Transcript
Page 1: Optimal Estimate of System State Lecture 1 Observed ...users.encs.concordia.ca/~ymzhang/courses/ESIF/ESIF_Lect1_intro.pdf · Mathematical Techniques in Multisensor Data Fusion, Artech

You

min

Zh

ang

Pho

ne: 7

912

7741

O

ffice

Loc

atio

n: F

UV

0.2

2E

mai

l: ym

zhan

g@cs

.aau

e.dk

http

://w

ww

.cs.

aue.

auc.

dk/~

ymzh

ang/

cour

ses/

ES

IF/in

dex.

htm

l

FP

8-4:

Est

imat

ion

an

d

Sen

sor

Info

rmat

ion

Fu

sio

n

FP

8-4:

Est

imat

ion

an

d

Sen

sor

Info

rmat

ion

Fu

sio

n

Lec

ture

1Re

view

on

Kalm

an F

ilter

s•C

ours

e ou

tlin

e•W

hat

is K

alm

anfi

lter

(KF)

and

app

licat

ions

of

KF?

•The

dis

cret

e-ti

me

Kalm

anfi

lter

(DKF

)•T

he c

onti

nuou

s-ti

me

Kalm

anfi

lter

(CKF

)•T

he K

alm

anfi

lter

s fo

r no

nlin

ear

syst

ems

3Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Co

urs

e O

utl

ine

Co

urs

e O

utl

ine

1.R

evie

w o

n K

alm

an fi

lter

and

its v

aria

nts

2.M

ultip

le-m

odel

bas

ed K

alm

an fi

lters

3.S

imul

tanu

ous

stat

e an

d pa

ram

eter

es

timat

ion:

Tw

o-st

age

Kal

man

filte

r

4.In

trod

uctio

n to

sen

sor

info

rmat

ion

fusi

on (

I)

5.In

trod

uctio

n to

sen

sor

info

rmat

ion

fusi

on (

II)

and

cour

se s

umm

ary

4Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Th

e P

rob

lem

T

he

Pro

ble

m ––

Wh

y d

o w

e n

eed

W

hy

do

we

nee

d K

alm

anK

alm

anF

ilter

s?F

ilter

s?

Mea

suri

ng

D

evic

esE

stim

ato

r

Mea

sure

men

tE

rror

Sou

rces

Sys

tem

Sta

te

(des

ired

but n

ot

know

n)

Ext

erna

l C

ontr

ols

Obs

erve

d M

easu

rem

ents

Opt

imal

E

stim

ate

of

Sys

tem

Sta

te

Sys

tem

Err

or S

ourc

es

Sys

tem

Bla

ck B

ox

•S

yste

m s

tate

can

not b

e m

easu

red

dire

ctly

•N

eed

to e

stim

ate

“opt

imal

ly”

from

mea

sure

men

ts

Page 2: Optimal Estimate of System State Lecture 1 Observed ...users.encs.concordia.ca/~ymzhang/courses/ESIF/ESIF_Lect1_intro.pdf · Mathematical Techniques in Multisensor Data Fusion, Artech

5Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

All

Ro

ads

Lea

d f

rom

Gau

ssA

ll R

oad

s L

ead

fro

m G

auss

1809

“ …

sin

ce a

ll ou

r m

easu

rem

ents

and

obs

erva

tions

are

no

thin

g m

ore

than

app

roxi

mat

ions

to th

e tr

uth,

the

sam

e m

ust b

e tr

ue o

f al

l cal

cula

tions

res

ting

upon

them

, and

the

high

est a

im o

f al

l com

puta

tions

mad

e co

ncer

ning

con

cret

e ph

enom

enon

mus

t be

to a

ppro

xim

ate,

as

near

ly a

s pr

actic

able

, to

the

trut

h. B

ut th

is c

an b

e ac

com

plis

hed

in

no o

ther

way

than

by

suita

ble

com

bina

tion

of m

ore

obse

rvat

ions

than

the

num

ber

abso

lute

ly r

equi

site

for

the

dete

rmin

atio

n of

the

unkn

own

quan

titie

s. T

his

prob

lem

ca

n on

ly b

e pr

oper

ly u

nder

take

n w

hen

an a

ppro

xim

ate

know

ledg

e of

the

orbi

t has

bee

n al

read

y at

tain

ed, w

hich

is

afte

rwar

ds to

be

corr

ecte

d so

as

to s

atis

fy a

ll th

e ob

serv

atio

ns in

the

mos

t acc

urat

e m

anne

r po

ssib

le.”

-F

rom

The

ory

of th

e M

otio

n of

the

Hea

venl

y B

odie

s M

ovin

g ab

out

the

Sun

in C

onic

Sec

tions

, Gau

ss, 1

809.

6Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Est

imat

ion

Bas

ics

Est

imat

ion

Bas

ics

•Pr

oble

m s

tate

men

t–

Obs

erva

tion

rand

om v

aria

ble

X

(Giv

en)

–T

arge

t ran

dom

var

iabl

e

Y

(U

nkno

wn)

–Jo

int p

roba

bilit

y de

nsity

f(

x,y)

(Giv

en)

•W

hat i

s th

e be

st e

stim

ate

yopt =

g(x)

whi

ch m

inim

izes

the

expe

cted

mea

n sq

uare

err

orex

pect

ed m

ean

squa

re e

rror

betw

een

yoptan

d y

?

•A

nsw

er: C

ondi

tiona

l Mea

ng(

x)=

E(Y

|X=

x)

•E

stim

ate

g(x)

can

be p

oten

tially

non

linea

r an

d un

avai

labl

e in

clo

sed-

form

.

•W

hen

X a

nd Y

are

join

tly G

auss

ian

g(x)

is li

near

.

•W

hat i

s th

e be

st li

near

est

imat

e yli

n =W

xw

hich

min

imiz

es

the

mea

n sq

uare

err

or ?

7Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Wie

ner

Filt

erW

ien

er F

ilter

1940

•W

iene

r-H

opf

Solu

tion:

W =

RY

X (

Rxx

)-1

•In

volv

es m

atri

x in

vers

ion

•A

pplie

s on

ly to

sta

tiona

ry p

roce

sses

•N

ot a

men

able

for

onl

ine

recu

rsiv

e im

plem

enta

tion

.

Sp

an

(X)

Y

Yli

n

8Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Kal

man

Kal

man

Filt

erF

ilter

1960

•T

he e

stim

ate

can

be o

btai

ned

recu

rsiv

ely.

•C

an b

e ap

plie

d to

non

-sta

tiona

ry p

roce

sses

.•

If m

easu

rem

ent n

oise

and

pro

cess

noi

se a

re w

hite

an

d G

auss

ian,

then

the

filte

r is

“op

timal

”.–

Min

imum

var

ianc

e un

bias

ed e

stim

ate

•In

the

gene

ral c

ase,

the

Kal

man

filte

r is

the

best

lin

ear

estim

ator

am

ong

all l

inea

r es

timat

ors.

•It

can

be

exte

nded

to n

onlin

ear

estim

atio

n –

exte

nded

Kal

man

filt

er (

EK

F)

1 11

11

kk

kk

kk

kk

xA

xw

yM

xv

+ ++

++

=+

=+

Pro

cess

Mod

el:

Mea

sure

men

t Mod

el:

ST

AT

E-S

PA

CE

MO

DE

L

Page 3: Optimal Estimate of System State Lecture 1 Observed ...users.encs.concordia.ca/~ymzhang/courses/ESIF/ESIF_Lect1_intro.pdf · Mathematical Techniques in Multisensor Data Fusion, Artech

9Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Wh

y D

o W

e N

eed

KF

?W

hy

Do

We

Nee

d K

F ?

A p

redi

ctio

n/co

rrec

tion

algo

rithm

use

d fo

r st

ate

estim

atio

nun

der

stoc

hast

icen

viro

nmen

t

KF

is u

sed

for:

Est

imat

ion

of a

val

ue w

here

mea

sure

men

ts a

re m

ade

in n

oisy

env

ironm

ents

(no

isy

inpu

t is

conv

erte

d to

less

no

isy

data

)

Pre

dict

ion

of ti

me

vary

ing

varia

ble

base

d on

a l

inea

r st

ate

mod

el (

base

d on

pre

viou

s m

easu

rem

ents

)

Dat

a fu

sion

(use

d fo

r ob

tain

ing

an e

stim

ate

of a

val

ue

who

se m

easu

rem

ent i

s ob

tain

ed fr

om d

iffer

ent s

ourc

es)

10Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Ho

w D

oes

It D

o It

?H

ow

Do

es It

Do

It ?

•T

he K

alm

anfil

ter

itera

tes

betw

een

two

step

s–

Tim

e U

pdat

e (P

redi

ct)

•P

roje

ct c

urre

nt (

k-1)

sta

te a

nd c

ovar

ianc

e fo

rwar

d to

the

next

tim

e st

ep (

k), t

hat i

s,

com

pute

the

next

a p

riori

estim

ates

.–

Mea

sure

men

t Upd

ate

(Cor

rect

)•

Upd

ate

the

a pr

iori

quan

titie

s us

ing

curr

ent

time

step

noi

sy m

easu

rem

ents

, tha

t is,

co

mpu

te th

e a

post

erio

ri es

timat

es.

•C

hoos

e K

k(K

alm

anga

in)

to m

inim

ize

erro

r co

varia

nce

()

ˆˆ

ˆk

kk

kk

kx

xK

zM

x−

−=

+−

11Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Co

mm

on

Ap

plic

atio

ns

Co

mm

on

Ap

plic

atio

ns

•T

rack

ing

mis

sile

s –

Tra

ckin

g m

ovin

g ob

ject

s•

GP

S•

Com

pute

r vi

sion

–E

xtra

ctin

g lip

mot

ion

from

vid

eo•

Dat

a fu

sion

/inte

grat

ion

–In

tegr

atio

n of

spa

tio-t

empo

ral v

ideo

seg

men

ts•

Rob

otic

s–

Rob

ust e

stim

atio

n an

d se

nsor

dat

a no

ise

redu

ctio

n•

Sta

te a

nd p

aram

eter

est

imat

ion

for

mon

itorin

g,

faul

t dia

gnos

is a

nd c

ontr

ol•

Dat

a sm

ooth

ing

and

curv

e fit

ting

•…

12Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Lin

ear

Lin

ear

Pla

nt

and

Mea

sure

men

t M

od

els

Pla

nt

and

Mea

sure

men

t M

od

els

Page 4: Optimal Estimate of System State Lecture 1 Observed ...users.encs.concordia.ca/~ymzhang/courses/ESIF/ESIF_Lect1_intro.pdf · Mathematical Techniques in Multisensor Data Fusion, Artech

13Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Est

imat

ion

Pro

ble

m

Est

imat

ion

Pro

ble

m --

Ob

ject

ive

Ob

ject

ive

Ob

ject

ive:

To

find

an e

stim

atio

n of

the

nst

ate

vect

or

rep

rese

nted

by

, a

line

ar fu

nctio

n of

th

e m

easu

rem

ents

th

at m

inim

izes

the

wei

ghte

d m

ean-

squa

red

erro

r

whe

re M

is a

ny s

ymm

etric

non

nega

tive-

defin

ite

wei

ghtin

g m

atrix

.

So

luti

on

?

ˆˆ

[]

[]

Tk

kk

kE

xx

xx

−−

kxˆ kx

,...,

,i

kz

z

14Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Th

e D

iscr

ete

Th

e D

iscr

ete

-- Tim

e T

ime

Kal

man

Kal

man

Filt

erF

ilter

A r

epre

sent

atio

n fr

om G

A01

15Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Th

e D

iscr

ete

Th

e D

iscr

ete

-- Tim

e T

ime

Kal

man

Kal

man

Filt

erF

ilter

A r

epre

sent

atio

n fr

om B

LK01

16Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Th

e D

iscr

ete

Th

e D

iscr

ete

-- Tim

e T

ime

Kal

man

Kal

man

Filt

erF

ilter

Page 5: Optimal Estimate of System State Lecture 1 Observed ...users.encs.concordia.ca/~ymzhang/courses/ESIF/ESIF_Lect1_intro.pdf · Mathematical Techniques in Multisensor Data Fusion, Artech

17Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Th

e T

he

Kal

man

Kal

man

Filt

er C

ycle

Filt

er C

ycle

Tim

e U

pdat

e(P

redi

ct)

Mea

sure

men

t Upd

ate

(Cor

rect

)

Adj

usts

the

curr

ent s

tate

est

imat

e

Pro

ject

s th

e cu

rren

t sta

te e

stim

ate

18Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Seq

uen

ce o

f O

per

atio

n o

f D

KF

Seq

uen

ce o

f O

per

atio

n o

f D

KF

19Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

A S

imp

le E

xam

ple

A

Sim

ple

Exa

mp

le --

Tra

ckin

gT

rack

ing

1k

kk

xF

xw

+=

+

kk

kz

Hx

v=

+

1

10

0

10

00

00

1

00

01

xx

xx

ky

y

yy

kk

pp

tp

pw

pp

t

pp

δ

δ

+

=+

10

00

00

10

x xx

ky

yk

yk

p pp

vp

p

p

=

+

X

Y(

,,

,)

kx

xy

yx

pp

pp

20Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Act

ual

Est

imat

e

A S

imp

le E

xam

ple

A

Sim

ple

Exa

mp

le ––

Est

imat

ed v

s. A

ctu

alE

stim

ated

vs.

Act

ual

Page 6: Optimal Estimate of System State Lecture 1 Observed ...users.encs.concordia.ca/~ymzhang/courses/ESIF/ESIF_Lect1_intro.pdf · Mathematical Techniques in Multisensor Data Fusion, Artech

21Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Th

e D

iscr

ete

Th

e D

iscr

ete

-- Tim

e T

ime

Kal

man

Kal

man

Filt

erF

ilter

An

Exa

mpl

e

22Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Th

e D

iscr

ete

Th

e D

iscr

ete

-- Tim

e T

ime

Kal

man

Kal

man

Filt

erF

ilter

An

Exa

mpl

e (c

ont)

–m

ore

exam

ples

in o

verh

eade

ran

d M

AT

LAB

23Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Th

e D

iscr

ete

Th

e D

iscr

ete

-- Tim

e T

ime

Kal

man

Kal

man

Filt

erF

ilter

Mea

s. u

pdat

e (c

orre

ctio

n)

Tim

e up

date

(p

redi

ctio

n)

24Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Kal

man

Kal

man

Filt

er v

s. R

LS

Filt

er v

s. R

LS

-R

LS(f

rom

F7-

1: S

yste

m Id

entif

icat

ion

cour

se)

Page 7: Optimal Estimate of System State Lecture 1 Observed ...users.encs.concordia.ca/~ymzhang/courses/ESIF/ESIF_Lect1_intro.pdf · Mathematical Techniques in Multisensor Data Fusion, Artech

25Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Kal

man

Kal

man

Filt

er v

s. R

LS

Filt

er v

s. R

LS

-K

F fo

r pa

ram

eter

est

imat

ion

(fro

m F

7-1:

Sys

tem

Iden

tific

atio

n)

26Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Th

e C

on

tin

uo

us

Th

e C

on

tin

uo

us --

Tim

e T

ime

Kal

man

Kal

man

Filt

erF

ilter

27Le

ctur

e 1

Lect

ure

1Le

ctur

e N

otes

on

Lect

ure

Not

es o

n E

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

nE

stim

atio

n a

nd

Sen

sor

Info

rmat

ion

Fu

sio

n, b

y Y

. M. Z

hang

(A

UE

), b

y Y

. M. Z

hang

(A

UE

)

Tex

tbo

oks

Rec

om

men

ded

tex

tbo

oks

Rec

om

men

ded

tex

tbo

oks

[BL

K01

] Yaa

kov

Bar

-Sha

lom

, X. R

ong

Li, T

hiag

alin

gam

K

iruba

raja

n, E

stim

atio

n w

ith A

pplic

atio

ns to

Tra

ckin

g an

d N

avig

atio

n, W

iley-

Inte

rsci

ence

, Jun

e 8,

200

1, IS

BN

: 04

7141

655X

.[H

M04

] Dav

id L

. Hal

l, S

onya

A. H

. McM

ulle

n, M

athe

mat

ical

T

echn

ique

s in

Mul

tisen

sor

Dat

a F

usio

n, A

rtec

hH

ouse

P

ublis

hers

; 2nd

edi

tion,

Mar

ch 1

, 200

4, IS

BN

: 158

0533

353.

Op

tio

nal

tex

tbo

oks

/ re

fere

nce

sO

pti

on

al t

extb

oo

ks /

refe

ren

ces

[GA

01]

Moh

inde

rS

. Gre

wal

, Ang

us P

. And

rew

s, K

alm

anF

ilter

ing:

The

ory

and

Pra

ctic

e U

sing

MA

TLA

B, 2

nd E

ditio

n,

Wile

y-In

ters

cien

ce, J

anua

ry 2

001,

ISB

N: 0

-471

-392

54-5

.[H

L01

] Dav

id L

. Hal

l, Ja

mes

Llin

as, H

andb

ook

of M

ultis

enso

r

Dat

a F

usio

n, C

RC

Pre

ss J

une

20, 2

001,

ISB

N: 0

8493

2379

7.28

Lect

ure

1Le

ctur

e 1

Lect

ure

Not

es o

n Le

ctur

e N

otes

on

Est

imat

ion

an

d S

enso

r In

form

atio

n F

usi

on

Est

imat

ion

an

d S

enso

r In

form

atio

n F

usi

on

, by

Y. M

. Zha

ng (

AU

E)

, by

Y. M

. Zha

ng (

AU

E)

Rea

din

gs

The

dis

cret

e-tim

e K

alm

anfil

ter

(DK

F):

•B

LK

01: S

ectio

ns 5

.1-5

.3

•G

A01

: Sec

tions

4.1

-4.2

The

con

tinuo

us-t

ime

Kal

man

filte

r (C

KF

):•

BL

K01

: Sec

tions

9.1

-9.2

•G

A01

: Sec

tions

4.3

The

ext

ende

d K

alm

anfil

ter:

•B

LK

01: S

ectio

ns 1

0.1-

10.3

•G

A01

: Sec

tions

5.1

-5.7

Intr

oduc

tion

to K

alm

anfil

ter:

Wel

ch &

Bis

hop:

Pap

er; S

lides

•M

aybe

ck’s

book

(19

79)_

Cha

pter

1