Outline 2(34)

Target Tracking: Lecture 1

Course Info + Introduction to TT

Course info
Umut Orguner Introduction to Target Tracking
[email protected]
Division of Automatic Control
Department of Electrical Engineering
Linkoping University
Linkoping, Sweden

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET

Course Info 3(34) Course Info: Literature 4(34)

A total of 7 meetings
The only book that covers all of it is
Meet once a week for two hours on Thursdays
Book:
G: Glashuset
S. Blackman and R. Popoli, Design and Analysis of Modern Tracking
Nr. Date Place Subject Systems, Artech House, Norwood MA, 1999.
1 21/10 G Course Info & Intro. to TT
2 28/10 G Single TT Issues A good reference in the long term if you are dealing with TT.
3 11/11 G Maneuvering TT Related section numbers to the lectures are given in the course
4 18/11 G Multi TT Part I: GNN & JPDA web page (lectures).
5 02/12 G Multi TT Part II: MHT When a more detailed coverage or derivations are necessary,
6 16/12 G Multisensor TT they will be provided in the class.
7 ?/? G Overview of Unconventional TT

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET
Course Info: Student responsibilities 5(34) Course Info: Optional 6(34)

Get an overview on the subject before the class by reading the

related sections in the book (see webpage of lectures).
Do not indulge in the details in the book while reading, just try to
For people who would like to do more and get more points
get an overview.
(3hp), there is an opportunity to do a project.
Complete the exercises and submit their small scale reports.
The aim of the projects is to gain additional knowledge on the
Nr. Subject tracking theory and algorithms.

1 Track handling in clutter and missed detections Depending on the amount of work required, you can work in
2 Maneuvering TT with multiple model filtering groups, at most, of 2 people.
3 Multi TT Part I: GNN & JPDA
4 Multi TT Part II: MHT
5 Multiple sensors and track fusion

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET

Course Info: Optional contd 7(34) Course Info: Optional contd 8(34)

Project Subjects We are going to implement most of the

conventional algorithms with standard sensor models in the course.
Procedure:
So possible project ideas are study and implementation of
1. Project proposal (a 2 page document)
Some conventional algorithm with unconventional sensors or
targets (important probabilistic modeling work on the sensor 2. Progress report (halfway during the project)
and the target is required) e.g., 3. Final report
More details are available on the webpage.
extended targets unresolved targets
Some unconventional algorithm e.g.,

track before detect random set based approaches (e.g., Any questions about the course?
(joint) integrated PDA PHD, CPHD, MeMBer filters)

probabilistic MHT symmetric measurements

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET
Introduction to Target Tracking (TT) 9(34) A Conventional TT System 10(34)

Definition
A target is anything whose state is of interest to us.
Measurement Gating & Estimation &
Processing Association Prediction
State we are interested in can change with time with a certain
dynamics which is itself unknown. TRACKER

Measurement origins are uncertain. Track

Predictions Handling
There are false measurements PFA > 0.
Some measurements are missing PD < 1.
We generally have no initial guess or estimate. USER

Definition
Target tracking, in its most general and abstract form, is a
non-vanilla version of dynamic estimation theory.

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET

Measurements 11(34) Measurements 12(34)

Definition Thresholding Illustration with a Radar

The term measurement contains all observed quantities included in
a (possibly processed) report output from a sensor.

Measurement (pre)processing generally includes a form of

thresholding (measurement detection) process.
Information loss during the thresholding is evident. In very low
SNR scenarios, thresholding might not be used, which leads to
Track Before Detect algorithms with high computation cost.
We consider point or contact measurements.
This means lots of processing that is out of hands of the
TT-engineer. What is PD , what is PFA ? Good modeling of Figures taken from: Y. Boers, H. Driessen, J. Torstensson, M. Trieb, R. Karlsson, F. Gustafsson,
Track-before-detect algorithm for tracking extended targets, IEE Proceedings Radar, Sonar and Navigation,
measuring process must be obtained. vol.153, no.4, pp.345351, August 2006.

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET
Measurements contd 13(34) Measurements contd 14(34)

Types of measurement sources:

Types of point measurements:
One of targets that has been previously observed;
Kinematic measurements e.g.,
A new target;
Position (pixel indices),
Range, False alarm (clutter).
Range rate (radar doppler),
Bearing. Clutter
Attribute measurements e.g., A false measurement (false alarm or clutter) in tracking
Signal strength, terminology generally refers to the concept of persistency.
Intensity, In other words, a persistent false alarm (clutter) is considered a
Aspect ratio, target to be tracked even if we are not interested in what or
Target type.
where it is.
We are going to be concerned with only the kinematic
If one of our interesting targets gets in the vicinity of
measurements.
uninteresting false targets, we become prepared.

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET

Measurements: Modeling 15(34) Measurements: Modeling contd 16(34)

Target originated measurements:

Target originated measurements:
No sensor is perfect.
j In addition to sensor measurement noise ek , there is a detection
yk = h(xk ) + ek
process with probability PD < 1 in many sensors.
Example models: Detection probability PD can be a characteristics of the sensor
Simple cartesian Range (raw measurement processing algorithm) as well as the target
state, i.e., PD might depend on the specific target position and it
iT q
j j can vary from target to target.
h
j j yk = (xk )2 + (yk )2 + ek
yk = xk yk + ek
It is generally difficult to find an exact formula for PD .
Bearing only log range (received signal strength (RSS)) Approximations and heuristics abound.

j j
 Every obtained measurement from the sensor includes two
yk =
j j
arctan2(yk , xk ) + ek yk = log (xk )2 + (yk )2 + ek sources of information:
1. Detection info PD ; 2. The actual value of the measurement yk .

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET
Measurements: Modeling contd 17(34) Targets: Tracks 18(34)

Clutter: Non-persistent measurements which are not originated from

a target.
Prior information is important. Definition
Clutter maps are sometimes existent. A track is a sequence of measurements that has been decided or
Sensor (processing algorithm) characteristics
hypothesized by the tracker to come from a single source.
Some characteristics might be available from the manufacturer.
Experiments might be performed.
The case of minimal prior info Usually, instead of the list of actual measurements, sufficient
statistics is held e.g., mean and covariance in the case of a KF,
Number of FAs in a region with volume V is Spatial FA distribution: particles in the case of a PF.
modeled as a Poisson distribution with clutter Uniform in every region
Generally each arriving measurement must start a track. Hence
rate FA (number of FAs per area per scan). with volume V .
PD is included in PFA . tracks must be classified and must not be treated equally.
1
FA
( FA V )mk exp( FA V ) pFA (yk ) =
PFA (mFA V
k )=
mFA
k !

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET

Targets: Tracks contd 19(34) Targets: Types 20(34)

We can characterize targets considered in target tracking into

categories depending on their size with respect to sensor resolution
Track types: According to their different life stages, tracks can be
(depends on target-sensor distance too).
classified into 3 cases.
Point target: A target that can result in at most a single
Tentative (initiator): A track that is in the track initiation process. measurement.
We are not sure that there is sufficient evidence that it is actually This means its magnitude is comparable to sensor resolution.
a target or not. However, an extended target can also be treated as a point
target by tracking its centroid or corners.
Confirmed: A track that was decided to belong to a valid target
Extended target: A target that can result in multiple
in the surveillance area. This is one end of initiation process.
measurements at a single scan.
Deleted: At the other end of the initiation process, this is a track Unresolved targets: This denotes a group of close targets that
that is decided to come from all random FAs. All of its info can collectively result in a single measurement in the sensor.
should be deleted. Dim target: This is a target whose magnitude is below sensor
resolution. These can be tracked much better with track before
detect (TBD) type approaches.
AUTOMATIC CONTROL AUTOMATIC CONTROL
U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET
Targets: Modeling 21(34) Targets: Modeling contd 22(34)

General state space model (Nearly) Coordinated turn model, i.e., nearly constant speed,
xk = f (xk1 ) + wk constant turn rate model
T
vxk vyk k

State with Cartesian velocity xk , xk yk .
Example models
(Nearly) constant velocity model sin(k1 T )
1cos(k1k1 T)

      2  1 0 k 1 0 T2 /2 0 0
xk 1 T xk 1 T /2 1cos(k1 T ) sin(k1 T )
xk , = + ak
0 T2 /2 0
vxk vxk1 0 1 0

0 1 T k 1 k 1

xk = xk1 +

T 0 0 k
0 0 cos(k1 T ) sin(k1 T ) 0
where ak N (0, a2 ) is a white noise.
0 T 0
0 0 sin(k1 T ) cos(k1 T ) 0
(Nearly) constant acceleration model 0 0 1
0 0 0 0 1
T2 /2
2
xk 1 T xk 1 T /2
xk , vxk = 0 1 T vxk1 + T k There are versions with polar velocity.
axk 0 0 1 axk1 1
This model is a little different than the model in the book. It was
where k N (0, 2 ) is a white noise. taken from [Bar-Shalom (2001)].

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET

Targets: Filtering 23(34) Targets: Model Mismatch 24(34)

When measurements corresponding to a target are obtained, the State estimators for kinematic models are sophisticated
calculation of the sufficient statistics is done via state estimators low-pass filters.
(filters). The bandwidth of the filter is a trade-off between
Early tracking systems: very low computational capacity = 1. Following the target motion
Steady state Kalman filters: and -filters. Kept 2. Reducing the noise (clutter,measurement noise)
an integer quality indicator instead of covariance. The model mismatch problem in target tracking is called
maneuvers.
Kalman filters (KFs) and extended KFs are the most common
Unlike the common concept of maneuvers in real life, a
approaches.
maneuver might not necessarily be a higher order motion than
Unscented KF (UKF) and other sigma-point approaches got
that of the filter model.
much criticism (and undermining) from orthodox people in the
When one is using a high order model and the target is
field when they were first introduced. Particle filters (PFs) were
assuming a lower order model than the filter, this is also a valid
despised.
maneuver for target tracking filters.
With the ever increasing computational resources, they are now This is because, the filter could have been using a lower order
commonly accepted as valid methods for target tracking. model and reduced the noise levels even more.
AUTOMATIC CONTROL AUTOMATIC CONTROL
U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET
Targets: Model Mismatch contd 25(34) Targets & Measurements 26(34)

Definition Definition
A maneuver is any motion characteristics that the target is assuming
Association is the process of assigning measurements to existing
other than the model used by the filter.
tracks or existing tracks to measurements (measurement-to-track
association vs. track-to-measurement association).
Maneuvers degrade the filter performance and makes the
estimators susceptible to noise and clutter.
In a classical air traffic control (ATC) application, there are
Maneuvers, hence, should be checked continuously. hundreds of targets and measurements.
Since target maneuvers are often in a set of finite number of Possible combinations are incredibly many.
models, multiple model approaches (Interacting multiple model
Not all of the possible associations are physically feasible.
(IMM) filter being the most famous) became popular.
One must exclude these highly unlikely combinations from
Using all possible models at the same time made hypothesis
further consideration as soon as possible.
checks about maneuvers unnecessary.

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET

Targets & Measurements contd 27(34) Targets & Measurements contd 28(34)

Definition Even if gating reduces the number possible association

Gating is the process of using the sufficient statistics of a track to combinations, there still remains some association uncertainty.
exclude the possibility of assigning a measurement to the track. The These are handled by some other association algorithms:
region in the measurement space that the measurements are Nearest neighbors (NN) (single, in general, non-global (local)
allowed to be assigned to the target is called as the gate. hard decision)
Global nearest neighbors (GNN) (single unique hard decision)
(Joint) probabilistic data associations (JPDA) (soft decisions, i.e.,
Using maximum velocity assumptions yk2 no decision or decision with probabilities)
yk1 3
about the target can give coarse gates. yk|k1
Multi hypothesis tracking (MHT) (making hard but multiple
More detailed gates are formed using the
y yk3
yk4
decisions and keeping them until sufficient evidence arrives)
yk5
1
yk|k1 2
predicted measurement means and yk7
yk|k1 What association algorithm to use depends on the SNR of the
yk6 system and amount of computational resources that can be
innovations (measurement prediction)
covariances of the track. allocated to association.
x

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET
Multiple Sensors: Centralized Tracking Case 29(34) Multiple Sensors: Decentralized Tracking Case 30(34)

Measurement
Measurement TRACKER
Processing
Processing

Measurement Track
Measurement Measurement TRACKER Association
TRACKER Processing
Processing Association & Fusion

USER Measurement
Measurement TRACKER
Processing Processing USER

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET

Multiple Sensors 31(34) Multiple Sensors contd 32(34)

Issues we will mostly neglect

Architecture Issues: Registration issues.
Centralized approach is optimal. Detection of biases and their compensation between various
Communication constraints make it unattractive because sensors.
measurement communication rates are generally higher than Out of sequence measurements: Measurements delayed in the
track communication rates. communication which arrive into the fusion center after a more
Some of the computations can also be distributed in the current measurement from the same source has already been
decentralized case. processed.
tk`1 tk` tk`+1 tk`+2 tk1 tk
Systems susceptibility to failures in the tracking center is also target time
an important issue.
We are going to concentrate on the decentralized architecture.
fusion center time
tk`1 tk` tk`+1 tk`+2 tk1 tk tk+1

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET
Multiple Sensors contd 33(34) References 34(34)

S. Blackman and R. Popoli, Design and Analysis of Modern Tracking Systems.

Norwood, MA: Artech House, 1999.
We will concentrate on
Y. Bar-Shalom and X. R. Li, Multitarget-Multisensor Tracking: Principles,
Track association: Process of deciding whether two or more Techniques. Storrs, CT: YBS Publishing, 1995.
tracks obtained from local trackers are coming from the same
Y. Bar-Shalom, X. R. Li, and T. Kirubarajan, Estimation with Applications to
source. Tracking and Navigation. New York: Wiley, 2001.
Track fusion: Process of combining the sufficient statistics of two
R. Mahler, Statistical Multisource Multitarget Information Fusion. Norwood, MA:
or more tracks from local trackers when they are associated to
Artech House, 2007.
each other.
Important issue: Correlation between the local estimation errors Y. Bar-Shalom and T. E. Fortmann, Tracking and Data Association. Orlando, FL:
although the local sensors measurement errors are independent. Academic Press, 1988.
S. Blackman, Multiple Target Tracking with Radar Applications. Norwood, MA:
Artech House, 1986.

AUTOMATIC CONTROL AUTOMATIC CONTROL

U. Orguner REGLERTEKNIK
U. Orguner REGLERTEKNIK
Target Tracking LINKPINGS UNIVERSITET Target Tracking LINKPINGS UNIVERSITET