4. B6 loc Kalman réi rac
25 loc Kalman cp dn bbl toén tng quét dude lug trang tl cia mét qué enh duge mé inh NBs mgt cdch Hs ae theo ene
‘91an ng mbt phucng tinh neu nhidn tuyde tinh nou sau
Te = At cee + Bt tea $e)
a do dee
ak = Hike + tk)
Trong dé w va vl 2 vector bidn ngBu nhién dai dif cho niu né ending va nhidu do oye. 2 bién ngBy niin nay dbc bp vA auge
(98 sil tuin theo phén B& Gauss v6 tung binh =0 vb ma trbn Np ign (covariance) Tn Lt la Q v8 R
ww #N.0)
v=mo.R)
Nu vector trang thx eb keh the Ibn, thi ma tris A 38 cé Kich toe Ibn xn. @ (nx) Hm tr phy thube vo du Knit
Wu V6 Ib vector eb ch thude I |. Vector do dae zc ich thu la m adn ma tba H sé Ih m x Chi ¥ rng ebe me edn QR,
A H cb thé thay d5i theo that glan (ting bude k), hung & dy ching Avge gié si KnBng i
Gin day ta thiy b3/ toda Ipc Kalman ehinh Idd tim oid ti ge hugng va ube dosn cia trang thél x khi ta biét duge sy
bbién thién eda m6 vi ta do duge mét dai lugng z ma phy thude tuyéa tinh vio x.
‘i dy trong bbl todn chuyén dng, ta bit Guge cui Bt thay ai cia vin &Bc,nhung tla é thé do dupe sy thay AB cia v\ te. Kh
4, ci ta ba tim Ib vn t6c wee zen.
uéy ta 98 sit "va Then luge 18 tién nghiém va tien naném ube heen cia gié trix tl th&t diém k. Gié tr én nghiém thu duoe
nl dye va0 mn n8 tng (1), 20918 7 Agung bg ty UNU QUGE aU Kr 68 c8 KR Qua do dac Zhe (2). KI dB sa 35 cL
Ube don ben nghiémn v8 bu ng 18 da
te
tk — Te
Ma dn np oo ca 2s rn et in ue tne cg thd
=
Pe = Blegen")
A= Fleet)
Me ten cd cing ta by oti Ws Ks co ha mn phg nh sau
gt (ce- HAE),
t=
‘én aay ta thy ,K cing chin alpha ma 63 aie thi 6 wen. Prvang tinh (3) cb nahi ib gd rt NEU nghiém ce We wong x
{uge tinh Bing gi tr ten ngnigm cla né va sau C6 tném/cc ci matt ava Veo $316 gia id Go duge v8 gb Wl do Gae vee
ain FE Ey 6 0ay cron 8 it (gala) cba macn lec Kalman
(chu ho 08 ra 1b tam tn no a8 enon K UU nt. TEI wu 6 dy theo Nghia 8 covariance cis sal <8 cla we lugng nu nghiém
cima ay 8 = KC» (24 — HE * FE), sna nn, sing chen nay ev rong ig thd inh P50 08 290
Ini 208 i theo K, ($8 ti r2 dug git K ma twang ing VBI nd Pk I AMS oh
Ky = Po HT (HP, HT + RY*
E.tnay a6: eo ts glan & v8 chin i a8 te cn tim cia mach ioc Kalman trong méi ude dase,
“Tém at mach Ige Kalman bao gm 2 oube : 1- Use don trang thal dn nahi, vd sau 8, 2-dva vio KB cud do a8 MU chin
Ue dean. Ta €& thé tOm UB at hoat €6ng cla mach foe Kalman bing ede Phuong tin 9Gia su Dan 68 06 oi8 tr ube dodm Ehe—0 ta tnd diém (k-1) vd Bit Cuge gid tr dig kniér Led. (Gib ti ban GBu tal thes aiém
0 auge chon £9 = HE % 20). use 66 ban chi vite 180
in trong nian cu dy
it tin Ninn che tinh tosn WW 1 Bn 24 DUBE AB RY an 3 trong Due 2
Measurement Update (“Correct”)
Time Update (“Predict”) aia eae
Kg = PyHT(HP,HT +R)"
(1) Project the state ahead
i can y
. ° (2) Update estimate with measurement 2
(2) Project the error covatianee ahead fy = 4+K,(4-AX)
Py = AP,_,AT+O (3) Update the error covariance
P, = (I= KH)P;
Initial estimates for f,_ and Py_y
Hinn trén tm tt c¢ bn ca mach lgc Kalman. Nhu G8 nhc cén & phibn 3.4 ci kd eda mach loc Kalman I lim thé ndo 68 md
hie ha ede trang tai vi do dae 68 eb dee 2 ova tinh (1) v8 (2) a cb thE 4p dung Kalman,
Dinh nghia covariance matrix
Definitionyedit)
Thvougnout his aril, bekifaces unsubserpted X ard ¥ are used to refer to random vectors, and unboldfaced subserpted X. and
YY ae uses to refer to random scalars.
the entries in he column vector
xX,
sre cnr vis, €2¢h wih inte wating, then he covaance mai isthe mati whose (i) ety is the covasance
Bij = cov( Xi, Xj) = Bl (Xz — pa)(Xj— 49)
were
i = B(X)
isthe exoected value ofthe th erty i the vector X. In other words, we haveEX: = aXe = pad] E[CXs = 14) X2 = pad] vo |
El(Xa = aX ~ pa)] Ell Xa = 2)(Xa—pa)] oo El X2 ~ 2) Xa - Hd]
El(Xu ~ stu)(%i— sy] ECA ~ sha) Ne al] oo EL Xn — sa) Xue = se]
-1
‘The inverse of his mat, D's the inverse covariance matrix, also known 8s the concentration matrix or precision
matrix see precision (statistics). The elements ofthe precision max have an interpretation in terms of partial
arelatons and pomession
Generalization of the variancefedit
‘The dfinton above is equivalant to the matrix equality
B=B|(X- EX) (X- E(X)"|
‘This form can be seen as s generalization of the scalar-valued variance to higher dimensions. Recall that fora scalarvalued
random varable x
a? = var( Xs) = El(X; — E(Xi))"] = El( Xi — E(Xi)) (Xs — EOG))].
Indeed, the entries on the diagonal ofthe covariance mavix are the variances ofeach element of he vector XX.
Correlation matrix(edit)
‘A quantity closely related tothe covariance matrix ie the correlation mati, the matrix of Pearson product-moment corlation
-oaffcients between each ofthe random variables in the random vector XG. which can be writen
pny = (sits) Ey (spent
i
snore EB) sno matix of agonal elamerts ot 20. onal mats ofeach fhe varancas of X)
Ecuivaeny te corlaton maa can be seen 2 the cvaance mati ofthe sandaied random varbis X10 (or = 1
Conflicting nomenclatures and notationsfesit}
[Nomenciatures differ. Some statisticians, following the probabil Wiliam Feller, cal the matnx 2 the variance of the random
vector XX, because tis the natural generalization to higher dimensions of the 1-dimensional variance. Others call it
the covariance matrix, because it's the matrix of covariances between the scalar components of the vector AX. Thus
yar(X) = cov(X) = E [(X — E[X])(X — E[X])"]
cov(X, Y) = E[(X — E[X))(Y — E[¥])"].
‘The var otaton is found in Willam Feller two-volume book An inveduction fo Probabilty Theery and ite Appicatione.* but both
forme are quite andar and there ‘sno ambiguty between them
‘The matric Dots also often called the variance-covariance matrix since the diagonal terms are infact variances.Propertiesjedit]
ne Ex"). Hu=E(X)
vanable and ¥ a random ¢-cimensional variable, the following basic properties apply
where X is a random p-imensional
S=E(XX™) — wet
Diispoetiesenisetnte and emetic
cov(AX +a} = A cov(X) AT
cov(X. Y} = cov(¥,X)!
cov(X, + X2, ¥) = cov(X,, ¥) + cov(X2, ¥)
tpsqmen Var(X + Y) = var(X) + cov(X,Y) + cov(¥, X) + var(¥)
cov(AX + a,BTY +b) = A cov(X,Y)B
cov(X,Y)=0
WX ana "Y are independent or uncorrelated. then
were XX ano Xz are random xt vectors, Yi arandom get voce, 2s a get vector, Dea pst vector ane AL
ane Bare gop matrices
This covariance matic @ useful tool ia many diferent areas. From ta lansformation matic can be derived, called a whitening
Iansformation that alows one to completely decorelate the data or from a diferent point of view, to find an optimal
basis fr representing the data ina compact way #8 s9¢ Rayleigh quotient fra formal proaf and addtional properties of
covariance matrices), This scaled pancipa| components anslsis (PCA) and tne Karbunen-Lotve transform (KL-ransfor)