Aparna Rathi Sc E
ELECTRONICS & RADAR
DEVELOPMENT ESTABLISHMENT
(L R D E )
LRDE
Introduction
Elements of Tracking
Basic Tracker
Mathematics and Modelling of Kalman Filter
Advanced Data Processing Algorithms
Case Studies
LRDE
What is Tracking ?
To determine targets ...
Position (co-ordinates)
direction
speed
acceleration
... Using radars output signals
LRDE
Origin of signals
Energy is beamed into open sky periodically
Pulses reflected by targets are detected by the radar
Reflections may come from more than one target
Coordinate information can be extracted from the detected
pulses by primary signal processing
LRDE
CONTEXT of RADAR DATA PROCESSING
BASIC SURVEILLANCE RADAR SYSTEM
TX
DX
ACU RX ADC SPU DE RDP
RADAR
DISPLAY
RMU
Line
Links
To
Remote
Center
Antenna
IFF
(SR)
MMI
Tracks
Update
Antenna Signals
SR Plots
PR Plots
TIMING,
CTL
Radar Data Processor
Using Discrete Time radar measurements which are corrupted with Noise, it is
required to:
1. Automatically detect the presence of moving objects (Targets)
2. Estimate their position with desired degree of accuracy.
3. Derive the parameters that are NOT measurable directly.
4. Track the targets motion during the track life time .
5. Utilize the track information.
6. Target classification
LRDE
RADAR DATA PROCESSOR
[RDP]
Radar Data Processor (RDP) is a Real time computer system with
a set of data processing algorithms, which uses successive scan radar
detections to automatically detect the targets, estimate kinematics and
track their movements
LRDE
RADAR DATA PROCESS SEQUENCE
SIGNAL
PROCESSOR
DATA
EXTRACTOR
RDP
2
9
4
#
1: - - - -
4: - - - -
7: - - - -
9: - - - -
RAW VIDEO
DETECTIONS
PROCESSING SEQUENCE IN A RADAR
Tracks
7
PLOTS
1
+
+
TRK R A S H M
LRDE
Sensor interface
Track Initiation
Plot-to-track correlation
Prediction
Track Filtering and estimation
Track termination
Track Display
MMI
Data Remoting
Diagnostics
LRDE
Correlation Track filter
Track init
Track
Prediction
Update Track
database
Detection reports
Correlated Trk
Uncorrelated rpt
Trk output to
display
Radar info
Tracking Process
LRDE
TRACKING CYCLE
First point Track
Tentative Track Level 1
Tentative Track Level 2
Confirmed Track
All the tracks are predicted periodically to
the next antenna visit time.
Gate computation for each track is done
and track gate is stored in target database.
TRACK PREDICTION
LRDE
- Generates list of reports and tracks eligible for correlation.
- Applies optimal association algorithm.
- Makes final observation-to-track assignments.
- For uncorrelated reports initiates a new possible track.
CORRELATION
LRDE
T1
TARGET CORRELATION
Gate
T2
P4
P2
P1
P3
PLOT TRACK
PAIRS
P4-T1
P2-T2
New track
init
T3
LRDE
- Filtering step incorporates the correlating observations into
updated track parameter estimates.
- For tracks that did not receive a correlating observation
the previous predicted estimates become filtered estimates.
- Kalman filter is used for filtering algorithm.
- Filtered tracks are sent to display.
FILTERING
LRDE
RDP IN SURVEILLANCE RADAR
- Automatic Track While Scan function (TWS)
- Simultaneous tracking of multiple targets.
- Suppression of False alarms.
- Real time display update, operator interaction.
- Capability to handle diverse target profiles
Speed
Evasive maneuvers
Crossings
Formations
- Process measurements from additional sensors
- Remote data Communication.
LRDE
CHALLENGES
- MEASUREMENTS:
--True Value + Stochastic Errors
- PROBABILISTIC DETECTIONS
--Missed data, insufficient data
- FALSE ALARMS/Clutter
--Ground Clutter, Unwanted echoes,
--Interference, environment
- UNPREDICTABLE / ASYNCHRONOUS DATA ARRIVALS
--Channel delay, misses, target motion
- AMBIGUITIES
--Closely spaced targets
--Excess / Missed reports
--Interferences
LRDE
Early warning for Air Defense (Track While Scan)
Air Traffic Control (ATC)
Maritime Surveillance
Air Borne Surveillance
Battle Field Surveillance (BFSR)
LRDE
Data Processing for Phased Array Radars
Filtered
Data
Filtered
Data
Receive &
Validate
Radar
Search
Plot Data
Radar
Track
Plot Data
Track
Beam
Request
Receive &
Validate
Plot
Manager
Correlation
Aircraft / Ballistic
Tracker
(Filtering & Prediction)
Track
Manager
RDP
Generation
of Beam
Request
Receive &
Validate
Radar
Search
Plot Data
Radar
Track
Plot Data
Track
Beam
Request
Receive &
Validate
Plot
Manager
Correlation
Aircraft / Ballistic
Tracker
(Filtering & Prediction)
Track
Manager
RDP
Generation
of Beam
Request
Radar
Search
Plot Data
Radar
Track
Plot Data
Track
Beam
Request
Radar
Search
Plot Data
Radar
Track
Plot Data
Radars for target and missile tracking (Rajendra for Akash)
Weapon Locating Radar
Airborne fire control radar (MSA for LCA, AESA for LCA)
Fire Control Radars
Long Range Surveillance and tactical radars (LSTAR)
LRDE
BENEFITS OF RDP
- Data Reduction & Information Enrichment
- Quantified performance measures
- Sophisticated algorithms for Data Association & Tracking.
- User interactions; Synthetic Display.
- Remote Communication for variety of applications
- Fusion of data from other Sensors.
- Optimisation of Radar performance, online resource
management, Beam Agility, (Multi Function Phased Array Radar)
LRDE
LRDE
Thus, RDP consists of
Tracking Filter
Track Data base
Correlation Algorithms
LRDE
Tracking Filters
Refer to Estimation Theory
Kalman Filter in Vector Form
LRDE
SYSTEM : Part of Universe which interests us.
ENVIRONMENT : Part of Universe which interacts with
System.
MODEL : - Representation of (Physical) system, its
behavior & performance.
- System is NOT known completely due to
uncertainties / complexities.
- A MODEL can be limited to the extent
feasible, to the extent sufficient for our
study.
- Different MODELs for the same system.
DYNAMIC SYSTEM : Except for few fundamental physical
constants, practically everything is
dynamic.
LRDE
STOCHASTIC MODELLING:
When precise Dynamics of a system is not known, we express our
ignorance more precisely by using probabilistic tools. Using
statistical information about the system behavior, we can
incorporate random characteristics in the model.
LRDE
PHYSICAL SYSTEM
MEASUREMENT
SYSTEM
OUTPUTS
MEASUREMENT
INPUTS
FIG. 1.1 DETERMINISTIC SYSTEM MODEL
DISTURBANCE
PROCESS
MEASUREMENT
ERROR PROCESS
DISTURBANCE
VARIABLES
MEASUREMENT ERRORS
CONTROL INPUTS
STATE SYSTEM
VARIABLES
MEASUREMENT
FIG. 1.2 STOCHASTIC SYSTEM MODEL
PHYSICAL SYSTEM
MEASUREMENT
SYSTEM
LRDE
Mathematical Models:
They represent the system & its behavior through fundamental
laws. The model is devised through empirical testing &
experience. They establish the relationship & interactions
between the parameters of interest.
Obtaining the mathematical model:
Process of translating knowledge about a physical system to a
mathematical relation is difficult but essential.
- Direct manifestation of Newtons law.
LRDE
ESTIMATION: Process of deducing STATE of a system
from what we can observe and what we
already know about the system.
STATE / STATE VECTOR: Set of variables of interest representing
the system. It includes estimates of
system characteristics NOT
measurable but derived from
observation sequence.
LRDE
ESTIMATION
- MATHEMATICAL MODEL OF TARGET MOTION
- MODEL OF OBSERVATION PROCESS
- MODEL OF UNCERTAINITIES
- in the system
- in the measurement
- ESTIMATE THE VARIABLES WHICH TOGETHER REPRESENT THE
STATE OF THE SYSTEM
- COMPUTE THE MEASURE OF UNCERTAINITIES IN ESTIMATION
- COMPUTE RELATED PARAMETERS: Track Quality, classification
LRDE
GAUSSIAN NORMAL PROBABILITY DENSITY FUNCTION
- Representation of uncertainty in measurement.
- Physical system noise / meast. Noise are caused by a number of
unknown, independent random process.
CENTRAL LIMIT THEOROM: Under certain condition SUM OF distribution of
independent RVs each having arbitrary
distribution tends to become NORMAL
as number of variables
|
.
|
\
|
o
o [
2
2
2
m x
2
1 ) (
exp
LRDE
Optimal State Estimation
It is a process of deducing a minimum error estimate of the state of the
system.
Estimation, utilises the measurement inputs, knowledge of the system,
assumed statistics of the system noises & observation errors, and the
initial conditions.
Minimization of estimation error is in a well defined statistical sense with
some stated criterion of optimality.
LRDE
Uncertainties in dynamics and external disturbances on
the system, use Normally distributed (discrete) white
noise
w
k
~ N(0,Q)
mean, = 0
variance, o
2
= Q
Clutter
Uniformly distributed in space
Poisson distributed in time
New Target
Uniformly distributed in space
Uniformly distributed in time
LRDE
KALMAN FILTER
Optimal recursive estimation algorithm. It uses
knowledge of system, uncertainties in models
and incorporates all the information provided
and evaluates state with better precision than
individual measurement.
LRDE
Linear system
System parameters are linear
function of parameters at some
previous time
Measurements are linear function of
parameters
White Gaussian noise
White: uncorrelated in time
Gaussian: noise amplitude
LRDE
System:
our
knowledge
of the
system
System
state:
unknown
system
parameters
model
+
|
noise
noise
? KF
measurement
LRDE
RDP_CEP
System:
our
knowledge
of the
system
model
measurement
KF
Parameters
x
x
i
= A x
i - 1
m
i
= H x
i
+
|
noise
noise
+ q
i
+ r
i
using vectors and matrices
LRDE
Prediction: x
i
-
= A x
i 1
Correction: x
i
= x
i
-
+ K (m
i
H x
i
-
)
x
i
= A x
i 1
+ q
m
i
= H x
i
+ r
x
^
x KF
Kalman gain matrix
minimize the difference x - x
^
^
^ ^
^
LRDE
KALMAN FILTERING SUMMARY
. . .
1. TARGET X
K
= [x y z x y z]
T
;CVM
STATE
MODEL
2. STATE
TRANSITION X
K
= A
K-1
X
K-1
+ W
K-1
|
K
: STM
W
K
~ N (0,Q
K
)
Q
K
: Plant noise representing uncertainty in model.
3. OBSERVATION MODEL
Z
K
= H
K
X
K
+ v
K
H
K
: Observation Matrix
v
K
: Additive ZMWG Noise ~ N(0,R
K
)
R
K
: Meast. Noise Covariance Matrix uncorrelated with W
K
4. STATE-OBSERVATION RELATIONSHIP
Z
K
- H
K
X
K
= Residue
LRDE
1. Predict the state to current discrete Time k
2. Compute Prediction Covariance
1
k k
x
1
1
=
k
k
X A
1 k k
P
~
1 1 1
~
+ =
k
T
k k
Q A P A
KALMAN FILTERING SUMMARY (2)
LRDE
3. Compute KALMAN GAIN
4. Estimate the State (FILTERING)
^ ^ ^
X
K
/
K
=X
K
/
K-1
+ K
K
[Z
K
- HX
K
/
K-1
]
5. Estimate Filtered Covariance
~ ~ ~
P
K
/
K
= P
K
/
K-1
- K
K
HP
K/K-1
~
= [I - K
K
H] P
K/K-1
1
K
T
1 k K
T
1 k K k
R H P H H P K
+ = ]
~
[
~
KALMAN FILTERING SUMMARY (3)
LRDE
Matrix description of system state, model and
measurement
Progressive method
Proper dealing with noise
prediction
correction
LRDE
NOISE
PROCESS
OBSERVATION
PROCESS
KALMAN
FILTER
GAINS
COMPUTATION
PREDICTION
PROCESS
State
Estimate
X
k
V
k
Z
k
X
k / k
X
k / k-1
P
k / k
~
P
k / k-1
~
^
^
LRDE
System estimation accuracy depends on the model we
chose to track aerial targets
Best Variance reduction is possible with a model closed
to actual target dynamic model
Minimum lag and Maximum variance reduction cannot be
achieved simultaneously
Automatic Adaptability
Minimum Maneuver detection delay
LRDE
IMPORTANCE OF MODELING
Challenge
Minimum lag and Maximum variance reduction cannot be
achieved simultaneously
Approaches
Plant Noise adjustment
Variable Dimension Filter
Interacting Multiple Model Filter
LRDE
Tracking Maneuvering Targets
INTEGRATION
FILTER - 1 FILTER - 2
PROBABILITY
UPDATE
x1(k-1) x2(k-1)
x1_0(k-1)
x2_0(k-1)
z(k)
x1(k)
x2(k)
U1(k) x(k)
P1(k)
P2(k)
U1(k-1)
IMM ALGORITHM WITH TWO FILTERS
OUTPUT
GENERATION
LRDE
To distinguish between targets and clutters
which is which?
LRDE
Kalman Filter require the knowledge of which
measurements to use to update and correct
Presence of clutters prevent direct application of
Kalman Filter
Single target problem
Probabilistic Data Association (PDA)
based on weighted probability of each measurement
Nearest Neighbour (NN)
choose the nearest measurement
Global Nearest Neighbour
The need for concept of gating to improve the
processing power requirement
LRDE
90
120
150
Target Measurement
True Target Trajectory
PDA Filter estimates
Clutter Points
Three model
IMM-PDA
Filtering using
range-rate
measurements
90
120
Target Measurement
True Target Trajectory
PDA Filter estimates
Clutter Points
Track break
Directio
n of
target
motion
LRDE
LRDE
Case Studies
LRDE
RDP in Airborne Radar
Systems (AEW&C, AESA)
Challenges in Airborne Tracking Algorithms
Detection problems in airborne radars
Detection of targets in clutter background
Tracking under dense clutter environment
Plot to track correlation
Passive/Active initiation to preserve radar resources under severe
clutter leaks
MPRF mode of operation with range rate as a filter state leading to
non-linear estimators
Tracking of maneuvering targets with surveillance as primary
function
Different modes of operation (A/A, A/S, A/G)
Track initiation and transition policies
Tracking of ground moving targets and STAP
Filter tuning for variable update rates is difficult
Platform motion compensation
Tracking filter to fully exploit the available information of the
platform
NORMAL SEARCH
TRACK INIT
SURFACE SEARCH
FULL TRACK
EXTENDED MODE
SEARCH
VERIFICATION
SPECIAL BEAM
LRDE
Cued search
Sea surface
search
Weapon
support
Air-to-ground
targeting
SAR wide-area
ground mapping
Electronic
protection
Exovolume
tracking
Selectable search volumes
Ground moving
targets
Exovolume
track
LRDE
Target Tracker Design
Requirement for Target Tracker
Requirement:
To quickly confirm new targets and maintain
tracks on Maneuvering Targets
Problem:
Frame time limitation, tracking targets with
highly different dynamics, 500 TWS tracks,
Narrow azimuth beam
3 Model IMM is Chosen
Model Plant Noise
Values
Constant Velocity(CV) Model 1
Constant Acceleration Model (CA1) 100
Constant Turn Rate Model 10
LRDE
Derivation of kalman filter
- State Transition matrix
X
k
- Target State at k
th
instant
Z
k
- Measurement (Position,velocity or acceleration) at k
th
instant
V
k
, W - Gaussian Noise
H - Measurement Matrix
K - Kalman Gain
P Prediction Error Covariance Matrix
Q Noise Covariance Matrix
R - Measurement Covariance Matrix
Kalman filter
Kalman filter equations
Kalman filter
Assumptions
Kalman filter
MODEL
Kalman filter
Assumptions
Kalman filter
Goal
Kalman filter
Derivation steps
Kalman filter
Step 1
Kalman filter
Step 1
-
-
X
k
Kalman filter
Step 1
Kalman filter
Step 1
Kalman filter
Step 1
Kalman filter
Step 2
Kalman filter
Step 2
-
Kalman filter
Step 2
+
^
Kalman filter
Step 2
-
V
Kalman filter
Step 2
Kalman filter
Step 2
Kalman filter
Step 2
R
Kalman filter
Step 2
Kalman filter
Step 2
Kalman filter
If so, which measurement come from which target or
clutter?
Organized and explore systematically all combination
Hypothesis construction
Hypothesis probability
LRDE
RDP_CEP
MHT is a multiple target tracker
It seeks to find the most likely hypotheses by exhausting
all possible associations between measurements and
targets
Implementation issues and software development
LRDE
RDP_CEP
Track-Report
Correlation
Ambiguity
LRDE
RDP_CEP
Real time constraints
Requirement of dynamic data structures
Data structure for storing huge number of hypothesis and
tracks
Efficient linking of multiple hypothesis records and target
tracks and measurement records
Data Management
Concurrent multitasking design
Design Challenges
LRDE
RDP_CEP
Scan1
m1
m2
Scan
2
m3
m4
New target
False target(clutter)
Existing target
LRDE
RDP_CEP
Software Flowchart
of MHT algorithm
No
New Observation Set
Gating with existing clusters
Loop: Observations
No
No. of cluster correlate
with this observation
finished
Yes
0 ,New Cluster > 1, Merge clusters
1
Loop: Clusters
New cluster
Yes
Generate feasible correlation hypotheses
Evaluate Probability of each hypotheses
Combining and pruning of hypotheses
finished
No
Yes
Filter all tracks and delete bad tracks
LRDE
RDP_CEP
Hough Transform
LRDE
RDP_CEP
Hough Transform-Simulations
Histogram in u space
LRDE
RDP_CEP
1/5
th
reduction in clutter after Hough
transform processing using sliding window
method (8 scans)
Hough Transform-Simulations
LRDE
RDP_CEP
RDP
RDP
.
k
X
.
k
X
.
k
X
RDP
RDP
