LEVEL 1 PROCESSING:

DATA ASSOCIATION AND CORRELATION

DONE BY
GREESHMA SHINE SHAJI
Reg no: 47321002
S2 Mtech INSTRUMENTATION TECHNOLOGY
CUSAT
INTRODUCTION
• A problem that is to data fusion involves assignment or
correlation.
• the problem is simple to state, but difficult to solve.
• problem involves how to determine which observation
form multiple sensors belong together, representing
observation of the same physical object.
fig shows:

• dynamic targets: circle, square, and star

• targets are observed by several sensors s1 s2 and s3.
• observations y1,y2,y3 sorted into groups and assigned as
belonging to tracks
Complications
• It is unknown a priori how many targets actually exist
• The targets may be closely placed and dynamically
manuevering
• There may be poor observing conditions, that cause false
alarm.
• An intelligent opponent may be deliberately create
confusion via electronic counter measure(ECM)
• Sensors may have relatively low resolution , causing an
inability to distinguish between closely targets.
Process model for correlation
• solution of data correlation problem occures within a
broader context of a multisenser data fusion system,
• Fusion of multisensor data may occure in a hierarchial
fashion with many data fusion nodes.
• Each node may accept data dirctly from:
• sensors
• Human inputs
• from internal and external data base
• other data fusion nodes
for example:
• A data fusion system onboard a single air craft may
involve multiple interacting fusion nodes.
• one fusion node associated with a multisensor
identification front foe( IFF) system, and another node
corresponding to an automatic target recognition(ATR)
and input data from external intelligence sources,
• Within each induvidual nodes, data processing involves
three key functions:
1. Data preparation
2. Data correlation
3. State estimation and prediction.
 data fusion tree nodes

prior data fusion nodes

and sources

source/sensor state
resource management control
• Data correlation occures within each processing nodes,
• Each processing nodes accept inputs from sensors, data
base, and other nodes and perform data fusion, to
estimate the position, velocity, attributes and identity of
target.
• Data correlation accepts input data and sorts the
observation in groups.
Required functions for discription of correlation
processing

1. Hypothesis generation(HG)
2. Hypothesis evaluation
3. Hypothesis selection

HYPOTHESIS :Possible explanation or interpretationof the

data
Hypothesis generation
• processes incoming data either sequentially or in batches.
Hypothesis Evalution
• HE processces the association matrix to evalutionate the
likehood of the feasible hypothesis output from hypothesis
generation.
• The output of HE function is a quantified evolution or
ranking of the feasible hypothesis.
• HE populates or fills in the association matrix with
numerical values represinting the fesability or relative
likehood ogf alternative hypothesis explaining incoming
data.
Hypothesis Selection.
• HS uses the results from the HE function to guide a
search for the best association hypothesis for all input
data.
The output from HS function are used in subsequent state
estmation techniques to combine the data for imporoved
estimation and prediction of the state.
HYPOTHESIS GENERATION
• HG function accepts input data from sensors or
other data fusion nodes and developes feasable
hypothesis or explanation of input data.
 Basic philosophical issues

• Assignment uniqueness
• incoming data can be assigned into only one list
Number of scans
• The word ‘scan’ discribes how some sensors dynamically
scan an obsevation area and collect snapshots of
observations.
• single sacn & multiple scan
advantages of multiple scans:
• more information
Advantages of single scan:
• computationally more efficent
• easier to implimemt
Disadvantage:
• provide more ambiguity in interpretation of observed data.
• assignment ambuigity
Multiple v/s single hypothesis
• single hypothesis approch:
In this approch we develop a single interpretation
concerning the assignment of the data to tracks.
• Multiple hypothesis approch:
in this approch an attempt is made to enumerate these
alternative hypothesis.These alternative hypothesis are
sysytamatically developed and are subsequently evoluted
and select the most likely hypothesis
advantage:
It allow explicit considerations and evolution of allternative
explantion of the data.

Disadvantage:
The number of alternative hypothesis grows rapidly with
increasing number of observation hence approch may
become computationally infeasable for real time
application.
Charecterizing the Hypothesis Generation Problem

The understading of the hypothesis generation ptoblem

depends upon three basic factors:
1. The charecterstics of the input data
2. The feasible hypothesis types
3. The measure of desired perfomance for hypothesis
generation.
Input Data Charecterstics
• The inputs to the HG function are the observations,
reports and priori information about the sensors, the
environment and entities being observed.
• Input hypothesis generation involves one or more sets of
sensed information.
• input datas can be: identity atributes, kinematic data ,
parametric attributes etc....
• The available data strongly influences the HG problem
definition.
• The ability to distinguish between observations based on
parameters such as location, velocity, attributes,or identity
depends upon wether those data are avilalable
• The extent to which a designer is willing to make
assumption about target motion, sensor accuracy signal
propagtaion etc...
• An additional factor that must be consider as an input
variable involves how the input data must be ingested
3 basic ways:
• process data sequentially: one at a time
• process the data sequentially in groups or batches,
• process the data after all data are received by the fusion
system
Output Data Charecterstics
• The output of hypothesis generation is a list of
hypothesisthat represents feasible interpretations of the
available data.
• This output is sometimes represented as a matrix and the
matrix map reports observations to hypothesis
• unlike hypothesis evalution and hypothesis selection
functions the definition .the defnitions of the output
hypothesis for HG is itslf
• Report to report:- Each individual observation or report
may be hypothesized to be related to a previously
observed report. Thus he incoming reports may be
grouped with other previous report, thereby hypothesized
that they belongs to the same physical target or entity.
• Based on measure of observation to observation
closeness or similarity.
• Report to track:- incoming reports may be hypothesized to
be evidence associated with an existing tracked object.
Thus one may consider as receive observation.

Track to track:- Same as track to tack except the sensor or

source proves a state vector as input.
• Spatio –temporal:- an incoming report belongs to one or
more geographically distributed reports, previously
observed.
• Multi- spectral:- Two (or more ) non- commensurate
sensor reports are related to the same physical target or
entity.
• Cross-level:- observations of different physical entities
belong to same complex entity
• Multisite sources:- observations from different
Sites or sources are hypothesized to be associated with the
same physical entity.
Multi scenes :- Tow alternate descriptions are developed to
interpret a collection of data , at a snapshot in time.
N-D set partitioning:- N mutually exclusive list of data are
matched to multiple hypotheses
• Labeled set covering:- Further generalization of N-D set
partitioning.
• Other factors in defining feasible hypotheses involves to
which knowledge is available about target behavior,
sensor perfomance , signal perfomance etc...
Perfomance Measure Charecterstics

The hypothesis generation technique that are selected to

implement a data fusion system should maximize the
system perfomance at a minimum cost of resources.
A particular challenge is to identify a sufficent number of
feasible hypotheses, while truly represent the physical
situation being observed , while eliminating unlike
hypotheses.
Perfomance measures for hypothesis generation includes:-
1. software life - cycle cost and complexity
2. processing efficency
3. Accuracy of identified hypothesis
4. Number of hypothesis
5. unity of available information
6. Robustness to errors and incorrect modeling
7. Hardware timing/sizing limitations
8. Ease of training on simple data
Overview of Hypothesis Generation Techniques
• The approch to developing a solution for hypothesis
generation can be separated into two aspects:

1. Hypothesis enumeration
2. Identification of feasible hypothesis
Hypothesis enumeration
• It involves developing a globel set of possible hypothesis
based on physical, stastical or explicit knowledge about the
observed environment.
• The first function required to accomplish hypothesis
generation is the identification of the set of potentially
feasible hypothesis.
• This identification of hypotheses based on a number of
models or techniques,
1. Physical models
Models of sensor perfomance, signal propagation,target
motion, intervisibility to identify possible hypotheses.
2. Syntax- based models
Use of syntactical representations to discribe the makeup
of complex entities such as military units
3. Doctrine- based senarios
Definition of tactical scenarios , enemy doctrine, anticipated
taregets, sensors, engagements to identify possible
hypotheses.
• 4. Probabilistic models
Probabilistic models of track initiation,tarck length,
birth/death probabilities to discribe posible hypotheses.
5. Ad hoc models
Ad hoc discriptions of possible hypotheses to explain
available data; may be based on exhuastive enumeration of
hypotheses.
Identification of Feasible Hypotheses.
• The second function required for hypothesis generation
involves reducing the set of possible hypotheses to a set
of feasible hypotheses.
• The challenge is to reduce thenumber of possible
hypotheses to a limited set of feasible hypotheses ,
without eliminating any viable alternatives that may be
useful in subsequent HE and HS processing.
Automated techniques include:
• 1. Pattern recognition technique:
use of pattern recognition techniques such as cluster
analysis, nueral networks, or gestalt methods to identify
natural grouping in input data.
2.Gating techniques
use of a priori parametric boundaries to identify feasible
observation pairing and eliminate unlikely pairs; techniques
include kinematic gating, and parametric range gates.
• 3. Logical templating
Use of prespecified logical conditions and parametric
factors to specify spatio-temporal conditions , causal
relations and entity aggregations for feasible hypotheses.
4. Knowledge based methods
Establishment of explicit knowledge via rules scripts ,
frames, fuzzy relationships , Bayesian belief networks,
nueral networks
THANK YOU