underwater objects is an ongoing challenge for
safe navigation of AUV. Bearing only tracking (BOT) is used
in many marine applications such as sonar based robotic
navigation, infrared seeker based tracking and underwater
weapon systems. The bearing and elevation only filtering
problem in 3D from a single maneuvering sensor is the
counterpart of the bearing only filtering in 2D. The important
field of research in the areas of aircraft surveillance,
submarine tracking, mobile systems and autonomous robotics
applications is passive target tracking of maneuvering objects
using line of sight (LOS) angle measurements [15].
Tracking in underwater environments, where objects have
multiple degrees of freedom or when the scene conditions
cannot be controlled is much harder. The bearing and elevation
only tracking in 3D is much difficult than bearing only tracking
in 2D. This is due to the fact that the dynamic model of 3D is
not such a straight forward model as in 2D. With this sense the
robotic platform could estimate its relative velocity and
position with respect to objects within its environment. This
would be useful both for navigation and interaction with
objects in the environment. Autonomous Underwater Vehicle
is a robot, designed to perform specific tasks in underwater.
The range, bearing and elevation measurements are often noisy
due to turbid nature of underwater environment and results a
nonlinear relation between the states and measurements. Due to
these measurement inaccuracies gives a direct impact on the
performance of the tracking algorithm.
There are two approaches that are generally used in the
underwater environment. First approach is the Linear Kalman
filter (KF) [6], which was designed by R.E. Kalman in 1960. In
this approach measurements are linear and designed for
prediction and estimation problem. It can be defined as an
optimal recursive data processing algorithm and
is
characterized by accurate estimation of state variables under
noisy condition. It is acceptable for robotic manipulators,
drives and other industrial applications. The algorithm is
developed in two steps which involve prediction and updating.
The second approach is Extended Kalman Filter (EKF) [7] and
measurements are nonlinear. It is well organized fact that in
EKF the initial covariance is based on the initial converted
measurement and the gain is based on the accuracies of the
subsequent linearization and therefore the overall performance
depends on these accuracies.
Discretetime EKF with 2D Cartesian coordinates is used in
the BOT problem in initial research [8]. The filter which was
implemented for bearing only tracking is based on the Nearly
15720316039292IJECSIJENS June 2016 IJENS
IJENS
International Journal of Electrical & Computer Sciences IJECSIJENS Vol:16 No:03
Constant Velocity Model (NCVM) and nonlinear measurement
model[9]. For 2D tracking problems, Extended Kalman Filter
(EKF), Unscented Kalman Filter (UKF), Cubature Kalman
Filter (CKF) and GaussHermite Kalman Filter (GHKF) are
implemented [10 and 11] .
EKF is implemented for both the predicted state estimate
and covariance using a discretized linear approximation[12].
The approaches which are mentioned earlier uses a 2D state
estimation problem. The performance of the EKF, UKF and
Particle Filter (PF) are compared for the angleonly filtering
problem in 3D using bearing and elevation measurements from
a single maneuvering sensor[13]. Estimation of
the
kinematics, such as position and velocity of a target, using
noisycorrupted measurements of the target from a single
observer is a nonlinear function.
In the primary research, the EKF algorithm results
uncertain performances, including poor track accuracy and
divergences [7,8]. A new method has been developed in this
research wherein tracking algorithm using EKF has been
extended to the Bearing and Elevation only Tracking (BEOT)
method. In this approach a single maneuvering sensor/observer
case was examined. Also BEOT problem was analyzed with
good accuracy and efficiency as the inaccuracies can be tackled
effectively in this method.
The target tracking basics is covered in [14]. Most aspects
of tracking covered in [15] . Comparison of different tracking
methods presents in [16][17] derives a tracking filter that is
well suited for angleonly target tracking. The approach that
has been followed in this paper is that the non linearities are
modified prior subjecting to the tracking algorithm with angles
only measurements. Modified gains for the bearings and
elevation problem have also been derived in a simplistic
manner. The results have been promising and have shown
improved performance over the conventional EKF.
Song T.L. and Speyer LL derived a modified gain extended
kalman filter (MGEKF) for nonlinear estimation problems[18].
This MGEKF algorithm was further developed by Galkowiski
P.J. and Islam M.A. based on the pseudo measurements[19].
Mo, Longbin, Liu, Qi, Zhou, Yiyu, Sun, Zhongkang presented
the improved modified gain functions for 3D anglesonly
tracking[20][21]. In this paper the nonlinearities are modified
and then applied to a tracker with bearing and elevation only
measurements. It shows an improved performance over the
conventional Extended Kalman Filter (EKF). In this paper,
modified gains for the bearings and elevations problem have
been derived in a simpler manner .
coordinates
, and
and the velocity towards those
coordinate axes ,
and . Thus, the dynamics of the
target is modeled as a state space model. The state of the
target is defined in the tracker coordinate frame (T frame) for
which the x, y , and z axes are along the local North, East and
upward directions, respectively as shown in Fig.1. The target
and sensor/observer states in Cartesian coordinates are defined
as follows
II. TRACKING ALGORITHM
The primary problem in bearing and elevation only tracking is
to estimate the trajectory of the object/target from noisy
corrupted sensor data[22]. In this scenario the observer tracks
a moving target/object with sensor(sonar), which measures
only the bearings and elevations of the target with respect to
positions of the sensor. There is one moving target/object in
the scene and one sensor(sonar) for tracking it. The state of
the Target Motion Model (TMM) is described by a Nearly
Constant Velocity Model (NCVM) and at time step (k+1)
consists of the position in three dimensional (3D) Cartesian
Where, the state vector ( ) consists of the position and
velocity components of the target moving in the 3D space.
(1)
And
(2)
The relative state vector in the T frame is defined by
(3)
Let the relative state vector in the T frame is
(4)
Then,
, etc.
The range vector of the target from the observer ( or sensor) in
the T frame is,
=
(5)
Then the range is defined as
=
(6)
The range vector can be expressed in terms of range,
bearing ( ) and elevation ( ), as defined in Figure 1, by
=
(7)
Target Motion Model (TMM) is described in the Cartesian
coordinate system by linear discretetime difference equation
with some additive noise as
(
i.e.
( )
( )
( )
( )
( )
( )
( )
(8)
( )
( ) ( )
(9)
Where ( ) and
( ) are the state transition matrix and
integrated process noise, respectively, for the time interval
(
) ( ) and process noise is assumed to be zero
mean white gaussian noise.
15720316039292IJECSIJENS June 2016 IJENS
IJENS
International Journal of Electrical & Computer Sciences IJECSIJENS Vol:16 No:03
(
( )
( )
[
(10)
given in Eq. (17). The updated state covariance matrix at time
(k+1) is a function of modified gain g and is shown in
Eq.(19).
(11)
x(k 1) F ( k 1) x ( k )
(15)
P(k 1) F ( k 1) P ( k ) F(Tk 1) Q( k 1)
(16)
x( k 1) x(k 1) K ( k 1)( z( k 1) H ( k 1) x(k 1) )
(17)
The target is tracked by a Sonar in the underwater and
provides measurements of bearing ( m) and elevation ( m).
The measurement model is given as
(
(
(
( )
(
(
)
]
)
(12)
)
)
)
(
(
)
10
K ( k 1)
P(k 1) H (Tk 1)
(18)
( H ( k 1) P(k 1) H (Tk 1) R)
P( k 1) ( I K ( k 1) g ( z( k 1) , x(k 1) ) P(k 1)
( I K ( k 1) g ( z( k 1) , x(k 1) )T K ( k 1) R K (Tk 1)
(13)
Where ( ) is uncorrelated, zeromean white Gaussian noise
with variances
,
in the bearing ( ) and elevation
( ) measurements respectively .
The measured range, bearing, and elevation from sonar are
converted to target positions in Cartesian coordinates with
respect to own ship as origin using the following relations:
(19)
Where ,
x (k ) = State estimate at time k.
F ( k 1) = State transition matrix at time k+1.
x(k 1) = Predicted state estimate at time k+1.
P (k ) = State covariance matrix estimate at time k.
( )
P(k 1) = Predicted state covariance matrix at k+1 time.
( )
Q( k 1) = Process noise covariance matrix at k+1 time.
( )
x( k 1) = Updated state estimate at time k+1 .
( )
( ) ( )
(14)
Where,
is the range vector of x, y and z positions.
Because the measurement model is nonlinear we replace the
Kalman filter algorithm with EKF. The dynamic model using
NCVM in 3D is linear and the measurement model for bearing
and elevation is nonlinear for this problem. In general EKF is
based on linearized approximations to nonlinear dynamic
and/or measurement models [9], [23] and is widely used. For
this problem, the linearized approximation is performed in the
measurement update step as described in [9], [23].
With the designed system dynamic and measurement model
given in Eq.(8) and Eq.(12), the process of estimating state
vector x (k ) with the proposed BEOT algorithm is elucidated
K ( k 1) = Filter gain at time k+1.
z( k 1) = Measurement data at time k+1.
H ( k 1) = Measurement matrix at time k+1.
P( k 1) =Updated state covariance matrix at time k+1.
R = Measurement noise covariance matrix.
g ( z( k 1) , x(k 1) ) = Modified gain function.
I = Identity matrix.
The main difference between the EKF and MGEKF is the
function g in the covariance update.
To determine the modified gain function g we write:
in Eq.(15) to Eq.(19). The algorithm estimates state vector
x(k 1) , from the initial state estimate x (k ) and its uncertainty
matrix P (k ) as given in Eq.(15) and Eq.(16). Kalman gain in
Eq.(17) is the function of uncertainty in state estimate and
measurement error , provides weight coefficients to the
measurements at time (k+1) used for state vector updation
(
(
)
) ]
g [(
(
)
)]
)
(20)
Where,
Estimated bearing from the states , and .
Estimated elevation from the states , and .
15720316039292IJECSIJENS June 2016 IJENS
IJENS
International Journal of Electrical & Computer Sciences IJECSIJENS Vol:16 No:03
Since g is not a function of target velocity we removed
those states for the derivation of g .
11
)
(28)
Equation (28) rewritten as
The measurement matrix H is given by
) =
(29)
(21)
IV. UPDATED MEASUREMENT OF ELEVATION DATA
As can be seen from the above analysis
III. UPDATED MEASUREMENT OF BEARING DATA
If the range in horizontal plane is
generates
(
) =
(30)
=
In the similar way
Then the estimated range be
with
By adding
+
+

Adding both sides with
+
)(
(
(
(22)
generates,
(31)
From the Eq. (24) we get
)(
) =
we can get,
) (
(32)
(23)
Now replacing Eq. (32) in Eq. (31),
Similarly,

2 =(
) (
) (
(
)(
]+
(  )
(25)
] = 2
)
] =2
)
) and (
(34)
(
[
(
)
) ]
) are then,
(
[(
(
)
)]
)
(35)
(27)
Now the coefficients of (
(26)
=2
)
(
= 2
(  )
Taking Eq. (25) and simplifying we get
[
) (
By rearranging the Eq. (33) becomes,
) [
(33)
(24)
) [
)=
Since the true bearing and elevation angles are not available in
practice, the measured bearing and elevations are used to
15720316039292IJECSIJENS June 2016 IJENS
IJENS
International Journal of Electrical & Computer Sciences IJECSIJENS Vol:16 No:03
compute the modified gain. Hence the Eq. (35)
rewritten as,
[
(
(
can be
)
) ]
(
[(
(
XZ plane
)
)]
)
P( ,z)
By considering the velocity components
is given by
g=
XY plane
P( z)
(36)
12
P( )
Fig. 1 Tracking Angles
]
(37)
V. RESULTS AND DISCUSSIONS
When extending the 2D estimation to 3D estimation, it is
important to redefine the parameters of target and observer.
For the validation of the proposed algorithm the simulation
has been generated with different measurement inaccuracies
with simulated data. Ownship is always considered at the
origin and also assumed that the target is moved with constant
velocity and travels in a straight path. The tracking angles of
the observer and target are shown in Fig.1.
Initial SONAR signal assumptions considered for the
simulations are,
1. Always own ship is at origin(O i.e. sonar at (0,0,0)
coordinates).
2. Target is moving at constant speed.
3. All angles (Bearing and Elevation angles) are considered
with respect to target.
For the implementation of the algorithm, the initial estimate of
target state vector is chosen as follows. As only bearing and
elevation measurements are available, it is not possible to
guess the velocity components of the target. So these
components are assumed as 10 m/s; which are close to the
realistic speeds of the vehicles in underwater environments.
The initial state has been defined as,
(38)
Where
and
are initial bearing and elevation
measurements.
Initial target velocity = 10 m/sec.
Initial target course = 135 deg.
Initial target range = 10km.
Initial target bearing = 0.5 deg.
Initial target elevation = 45 deg.
= 0.0015
= 30
= 0.0015
Where,
and
are the variances
bearing ( ) and elevation ( ) respectively.
in range (r),
The noise in the bearing is assumed to be additive in nature
and follows normal distribution. The measurement contains
noise generated by several noise sources according to central
limit theorem, i.e. the sum of noises of any density function
leads to Gaussian density function. So, it is assumed that the
noises in the measurements are of Gaussian nature. Many
random processes occurring in nature actually appear to be
normally distributed. So the noise in the bearing is assumed to
be additive in nature and follows normal distribution.
Bearing and elevation measurements are taken for every 1
second for 1000 Monte Carlo updates. Fig. 2 shows the
trajectories of ownship, true target and predicted target. Initial
covariance matrix is assumed as per the standard procedure in
[24].
15720316039292IJECSIJENS June 2016 IJENS
IJENS
International Journal of Electrical & Computer Sciences IJECSIJENS Vol:16 No:03
13
140
120
1.5
Course Error (degrees)
100
Simulated Target Trajectory
Predicted Target Trajectory
Observer Position
x 10
80
60
rz (metres)
40
1
20
0.5
100
200
0
8000
6000
300
400
500
Time(seconds)
600
700
800
900
1000
Fig. 6. Course Error Vs Time
200
150
4000
100
2000
50
0
rx (metres)
0
ry (metres)
Fig. 2. Observer, True Target and Predicted Target Trajectories
700
600
The estimated errors in simulations 1 are plotted in Fig.3 to
Fig.7. It has been observed that the convergence duration is
300 seconds in case of range, 80 seconds in case of bearing,
200 seconds in case of velocity, 800 seconds in case of course
and 800 seconds in case of elevation, which indicate the
suitability of this method for aiding the AUV for its safe
navigation.
45
400
40
300
35
200
100
100
200
300
400
500
Time (seconds)
600
700
800
900
1000
Elevation Error (degrees)
Range Error (meters)
500
30
25
20
15
10
Fig. 3. Range Error Vs Time
5
0
100
200
300
400
0.25
500
Time(seconds)
600
700
800
900
1000
Fig. 7. Elevation Error Vs Time
0.15
The performance of the algorithm is evaluated in terms of :
i. The percentage fit error (PFE) in and
0.1
0.05
(38)
( )
0
100
200
300
400
500
Time (seconds)
600
700
800
900
1000
Fig. 4. Bearing Error Vs Time
(39)
( )
(40)
( )
Speed Error (mts/sec)
Bearing Error (degrees)
0.2
ii. The root mean square position error
(41)
100
200
300
400
500
Time (seconds)
600
700
Fig. 5. Velocity error Vs Time
800
900
1000
Where, N= 1,2,31000 Monte Carlo runs.
iii. The root mean square velocity error
(42)
Where, N= 1,2,31000 Monte Carlo runs.
iv. The root sum square position error
15720316039292IJECSIJENS June 2016 IJENS
IJENS
International Journal of Electrical & Computer Sciences IJECSIJENS Vol:16 No:03
14
(43)
RSSPE
40
20
0
100
200
300
400
500
600
700
800
900
1000
100
200
300
400
500
600
700
800
900
1000
100
200
300
400
500
600
700
800
900
1000
100
200
300
400
500
time in sec.
600
700
800
900
1000
RMSPE
10
v. The root sum square velocity error
5
0
(44)
RMSVE
1
0.5
0
RSSVE
1
0
Fig. 10. RMSP, RMSV, RSSP and RSSV in predicted position
30
RSSPEx
where
and are the measurements, and are the
estimated target positions,
and are the measurements,
and are the estimated target velocities in
and
coordinates, respectively. Here RMSPE, RMSVE, RSSPE
and RSSVE are calculated at each time step as given in the
equations (41 & 44).
20
10
0
PFEx
0.4
0.2
100
200
300
400
500
600
700
800
900
1000
100
200
300
400
500
600
700
800
900
1000
PFEy
10
100
200
300
400
500
600
700
800
900
1000
100
200
300
400
500
600
700
800
900
1000
100
200
300
400
500
time in sec.
600
700
800
900
1000
RSSPEy
10
0
30
RSSPEz
The Percentage Fit errors and Mean absolute errors in x, y and
z positions of scenario1 are plotted in Fig. 8 and Fig. 9 and
found to be in acceptable ranges. The Root Mean Square
Errors in position (RMSPE) and velocity (RMSVE), Root
Sum Square Errors in position (RSSPE) and velocity
(RSSVE) are shown in Fig.10. It is observed that the errors are
small and settled down after filter learns dynamics.
Root sum square position errors in x, y and z positions are
shown in Fig.11 and observed that the errors are small.
20
10
0
Fig. 11. RSSP in x, y and z positions for simulation 1
Comparison of Proposed BEOT algorithm with
Conventional EKF:
The computational time of Extended Kalman Filter (EKF) is
generally less when compared to Unscented Kalman
Filter(UKF) and Particle Filter ( PF) [25, 26]. In real time
scenario the computational time is more important than the
accuracy of the position estimation. Because of this reason the
results of the proposed algorithm are not compared with the
UKF and PF.
PFEz
0.4
0.2
100
200
300
400
500
time in sec.
600
700
800
900
1000
Fig. 8. Percentage fit errors in x, y and z positions
In Fig. 12 to Fig. 16 the range, bearing, velocity, course and
elevation errors for proposed BEOT method is compared with
conventional EKF method for every time step of 1000 Monte
Carlo runs and found to be optimum.
It is observed that the range, bearing, velocity, course and
elevation errors are optimum for the proposed algorithm.
MAEx
0.5
700
100
200
300
400
500
600
700
800
900
Proposed method
EKF method
600
1000
500
1
0.5
0
100
200
300
400
500
600
700
800
900
1000
MAEz
0.5
Range Error(meters)
MAEy
1.5
400
300
200
100
200
300
400
500
time in sec.
600
700
800
900
Fig. 9. Mean absolute errors in x, y and z positions
1000
100
100
200
300
400
500
Time (seconds)
600
700
800
900
1000
Fig. 12. Range error of proposed BEOT algorithm and EKF
15720316039292IJECSIJENS June 2016 IJENS
IJENS
International Journal of Electrical & Computer Sciences IJECSIJENS Vol:16 No:03
Operating frequency (high)Chirping from 620 to720 kHz
(670kHz).
Optional high frequency is 1 MHz.
Maximum range is 300 m [300kHz].
Maximum range is 100 m [670kHz].
The simulated results with the assumed data and the results
obtained from the Sector Scan Sonar(SSS) data are similar.
The following figures corresponding to SSS data. Fig. 17 to
Fig.19 shows the range, bearing and elevation errors.
0.25
Proposed method
EKF method
Bearing Error(deg)
0.2
0.15
0.1
0.05
100
200
300
400
500
Time (seconds)
600
700
800
15
900
1000
180
Fig. 13. Bearing error of proposed BEOT algorithm and EKF
160
140
8
Rng Error (m)
120
7
Proposed method
EKF method
100
80
Speed Error(m/sec)
60
5
40
4
20
3
0
100
200
300
400
600
700
800
900
1000
Fig. 17. Range error vs time
500
TIME (seconds)
100
200
300
400
500
Time (seconds)
600
700
800
900
1000
0.9
Fig. 14. Velocity error of proposed BEOT algorithm and EKF
0.8
Bearing Error (degrees)
0.7
140
Proposed method
EKF method
120
Course Error(deg)
100
0.6
0.5
0.4
0.3
0.2
80
0.1
60
0
100
200
300
400
40
500
Time (seconds)
600
700
800
900
1000
700
800
900
1000
20
Fig. 18. Bearing error vs time
0
100
200
300
400
500
Time (seconds)
600
700
800
900
1000
45
40
Fig. 15. Course error of proposed BEOT algorithm and EKF
Elevation Error (degrees)
35
45
Proposed method
EKF method
40
Elevation Error(deg)
35
30
25
20
15
30
10
25
5
20
0
15
100
200
300
400
500
Time(seconds)
600
10
5
0
Fig. 19. Elevation error vs time
0
100
200
300
400
500
Time (seconds)
600
700
800
900
1000
Fig. 16. Elevation error of proposed BEOT algorithm and EKF
The same algorithm is validated practically with the sector
scan sonar that operates with the ranges of 100m and 300m.
The sonar which was used to collect the data is Super Seeking
DST (Digital Sonar Technology) Dual Frequency CHIRP
Sonar. Sector scan sonar with the following specifications:
Operating frequency (low)Chirping from 250 to 350 kHz
(300kHz).
VI. CONCLUSIONS
In this paper , simulations has been done in 3D estimation
algorithms based on the 2D estimation BOT problem. A new
Bearing and Elevation only Tracking algorithm was designed
using modified kalman
gain to evaluate the system
performance and has been developed for underwater object
detection/tracking. The experimental results also show with
1000 Monte Carlo simulations and the convergence issues
associated with the developed algorithm has been analyzed. It
15720316039292IJECSIJENS June 2016 IJENS
IJENS
International Journal of Electrical & Computer Sciences IJECSIJENS Vol:16 No:03
is found that the errors are small and settle down after the filter
learns the dynamics and will converge more quickly.
In real time underwater scenario the computational time is
more important than the accuracy of the state error rate. The
results of the proposed BEOT method are compared with the
traditional EKF algorithm and observed that range, bearing,
velocity, course and elevation errors of the target are optimum.
The same algorithm is validated practically with the sector
scan sonar that operates with the ranges of 100m and 300m.
It is also observed that the simulated results with the assumed
data and the results obtained from the Sector Scan Sonar(SSS)
data are similar.
Acknowledgements
This research paper has resulted from the NSTL (DRDO),
Visakhapatnam, Andhra Pradesh, INDIA sponsored research
project.
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
REFERENCES
Goutam Chalasani,, Shovan Bhaumik, Bearing Only Tracking
Using GaussHermite Filter IEEE, 2011 ISSBN:978145772119
S.Sadhu, S.Bhaum`
ik, and T.K.Ghoshal, Evolving
homing guidance configuration with Cramer Rao bound,
Proceedings 4th IEEE International Symposium on Signal
Processing and Information Technology, Rome, 18th21st
December 2004.
T.L.Song, and J.L.Speyer, A stochastic analysis of a modified
gain extended Kalman filter with applications to estimation with
bearing only measurements, IEEE Trans. Autom. Contmi, 1985,
AC30, No.10, October 1985,
pp. 940949.
Ronghui Zhan and Jianwei Wan, Passive maneuvering target
tracking using 3D constant turn model, Radar, 2006, IEEE
Conference.
K. Dogancay, 2005 Bearingsonly target localization using total
least squares, Signal Process, vol. 85, pp.
16951710.
R. E. Kalman, A new approach to linear filtering and prediction
problems, ASME J. Basic Eng.,1960.
Nardone, S. C., Lindgren, A. G., and Gong, K.F., Fundamental
properties and performance of conventional bearingsonly target
motion analysis. IEEE Transactions on Automatic Control, AC29,Sept. 1984 9, pp. 775787.
V. J. Aidala, Kalman filter behavior in bearingonly tracking
applications, IEEE Trans. on Aerospace and Electronic Systems,
Vol. AES15, No. 1, January 1979, pp. 2939.
Y. BarShalom, X. R. Li, and T. Kirubarajan, Estimation with
Applications to Tracking and Navigation, John Wiley & Sons,
2001.
B. L. Scala, M. Morelande, An Analysis of the Single Sensor
BearingsOnly Tracking Problem, Information fusion, 11th
international conference, 2008 .
Jouni Hartikainen, Arno Solin, and Simo Sarakka, Optimal
filtering with Kalman filters and smoothers, A manual for the
Matlab tool box, AaltoFinland, August 2011.
R. Karlsson and F. Gustafsson, Range estimation using angleonly target tracking with particle filters, Proc. American Control
Conference, 2001, pp. 3743 3748.
Mallick, M., Morelande, M.R., Mihaylova and L, Arulampalam,
S., Yan, Comparison of Angleonly Filtering Algorithms in 3D
[21]
[22]
[23]
[24]
[25]
[26]
16
using Cartesian and Modified Spherical Coordinates, 15th
International Conference on IEEE, 2012, pp. 13921399.
Y. BarShalom and X. Li. Estimation and Tracking Principles,
Techniques,and Software, Artech House, 1993.
S. Blackman and R. Popoli. Design and Analysis of Modern
Tracking Systems, Artech House, 1999.
B. Ristic, S. Arulampalam, and N. Gordon, Beyond the Kalman
Filter, Particle Filters for Tracking Application, Artech House,
2004.
V. J. Aidala and S. E. Hammel Utilization of Modified Polar
Coordinates for BearingsOnly Tracking, In IEEE Transaction on
Automatic Control, volume AC28, March 1983.
T.L.Song, and J.L.Speyer A stochastic analysis of a modified
gain extended Kalman filter with applications to estimation with
bearing only measurements, IEEE Trans. Autom. Contmi, 1985,
AC30, No.10, October 1985, pp 940949.
PI. Galkowski, and M.A. Islam, An alternative derivation of the
modified gain function of Song and Speyer, IEEE Trans. Autom
Control, 1991, Vol.36, Issue 11, pp. 13231326.
Mo, Longbin, Liu Qi, Zhou Yiyu, Sun, Zhongkang, New
modified measurement function in 3D passive tracking,
Acquisition, Tracking and Pointing XI, Proc. SPIE,
Vol. 3086, pp. 311314.
MO Longbin, Liu Qi, Zhou Yiyu and Sun Zhongkang,
Utilization of the Universal Liniarization in target tracking,
IEEE Transactions on Aerospace and Electronics conference,
1997.
K. Radhakrishnan, A. Unnikrishnan, and K.G Balakrishnan,
Bearing only Tracking of Maneuvering Targets using a Single
Coordinated Turn Model, International Journal of Computer
Applications, 2010.
A. Gelb, editor, Applied Optimal Estimation, MIT Press, 1974.
Y. T. Chan and S.W. Rudnicki, Bearings only Doppler Bearing
Tracking using Instrumental Variables, IEEE Transactions on
Aerospace and Electronic Systems, Vol.28, No.4, October 1992.
T. Fiorenzani. C.Manes, G. Oriolo and P.Peliti, Comparative
study of Unscented Kalman Filter and Extended Kalman Filter
position/attitude estimation in unmanned aerial vehicle, Itallian
National research Council ( IASICNR,), ISSN: 11283378.
Nak Young Ko and Tae Gyun Kim, Comparison of Kalman
Filter and Particle Filter used for localization of an Underwater
Vehicle, 9th IEEE International Conference on Ubiquitour
Robots and Ambient Intelligence (URAI), Daejeon, Korea/
November 26th  29th, pp: 350352, ISSN: 9781467531128/12,
2012.
15720316039292IJECSIJENS June 2016 IJENS
IJENS
Гораздо больше, чем просто документы.
Откройте для себя все, что может предложить Scribd, включая книги и аудиокниги от крупных издательств.
Отменить можно в любой момент.