21.01.2014 Views

Experiences with self-calibrating bundle adjustment - IDB

Experiences with self-calibrating bundle adjustment - IDB

Experiences with self-calibrating bundle adjustment - IDB

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

ACSII-ASP CCNVENTION<br />

I^IASHINGTON, D,C,<br />

FEBRUARY/l4ARcH 1978<br />

PRESEIITED PAPER<br />

EXPERlENCES 1.1ITH SELF-CALIBRATING BUNDLE ADJUST|,IENT<br />

ARMIN GRÜN<br />

INsT I TUTE oF PF]oToGRAMNETRY<br />

TEcHr'.u cAL UNTvERSITy oF iluNlcH<br />

ARC I SSTRASSE 21<br />

D-8000 f]üNcHEN 2<br />

FRG


EXPtRIiNCIS IIITN S i L F - C A L I B R A T I I C B UJIDL t A!JUSTI'1INT<br />

1. Introduction<br />

As a result of the coapensation of systematic ima!e errors mcdern photo!rammel.ic<br />

methods of point deternination have achieved an accuracy level, ,rhich Fakes then<br />

competjtive |,/ith traditjonal g€odetic irethods even ir the fleld of densificaticn<br />

of control netHorks and cadastral survevlng.<br />

Hovever, a successful application of photo_arammetric nethods to hiqh precision<br />

point determination requires a conprehensjve and efficient conrpensation of the<br />

systematic image errors. D!ring the last feI.l years various technics of treatment<br />

of systematic errors have been p.oposed. Thereby !he slfi!ltaneo!s <strong>self</strong>-calibration<br />

techniq!e by using additional parameters ltsauer /3,/, prcüa /3/, Ebret /t/,<br />

clin ,/13/) becdne the ncst promising procedure, rdther than the ä-priori<br />

field correction lrtpfe! /16/) and the conpensarion by flight arrangenent<br />

(rhcücs /23/).<br />

iire advanta!es of s e I f - c a I i b r a t 1 o n are quite evidentl<br />

ihe extension of the mathenatical mod€l !f <strong>bundle</strong> <strong>adjustment</strong> allows for the<br />

c0fnpensation of the actunl systematic errors, whjch may vary f.orTt project to<br />

pr0Ject; tnere is no ddlitional efforr necessary for es!abljshing and ftyin! a<br />

test field,<br />

for supplementnry ima!e point |rreasurefxents and cofipLrtatjons.<br />

8y applying the methods of test field calibratlon and <strong>self</strong>-calibratior to a part<br />

of the '0bersch!/dben' tes! block the superiority<br />

rs derf0nstrated. iloreover, !re results of sorne other practical<br />

projects are refered<br />

and conpared ,{jth the theorelical<br />

test<br />

of s e I f - c a I i b r a t i o n technique<br />

accurdcy e^pecta!ions of <strong>bundle</strong> adjustllokever,<br />

by !sjn!<br />

the s e I f - c a I i b r a t i o n techniq!e some prob I ens arise \ritn respect<br />

to the choice and handling of the aiditional<br />

parameters. lniending the compensation<br />

of the rhole systematic errors requires d hignly general and flexjbte<br />

parameter<br />

nodel. This may increase the danger of overparametri:ation. So one has tc<br />

take cdre of the stability of th€ system of equations (e.9. normal eq!ations) jn<br />

eacn partrcular case because of changing geometrical condjtions jn Cjfferent Droiects<br />

(control point distrjb!tion, fli!ht directr'on, image oveftap, image point<br />

djstribution, shape of the terrain). Therefore the q!estion of a Droper choice of<br />

tyDe and nu rber cf additiondt pardneters ls disc!ssed and suitabte statisti.dl<br />

!ests are pf0posed to clreck the sjqnificaice<br />

of the conputed Dararneter amounts,<br />

.rurte r3cently piotograrnetric research Lrnderrakes !r--at efforts io stabirize the<br />

reliability of protoJranretric systems. It is !/r'dety known thar grcss errors are<br />

rery hdrdly to ietect sjnDte r€t1oCs after<br />

"ith<br />

th€ perforrßan.e of brnJle adjust_<br />

nent. io get mostly objective decisrons for the rejection of observations re_<br />

ference is nad€ !o B.rdrC. r retiability theory (i.iri. /1,/, /2/ t.<br />

All ccmDLrtaiions presented Ii thjs paper !ere perforFed djth the bLrndle pro_cran<br />

il80P ( 'Iu'rr'ch Cuidle 0rientation . jth Adcitional para,neters,), !lhich the authcr<br />

developeij i, I9ll/Ja at the Iistitute of photo!rannetry, Technical Universr.ty of<br />

llirni ch (ar:ir ./tr-l,).


2<br />

In the follo\{ing chapter the rTrathematical model of this program is described<br />

together !lith some remarks or gross error detectr0n.<br />

2.<br />

The b!ndle program li80P. FLrnctional and stochastic model.<br />

The detection of gross errors<br />

The linearized mathematical model of the s e I f - c a I i b r a t i n g bundl e progran<br />

yields the follov/ing:<br />

Z<br />

+ A2 dx' +<br />

I dx P<br />

Ldz<br />

tB r<br />

]Z ;<br />

0;<br />

P8 (=r )<br />

Pz<br />

(1)<br />

,2, ,P<br />

Yectors of residuals of image coordinates<br />

-h.r,-t .... - ...rrindtes<br />

, äddi tional<br />

dt, dz -<br />

Correction yectors of point coordinates,<br />

coordinates, j riage pdrameters, additional<br />

control<br />

poi nt<br />

A1' 42, 43, A1<br />

= C 1.r-r ' 1. rrp'.icierrS<br />

vectors of corstants of image coorainares. aodi! oral<br />

..n!.11 ^. in. . ,, -.rinates<br />

B''Z''P<br />

= L/r'eiqht matrices (diaqonal form) for ima!e coordinates<br />

additional<br />

pararneters, control point coordinates<br />

= ldentity natrix<br />

The matrix Pp allows to inirod!ce ä-priofi knawn accuracy standards of the control<br />

point coordinates; the snfie dDplies uith resDect to the matrix PZ and the<br />

ddditional parameters.<br />

Under certain circunstances and <strong>with</strong> certain pararietef sets the treatment of<br />

sJrtemetic errors as free !nKno\rns (PZ = 0) nay lead to renarkable deteriorations<br />

of the conoition of the system of nornral equations. lherefore it is expedient<br />

to introd!ce PZ+ 0, thus protectrng against överparanetrization.<br />

(cltn /13/).<br />

l,lith<br />

Pp + 0 He obtain a weighted minimal fitting (l/lith respect to a 3-dimensional<br />

sr'uildrity transfarmation) of the p h o t o q r a nr m e t r i c points anto the control pojnts<br />

Collecting the matrices and veciors of (1) jn the foll<br />

owi ng<br />

At Az A: A,t<br />

000I<br />

0I00<br />

)<br />

P<br />

(<br />

Pa o<br />

a Pz<br />

00<br />

a\<br />

ol<br />

x<br />

1,u<br />

= {dx'<br />

"T<br />

-<br />

t)<br />

,<br />

t,n<br />

(IBT<br />

.tr<br />

(r)<br />

1,n


the system (l) farnally results in<br />

v = Ax - I i P (2)<br />

and by using the operator of expectati.n t and th: operatcr ! vhich leads tc the<br />

,dridrcer<br />

d d c va. e1!es ?e 9et<br />

At r - rll F{.(l =:<br />

^"-l'<br />

c(r ) = d: P ', (.^ = standard deviation of unit (3)<br />

weight to be e5timated)<br />

E(v) = 0<br />

Thereby the correctlon vectors of the grorrnd point coordinates Cx and of th€<br />

orientaticn<br />

parameters dt are all!ays introduced as free !nknowns,<br />

System (3) is cdlled a generalizen Gduß-Markov llodel (.{ocl /1s/). For known<br />

e^pected values tr- and for P7 + 0 i t may be interpreted as Reqression l'lodel<br />

\:o:h.;:/), far Pa + I, E(lr)= 0 as Collocatior, lladel (xble? /t/).<br />

Using Pp - 0 the system becomes srf!uldr. If ihe geodetic control point coordi_<br />

naies are introduced as errorfree quantiti€s (diaq (Pp)r -), like it is often<br />

required in oractice, and if the additronal parameters are tred!ed as free unknor'rns<br />

(PZ : 0), the ordinary GauS-llarkov Model, i.e. the usual case of Adjustment<br />

of 0bservations is obtaineC.<br />

Llith the set of additional paraireters dz fron {l) we are fLrlly flexible, there<br />

rs no res!riction concerning th€ir type and number. Tiis enäbles us to jntroduce<br />

blockinvarjant<br />

parameters, just as parameters l1hich belonq to a sinqle strip,<br />

to a ce"roin grouo of |a jes or evel to a silgle<br />

All practical<br />

r_aq".<br />

comoLrtations presented in this Daper were performed by applyinq<br />

b I o c k i n v d r i a n t .r a r a n € t € r s only.<br />

Besides a sophisiicated funcrional and stochastic model a high precision pretending<br />

and operational <strong>bundle</strong> orogrdm requires the aLrtomatical determination of<br />

the approxinate val!es of the aaj!sLment dnd the autonatical detection and localisation<br />

of gross errors to cone out <strong>with</strong> a nininrun of program runnjngs and to<br />

oatain re 1i abl e resul ts.<br />

0ne imporiant Droblem consists in the procedLrre of detection of small qross<br />

errors. A decision concerninq the rejection of an obser!dtion rs forndlly<br />

uprn the dndlysis of the resjdudls of adj!stments. If !e try to objectify<br />

based<br />

decisions Lve have to refer to st.tistical concepts. Il r/as ,rrrdc /1/, /2/, rtho<br />

de!elooec e starisr.cal corcept of e.ror derpctron, !.r'c s dl.ead/ dip'ied<br />

-o<br />

qeodetic netvork olanning and adjustflent (tca"Ca /2/, Bailncf;n /4,/, larita /2a/),<br />

Furthermore some inv€sti!ations vrere performed concerning photogrammet.lc systeris<br />

- the block adjustnent of inderendent nroCels (.1örstr?, /i1/, !!6Leüt /1!/).<br />

!a:rid'E theory nust be regirded as a "reliability" theory. This ternr ha! to be<br />

seperat€d very strictly<br />

from the term riacc!racyii. A meäsurjn! system may be<br />

accurata, \,rithoLrt bejn! reliable at all.r,le consiCer a system as not reliable',<br />

lf there is a risk that one cannot letec! mocel errors, for exdriole gross errors<br />

or systemetjc error!, \rithin..€rtain statistical securjty. For example, if in<br />

b!ndle ddlusiment tho p\otoqrtsmnetric coordinates of a ground poin! have to Je<br />

deri ved fror only ivo image point neasurefrents (tro ray5), then these coordinates<br />

may be of moderate acc!racy, i./hereas their relidbiliry is !rorse, beca!se d !ros5<br />

cur


error in those image coordinätes, whi.n belon! to the epipolar plane, cannot be<br />

:,rdrda's theory ma,v be equally apDlied to theoretical reliability studies of<br />

bLrndle adjustnent systefis as to the detection and localization<br />

of qross errors<br />

0n€ resLrl t af ?aa!d3'<br />

mÄrn. fho .hp.l inn ^f<br />

For that the tes t cri<br />

.<br />

'nvesticatio.s is rFe 'd"ta-sroopino' !ecl^nioue. hlicl'<br />

the inCividual resiCuals for gross error detection.<br />

-vj<br />

'lvivi<br />

q-rvr ! j are rne vdrrdrces o';i a.d oo is tre standard oevration of unit Lergtt.<br />

(4)<br />

i s proposed, vlh€reby the<br />

ro< i., ,l c ,,-,, \., J.e oe.]0Ied urr!n vt;<br />

fo. Ur, i.e. tqe reo or krere Lhe r"ll hyoorhesis (the<br />

s nornally distrib!teJ) is valid is then<br />

- F I , 2 \ 1 - , o , I . co J = \ . = Fl 2rt-,o,',ooy (5)<br />

vhere do is the type l.error siz€ {.0 - P(l)) for the test criterion Vi.<br />

Tire essential restrictjon for lar!e systems concernrng the conrputinq anount in<br />

(4) are the values Qv.v.. Thereby t\.io oossjbilities for a !./ay cut arise. The first<br />

is to modify the photogranrmetric systems in d \,lay which allows to use equal vat!es<br />

for al I Qvivj sucn that<br />

-p vi<br />

(6)<br />

This nay not be feasible jn most of the practj€al<br />

pfojects.<br />

The other possibility is to cornnLrte the dctudl values for Q",u tlen on€ has to<br />

subdivide the Nhole system into subsystems to reduce compLrtati;n tin,e. This is<br />

also required from a theoretjcal<br />

point of view, oecduse ihe derjvation of (4) is<br />

Ldsed upoir tlre presLmption that only one _aross error appears, so that rith<br />

!se of subsysiens !Je are closer to theory,<br />

The procedr,rre öf coaputation of approxi ate val!es (coordinates of qround points<br />

carameters of exterior orr'entatlon) for the <strong>bundle</strong> adjustnrent in ll80P (relativ<br />

model ori€ntation, strip forflration , absol!te orientdtion of strips) is a yrellsult€d<br />

tecfnique for the introd!ctron<br />

the<br />

of subsJstems for a p.eliminary cnecking,<br />

l/hereas the fjnal ciecking has to be oerforfired <strong>with</strong> <strong>bundle</strong> subsystens.<br />

l' ',<br />

r^, cn.hir.j..-o4 s-1-rs-- prlr. ,---ction<br />

are nol i'-<br />

cluCed in [180P, because sone further investiqations<br />

have to be performed,<br />

:Lcording ta other ccncepts as published by ?ote /19/ drd 9ezt":vg /5/.<br />

Finally it rnust be stated, that the Dhotogramnretri c systefis, nowddafs L,sed for<br />

highly precise point determinatior as occuring in densifjcation of control networks<br />

or caaastral surveyin!, nostly are qui te !/ell sLrited for gross error<br />

detection and lokalization.<br />

This is due to the great nunrber and approprjat€<br />

distribution of image points, to the dpplicörion of 60i sidelrp and to the !se<br />

of groLrps of pojnts i2,3 or even rore targets per qroup) instead of determinin!<br />

isolated, \.reak poi nts.


Conrpali9or of ä-Driori test field calibrdtion and s e I f - c a I i b r a t i o n<br />

It was neniiofed in tte introduction that block adjustnent Hith s e I f - c a I i b r a t i o n<br />

is consjdered as rhe nost effective<br />

ard economic nethod for the conpensation of<br />

systenetjc jriaqe errors, This ho'rever, is not !enerally acceDted dno above all<br />

jn dispute ir lJest Gerrnany. EsDecially the test field calibratron<br />

persation b"v ili!ht arranqefient L.rere proDosed as competjtive nethods (i!pr-r! /.tdl,<br />

r.!eLskdEen /J7/, ritont, ,l2J/). Ihe.eby the advanidge of the tes! fjeld catibration<br />

consists in the possibilitT<br />

tion terms than it may be feasible by s e I f - c a I i b r a t i o n .<br />

of applyjn! a hi!her deqree of inage corre.-<br />

To get an iCea of the efficiency of possibly alternative nrethods as ä-priori<br />

test field calibratian and sirnultareous s e I f - c a I i 5 r a t i o n both methods !rere<br />

appljed to a comnon practical project. For this purpose d part of the ,0berschwaben"<br />

- testblock, as it lras used jn rlaer /t/ and Grü4 /1;/, 'las selected.<br />

It ls a bl0ck consisting of 4 strips \rith 26 photos each (totat 101 photos),<br />

6Ci1 forward overlap and 20; sidelap. Soth yrere como!ted, the !Jide dn!le (HA)<br />

and tne;!per |lid€ angle (SliA) flignt.<br />

The test field fli!ht Darameters arei<br />

iiA: RirK 15/23,4<br />

strips <strong>with</strong> 3 photooraphs each in preftight<br />

and costfliqht, image scale 1:10,,100<br />

SllA: RllK 3.5/23, L strip <strong>with</strong> 3 photoqraphs in nreftiqht and<br />

4 strips !]ith 3 photoqraphs each in postflight, inraSe scale 1j10,400<br />

The polynomial coefficieits<br />

for the Sl.lA systematic errors,<br />

computed as mean.,<br />

irofl or€ and post test field fljqnts (test 3rea ,,Rher'dt,') Nere taken over from<br />

tfe Irstitute of Plrotogrammetry, University of B.nn (trare jsnc:,1ek /17/).<br />

All other compLrtations \./ere exec!ted by the author. The coefficients for the t{A<br />

syst€natrc errors were oL\tained by apelyjng an orthogonal bivaridte polynonial<br />

for a reqular 5 x 5 inäqe point screen (see chapter 5)t the results from preand<br />

Dostfljqht are sho\rn separately. The results of <strong>self</strong>-catjbrdrron were ootained<br />

by applyinq a blockinvariant orthogonal bivariate potynom for d regutar<br />

I r 3 inaqe point screen lEbse! /3,/), due to the jmage point distfibutr'on<br />

0berschr/aben test bl ock fl i qht.<br />

of the<br />

Fil<br />

tl<br />

pranrder ry<br />

Fiqure ldi Co.trol point distribution of rfe<br />

(strips nos. 5,7.9,11), i = bridqi<br />

'tberscir!aben' tes t<br />

.n rll(t:nra ,n ,i


Table I contains the resul ts. eoth |'€re<br />

(P4) control point dlstribution as it is<br />

ve ry dense (P1)<br />

in Fj!ure 1d.<br />

and a very sparse<br />

r0<br />

rao<br />

Fi !aure 1b: Test<br />

Iach<br />

of 3<br />

area 'Rheidt".<br />

pornt represents a gr0up<br />

gr0un0 p01nrs.<br />

550<br />

Tdble 1:<br />

Cofioarison of d-priori test field calibration and s e I f - c a I i b r a t I o n<br />

Test block<br />

'0berschwaben" (strips nos.5,7,9,1.1), 104 photos,<br />

20X sideldp, imaqe scale - l:28,300.<br />

Cariera<br />

Coitrol<br />

distrrbutior<br />

wi thout correction<br />

of systemati c<br />

<strong>self</strong>-calibration<br />

d /., ,, !,<br />

o n,I<br />

l"l t'.1 [,'] t,.l t,.l<br />

,z<br />

t"'l<br />

<strong>with</strong> test field<br />

prefliqht<br />

I<br />

cal ibration<br />

postfli-oht<br />

oo ux,Y uz<br />

l"'l [,'] t,-l<br />

P1<br />

5.3 8.8 15.8 : .3 4 .2 L2.2 4.7 4.3 8.8 9.4 17.1 lB.2<br />

acy i nprovenent 1.6 1.7 1.3 Ll L.l 1.0 1.9 1.9 0.!<br />

ii | 4.r 24 e '!4 2 1.2 8,0 18.9 3.6 3.5 22. t 22.2 31.4 56.3<br />

s;l<br />

11<br />

sll<br />

Accuracy r nprovenent 1.3 3.1 2.3 1,1 1.2 1.1 l.l 1.1 0.3<br />

I s.: to.: t;.0<br />

|<br />

6.0 9.8 14.3 i.2 12.6 15.7<br />

Acc!racy ii:provefien t L-L L.] 1.1<br />

I s. e :e.q q,1.2 5.9 t!..1 r7.6<br />

Accuracy i nprovement 1.1 2.4 2.5 1.0 1.4 1.6<br />

Analysing tl're results we recotni.e significantly different accuracy levels in<br />

\'/ide angle (llA) and super wid€ anqle (5liA) fliqhts.<br />

iloliever, th€ accuracy imorovenent<br />

by s e I f - c a I i b r a t i o . is simjlar<br />

in botir cases. A decreasing control distrjbu-<br />

!ior results in an increasinq accuracy irnprovenent (up to a factor 3.1) This ls<br />

due to the fict<br />

that -.he propaqation of systenatlc errors is the rnore Lrnfavourable<br />

the le5s control points are Lrsed. Thus s e I f - c a I i b r a t i o n yields I arger a.curacy<br />

imDrovements in the case of poorer control distribLrtions. The;-priori test<br />

field calibratr'on acts quite Ciff€rently. ljjthin the SllA fli!ht r'/e obtain a re-<br />

'iarkable improvem,ent ccnrpared ,iith the ccrioutation frithoLlt correctian of systefiatjc<br />

errors. This is nrt true in the case of llA flight, fere an inpr.venent {as<br />

achieved only 5y ti.e pre'_liqht and the sparse control distrjbution (iactor 1.4<br />

in heiqht). vhereds tne postfliqht calibration larqely deteriorates the results.


7<br />

This clearly sholJs the unreliability of the ä-oriori test field calibration<br />

technjque. Though test fi€ld cdlibration oDens the possibiljty of the applicatjon<br />

of highly developed inrage correction terms this advanta!e m!st nct be<br />

overvalued. I' decisive dlsadvanta!e results from the fact that lhe sJ/stematrc<br />

errors can chenge !ihen the act!al pr.ject fli!it is perforned. This may lead to<br />

a remarkable insec!rity ,,rher apDlyin! test field calibration. There is no real<br />

indicator that perflits lhe conclusion whether a test field calibration is representative<br />

for the actual project or not.0nly s e I f - c a I i b r a t i n I block adj!stment<br />

is able !o coripensate the actLrdl systematic errors kith sufficient reliability.<br />

Therefore the use of additional parameters in a modern <strong>bundle</strong> adjustnent progran<br />

is an j|ldi:pensab]e demand.<br />

4. The accuracy potentjal of s e 1 f - c a I i b r a t i n q <strong>bundle</strong> <strong>adjustment</strong>. Theoretical<br />

expectations and oracti cal resLrl ts<br />

Asking for the theoretical<br />

dccuracy expectations of an actual photogrammetric<br />

block requires the determinatioi of the standard deviations of the point coordinates.<br />

Generally thjs is dn expensive paocedure beca!se tt presu'nes the inversion<br />

of the matrix of the nornal equatjons. Ho,rever, photogrammetric point<br />

d"-".".r,-j i. r,.(p - ^a.-rltstton sets.<br />

iherefore th€re is no need to coflrp!te the accul.acy exoectations for each project<br />

ane\r, but we can lrse accuracy moCels, which 1,rere derjved frorn synthetic<br />

da!4. In a.ecent pJblrca-ro1 E.aer) t?ack, lc.Lbe?: I nave p.ese.ted sucr<br />

dccuracy models for bLrndle adjustnent, \,rhich !Jill be appljed to the follo,'{ing<br />

inyestigations.<br />

If the efficiency<br />

of s e I f - c a I i b r a t i n g bunCle d d j Lr s t L , e n t rr a s to be checked we have<br />

to compare rts.esults,iith those theoretical öccLrracy nodel:. This comparjson<br />

Coes nol yield e(act cri lerions,<br />

becaLrse sometimes rerial"kable differences betreef<br />

theoretical block conditions dnd practical test circumstances irith respect to<br />

image point Cistribution,<br />

inage overlap, position of control points, do occ!r.<br />

l!evertheless the aoplication oi those accuracy nodels yields d good saanCard of<br />

conrparjson to check the practically<br />

obtained accrrracy values,<br />

Table 2: Slock specifications<br />

of test blocks<br />

Block<br />

pnotoqra0ns<br />

0verl ap<br />

plan.lhei!ht<br />

ol an.I hei ght<br />

testblocl (Pl<br />

itA + st,iA<br />

t04 1 : 28 ,300 60 20 3l 17 225 111<br />

"Steinbergen"<br />

top.height block 109 1 8,000 60-90 50 2i 55<br />

''i,loosach'<br />

cadastral bl ock 93 1: 3,300 6C 60 25 10 113<br />

cadastral block Lq I '1 , ,101 7C 60 10 tl 15


ln the followinq the results of sone blocks are presented, adjLrsted tl,lice vrith<br />

blockinveriant additiondl parameters and wtthout addrtronal paraneters. The aajn<br />

redson oi this representaticn is to sholl the extent of the accuracy iriprovement<br />

\{nen appl_vin! sinrultaneo!s s e I f - c a I i b r a r i o n technique.<br />

Tabl€ 2 incicates the block specifications,<br />

lable 3r Results of p h o t o g r a n fir e t r i c accuracy tests<br />

Block<br />

oo ux,Y rz<br />

t,'l<br />

ith<br />

while Table 3 sho,r'Js the results.<br />

ddd.pdr.<br />

Appr0x.th€oret.<br />

accuracy €xoect<br />

lu "l I, rttl r,;t t;;i t";r l"'l [u']<br />

PI 5.3 8.8 14.8 3.3 5.2 12.2 3.0 a.l<br />

P1<br />

5 'riA<br />

Accuracv irrrov€n€nt 1.6 1.7 1.1<br />

8.2 16.3 17.6 6.0 9.8 14.3 5.4 5.2<br />

AccLrracy improverien t t.4 1.7 r.2<br />

5.6 - 14.3 5.4 - 8.0 - 5.9<br />

Accuracy r nrpr0vemen! 1.0 - 1.3<br />

5.0 10.6 4.2 5.1 2.6<br />

AccLrracy i mproYement 1.2 2.1<br />

4.2 I .7 4.0 1.8 2.1<br />

Accuracy inproveoent 1.1 1.5<br />

These results allo!] some interesting<br />

conclusions.<br />

First we have to state th€ large acc!racy improvements, obtained by <strong>self</strong>-calib.aiion<br />

techniq!e (!p to fdctor 2.1). As jt f/as nentjoned alreadv ln the fore-<br />

!oing chapter, the extent of inrorovement depends on the con!rol distribLrtion,<br />

i,e. the lar!es! improvenent usually is achjeved lrith a very sparse control distribution.0n<br />

the other hand the tneoreticdl accurdcv e(Dectalions are not fnet<br />

completely. Firstly<br />

this can be explaineC bv the fact that the enrployeC theoretical<br />

.ccuracy rnodels do not correspond perfectly !./ith the practical<br />

arrangements.<br />

Arother djsturbinq effect is exerted by the i ..-e..rr4 1..r.r. ip


9o t'loL Lhe corcl Jr'o<br />

is allo.eL<br />

deformations are to ir!rute tc the<br />

cadastre block M00SACH<br />

that wi th great<br />

!eodeti c point<br />

n..h:hi I i fw thr


10<br />

satisfy this clajm b!t there is no secLrrity at all that these model concepts of<br />

systematic defornations cofrespond ,rith the actual sy!ternatic inflLrences in any<br />

practi cal pro ject.<br />

Therefore, tbaep /9/ !lent another rjay, asking for oarameters, which cornpensate<br />

the total systenatic effect in a presumed ind!e fornt screen rrithoLrt expldining<br />

the errar sour€es. He developed a bivariate<br />

orthogonal oaranreter set in relation<br />

to a regula. l,{J image poinl distribution. lhereby the orthogonality is not only<br />

valrd l{ith respect to the addrtional parameters themselves but also in relation<br />

to the elements of the extericr orientation.<br />

In /1i/ the author proposed a redLrced third de!ree polynomial, Hhich fits to the<br />

Jrain conponents of the systenatic deformation's model concepts,0ther arrangeirents<br />

are found in Pduer /3/, Salneape?c a.o. /21/, schtt /22/.<br />

polynomials have prcved to be a suitable mocel for st/stenatic imaqe err0rsi<br />

the treatfient of polynomials is cansidered fron a more !eneral point of vie$.<br />

ihereby the favourable pl.ooerty of bivariate polynomials nust be emDhasi.ed:<br />

They dllolv for the compensation of the t.!dl systeflaLic effect in all points<br />

oi d presurned inage pornt distribution.<br />

th!s<br />

The natnenatical backqroLrnd for the construction<br />

of bivariate polynonial sets of additional Daraneters is the follolring<br />

The displacenent of inrage coordinates, modeled by bivariate<br />

AX - f(x,y)<br />

c-v = g(x,y)<br />

- ,vTAx<br />

= yT3x<br />

polynomials is<br />

!here A,B denote the pdraneter matrices and the ve€tors x,-v include the po,/ers<br />

of i nage coordi nates as<br />

(6)<br />

I<br />

(L x x2<br />

(tvv2<br />

xn)<br />

vn )<br />

(7)<br />

Tne bi vari a!e transfornation<br />

i<br />

^ie-v"i_:5<br />

1 r ,' l. ,.)/t, r" y<br />

^ 1,.- | ,<br />

y ^v ^zv,(^'v l,ev ',...<br />

z: -. -l<br />

,2 ^r2.42r2t, r) rz t,^a r2l...<br />

-',-1t----t-!)<br />

t) xY' ,'Y'<br />

"Y' "Y'1..<br />

,.4 .l 2l r 4 .r .rl<br />

:::;:<br />

yn *yn *tyn "tyn<br />

*"yn. . .<br />

-<br />

l<br />

nz<br />

., n,,3<br />

.. n.,4<br />

:<br />

(3)<br />

!hereby the li|lits of a 3x3 ina,qe point distributicn (S9) dnd a 5x5 inaqe poini<br />

distributian<br />

(S2S) are narked.<br />

If th€ ina!e points are not disrributed eq!all,v in x- and y- direction a proper<br />

reLLdroL-er se-ectior car oe !orr'.C o.r ds rfdrcr_eC . { .<br />

lf someone r?an!s to confirfi hin<strong>self</strong><br />

diaaonal lines have to.\e<br />

to a certain deqree of polynomials, then the<br />

follo..red. Thouqh this concept is occasionally applied,<br />

it rusl be reqarded as a quito drbirrary proceedin-o, beca!se it leeCs to en inco<br />

Dlete bivariate polynomi dl .


11<br />

In<br />

this cage rne c0rresponcr ng<br />

parafneter malrices are not ful<br />

ly popLrlateC, as<br />

aLl a12 a13<br />

'2t "22 "?.,<br />

'31 o32<br />

a41 | 0<br />

I<br />

0<br />

34<br />

"11 -12 "13<br />

b2t hz2 lzs<br />

b3l b3z c<br />

b,t1 0 0<br />

ot,<br />

\<br />

ol<br />

,l<br />

ol<br />

Th!s<br />

in 9<br />

.1nans !l-,rt possibl/ si!rr"-"r L<br />

t.]e ser of ddoitrolal pd.dr.r!ers<br />

inage points is<br />

parümeters are let o!t of consideration.<br />

c0rnpensatinq the whole systematic effect<br />

13(x,y)<br />

-o3(x,yl<br />

, t0,,<br />

,tu,"<br />

(e)<br />

wl tn<br />

A3<br />

I'11 '12<br />

a2t uzz<br />

a3t 432<br />

"13<br />

423<br />

u3:<br />

"3<br />

(<br />

"11<br />

b12<br />

"2r -22<br />

":l<br />

D:z<br />

ot3<br />

ozz<br />

o3:<br />

)<br />

(r0)<br />

rx = d11+ar<br />

22<br />

2x+d2ly+ar :x2+a zzxy+a:t<br />

y2 r u yr 2312 u 3zry2<br />

t u 33,<br />

r,y = b + ,12<br />

1I<br />

b +b l2x 21v+<br />

b,. x 2+b<br />

(11)<br />

r rxy+<br />

b,, y2+t: y +t.<br />

rrx2 rxy2+a rrx<br />

For vertical photoqraphy and flat terrain the 6 parajreters of<br />

tation are included in (11). To avojd hiah ccrrelations<br />

ted. Hence ve obtain for exanrle<br />

u rrf, ^ ,r*!,<br />

u rr!2<br />

*u z: "<br />

2y*<br />

u::ry 2n.:: * 2 y 2<br />

brrr+o. rx2+brrxy+<br />

t ra^?y<br />

ra<br />

"<br />

r"v2 ra r.*2v2<br />

these<br />

extert0r ort ennave<br />

!0 !e reJec-<br />

(12)<br />

ilLrnerical aid statistical considerations lead to the idea to construct orthoqonal<br />

parameter sets, i.e. additional paraneters, which are independent amonq<br />

each 0ther and <strong>with</strong> respect to the elements of exterjor orientation. At this<br />

point it must be emphasized tirat orthoqonality amon! the addjtional pa!"aneter:<br />

js strongly valid only eith resp€ct to a certain reqLrlar imaqe point djstri5utron,<br />

aFong idditional paraneters anC the exterior orientation elements of the<br />

individLral imaees only in tfe case cf exact vertical Dhotoqraphy änd flat terrain.<br />

F'rrthermore one has io notjce the fac! that even ir thos€ extrenel_v regular cases<br />

correlation between the ddditional par3fieters and the ccorCinate Lrnknc!ns of tne<br />

qroun0 pornts aay 0ccur. l_rp t0 no,v tnjs t/as not sufficiently t3ken jnto consideration,<br />

?hat fiay result r'r serious conseq!ences as jt wjll be sho,{n later on,<br />

The o r t n o _o n a I i ? a t i a n of {12) leads to ttre follo!,irng oa.emeter set, firs!t_v<br />

publ i shed by Et,te! /-a/.<br />

l2<br />

l


12<br />

rr = f{x,y) = blx+b2y-b32k+b4xy+b5l+ 0 +b7xl+ 0 +b9yk+ 0 +bIlll+ O ,<br />

-- (lrl ,,,,<br />

,1y = S(x,y)<br />

=-b1y+bZx+b3xy-b421+ 0 +böl+ D +b8yl+ 0 +510(l+ 0 +b12kl;<br />

Thus, for edch reqLrlar screen of imaqe poinLs on orthogoral set of parameters<br />

can be constr!cted and nodified i:n hF if.lp.endenr <strong>with</strong>.ecnp.t<br />

to the orientation.<br />

elerrients.<br />

Sesrdes !he l,(l image point screen tne 5x5 i age point screen (Fig!re 3) is of<br />

greaa l nportance.<br />

Fi gure 3: 5x5 1flage point<br />

distribution<br />

L<br />

I<br />

buti on starts from<br />

3y<br />

f5 (x 'y)<br />

ss(x,y)<br />

,too*<br />

,tuu"<br />

(14)<br />

_'e<br />

utt '<br />

As ' Bs =<br />

:<br />

ust '<br />

)<br />

o . L _t ^ g o n o I I z J r i o n o.C tre<br />

r"jecrion<br />

the 6 exterior orientation elements lead<br />

(15)<br />

!{hi ch col. respond<br />

f. fha f^ll^!inn '14 paremerer ser;<br />

to<br />

n:rimEtp.< fn. fhi< reg!lar i maqe point distri-<br />

'-*<br />

-Tt2<br />

, 1 -,2<br />

2 t7 2<br />

t7<br />

o.12-<br />

2A<br />

2A<br />

. = r2('*2 - , = ? 31<br />

,2(, 12) ' -! ca<br />

2A<br />

2A 1n<br />

2<br />

Äx = at2x<br />

a21y +a 2rxy<br />

+ a31l<br />

0 -b ^^i:k+<br />

rv =-4. -v<br />

urr" -urrllt*<br />

o<br />

(r,(.<br />

(rY.<br />

)+ ari.(f+ a.ryk<br />

){ 0 + 0<br />

a32xt+a41yq<br />

0+0<br />

0<br />

b l<br />

Z3yk+bj2x


L:<br />

a^..1rd..r -',1 ra,^^vo-r..5 - 0 r 0 .1<br />

trl..tr 0 - 0 - 0 | 0 .ol;r -D2I,yorl L]+bII yq-b5.s'<br />

1;r..) ra".vr +r."^lo +a,.vlo+a""xs + 0 + 0 + 0 + 0 i<br />

(ly..) + 0 + 0 + 0 + 0 +b2./r + b r I p + b : 4 .t I y I q + b 52<br />

j( s +<br />

,:^..) + a " . I r ' a , , x v o c ' a . . I s + 0 ' 0 0<br />

irl..) + 0 + 0 r 0 +oJ-" rb41\yoo-b531's -<br />

{r{,.) -a../rr-a,^os ! 0 r 0 -d..rs r 0 ,<br />

,.y..)r 0 ' 0 rb..yortF "<br />

- (16)<br />

,, 5l "' " -55 - '<br />

li(.,


Therefors the probler oi statistjcal testina !/ill be treateC unio.<br />

(.d. n.nr.:l<br />

SupDose the linear roCel cerived from (1)<br />

v = Ar - I i<br />

or in statistical<br />

P<br />

nötation<br />

L\r/ r 2"-I<br />

.( , -(.) . '<br />

oo<br />

i f(v) = 0<br />

(17)<br />

vrith the general linerr<br />

hypothesis<br />

Ho: 3x = l"l (1s)<br />

llhere B is a bxu matrix \"lith rank (e) = b=u.<br />

The minimum variance unbiased estimators lor r,.l from (1i) are,rith r = n-u<br />

r = (q Pq) ,A Pt (19.)<br />

':<br />

= - 'r -Ai, 'P, t -Ai ) (20)<br />

LJnder the addi tional constrain (i8) they becorne<br />

i = i-1rroa1-ls- s,,'"o11-lnrr-lqei-nt , (/1,<br />

-2 L .. "^. T^. - .-.<br />

.; - : -A


15<br />

i nfl uence does occur<br />

unknoLlns in (1):<br />

I dz = dz'<br />

alto!ether, Ee set vrith resDect to the sLrbdivisioi of the<br />

(l = I den ti ty ratrix) (?6)<br />

then T becones<br />

T=+(d?<br />

"";<br />

-dz<br />

tTl;l {a: - a'')<br />

(m d number<br />

of addi tionä l Daranreters )<br />

127 )<br />

Accordina to (i)<br />

the symrietrr'c matrix it of the normal equatjons is<br />

qlp"r,<br />

oJp^r^<br />

./. rln re r* e,<br />

n] e rr,<br />

'r lr un,<br />

"1'B^4<br />

AiPBA4<br />

t,zej<br />

rl n nr, r]rrno<br />

\ ttT ::)<br />

AqPAA,l*Pz<br />

'zz - 'zz -ll<br />

l-ll<br />

is the 0artly reduced |l.atfix of t'1€ norral equations for thp se! of addr'tionäl<br />

paraneters. If the aCditional Daraneters are arranqed at the end of the cornolete<br />

vector of unkno,r'/ns as it is s!pposed in (l). then 0;l i! obrained<br />

additioral effort during the factorizati.h of .<br />

The test criterion T (27) cofresDonds to potelljrg's T2-tesi, i.e. the muttivarjate<br />

eriension of the univarlate t-test.<br />

reduced to a t-test.<br />

:.t i th<br />

Thjs imDticates that (27) can be<br />

-'i<br />

-_ i<br />

(30)<br />

' |,2<br />

-ozi J<br />

(31)<br />

where T is distributed as St!de11t's t-test,rith r deorees of freedo,n.<br />

The global test (27) on all additionat paraneters atto,rs the ascertajnnent Df d<br />

global siqrificänt svstenatjc influerce, Bhereas to check the siq,lifjcance of a<br />

subset of ödCitional pararneters one has to nodify Ho in a suitable manner,<br />

If only ä single parameter has to 5e tested and if only sliaht corr-"lations of<br />

this param€ter trr all otirer unknovrns do occur {:l) can be aoDlied. So the corj.elatjons<br />

bet!/een Daraneters are at least of the Same inDortance !nan ahe<br />

But since the corneutation of a renarkable nuFber of €orrelations (.ddjtjonal paraneters<br />

to Doint Coordirates, ddditjonal para.retefs to orientation elejnentS)


16<br />

becones a very expensive orocedure, tfe advanta€e in usin! orthogofal parameter<br />

sets is qui te obvious.<br />

lf 5!fficient independency of a set of addjtional garameteI.s is ensured \,rith<br />

resoect to t e orienLdL'o' ele-e.ts ano a so to t\F pojnt coo"dindt.s tlen 0;]<br />

from (29) incl!des all necessary irformation for a sophisticated testing of<br />

additional parameters. If a correlation coefficient for tvo additional pdrarßeters<br />

exceeds a certain limit,<br />

tested tosether.<br />

This I eads to<br />

let us say 0,75, then those two parametera have to be<br />

IT<br />

2,2<br />

2,1 2,7<br />

(32)<br />

- 0 dnd d2 , (d!. .lz, ) then T resul !s<br />

I<br />

26t<br />

(däj d2k )<br />

Q,i.;<br />

Q,*.1<br />

Qrl.1<br />

qrg,1<br />

)'(,,,)<br />

(33)<br />

l.ih i le for a type I error si:e o we mostly use<br />

power I depend on the al ternati ve hypotheses.<br />

To de-onst.ate Lhe aoove -enElored Droolens a<br />

tr'ca1 example is introd!ced.<br />

Thereoy lle DraLe use of a test field flight abo1,'e the test field<br />

Fi_oures 4a änd 4b).<br />

- 0.01 or s = 0.C5, I and the<br />

Rheidt (see the<br />

Fi!!re 4a: Test field fli!nt "Rheidt 74" Figure 4b: Inraqe point distribution of<br />

(tlay 13, r974). R:ll( AR l5123, test flight "Rheidt 74'<br />

3 strips ur'th 5+6+5 = 16 images,<br />

60X forvrard overlap, 60.1<br />

sr'delap, alternative fl i qht<br />

di recti on, i mage scale 1 :5,300<br />

The testfield Rheidt consists of 4l much re-culary distributed groups of ground<br />

points \aith 3 p0i nts each group<br />

Tne fli!ht arrangement leads to the imaqe point dlstrjbution of Figure 4b.


I]<br />

In the follovrin! demonstration He \vill use only 4 control points (nos 110' 15c'<br />

510, 550), all the other points serve as check polrts.<br />

lleglecting all other !eometrical corditions the image ooint distribution seens<br />

to allovr the introductjon of the (nonorthogonal) paranret-'r set derivec frcm (8)<br />

a21y+a22xy+a3ly2+uru*3*ur,r2Y*u,rtY2*"0.V3<br />

(34)<br />

aY = !lx'YJ brrv+br,x2+urrxv+orox3+lr,x2v+l,rxv2+bntv3<br />

It must be emphasized, lhat this set of a reduced bivariate<br />

being regarded as representative of a suitable parameter set. It<br />

the demonstl^ation of some inconvenient effects \rhich result fronr<br />

of an unqual i fi ed set.<br />

Table 4 shows the results of vario!s prograrn runnings rith<br />

presufrptlons.<br />

fne medrrng o' tne oifferenL ruf ings 1s:<br />

Al1 paraneters of equal wei ght pi =0 ( free unkno!]ns )<br />

paramelers<br />

rmage p0rn!<br />

1*l = lvl "<br />

of equdl weighi, !Jhich is equivalent t0<br />

di.nl).amcnr.f 1q m fn. prrh n;.rE)far<br />

100 mm<br />

t-vpe is far allay from<br />

is only used for<br />

the i ntroducti on<br />

changinq parameter<br />

B, but <strong>with</strong> oo (parameters al4, b41 are excluded)<br />

As B. but wilh o : o-<br />

'22 dll<br />

a.,,<br />

b,, are e)(cluded)<br />

+.o rDdra''leLers a-2. ell<br />

J41<br />

iatrle a: Pi.anerer<br />

(4 coiüol pol^s, L1s.n..( 'oln!3).<br />

(r4), dpprled<br />

,i:, :;,1 .',']<br />

13.r r,0 .2<br />

'F<br />

9.I -d l. 1.9<br />

--_-t-<br />

rortiiieirv<br />

reiqiil


IB<br />

case A, vrhere all parämeters are regarded as free unkno!,/ns, jndicates qLrite in_<br />

ferior res!lts, especially in lre jqht (rrZ = 150.2rm). This is mainly due to the<br />

fact that the qeometrical conditions (only 4 perimeter control points) don't<br />

al'lo!J a seperation of the paraneters a14, b23 and a32, b41 (correlation coefficrents:<br />

r- * = 0.995, r" = 0.987). This becomes evjdent if the effects<br />

o 14' 23<br />

d32"41 "<br />

of these pdrameters are st!died. As it was sho\'/n for a sjnOle strip in schtt /22/,<br />

t h e d i s p I a c e m e n t s Äx = a,,x", Äy = b,, r'y lead to a lonqi tudindl hei ght curvature,<br />

L1e dlsDlacererts a^ = a3zry' \ ay r o4lyJ le"o lo d heiort curvoLLre across sl''D<br />

direciion.<br />

For n block this is valid in like manner, so that these deformations<br />

cannot be determjned by only 4 perineter heiqht control points<br />

To avold the larqe and unreliable paranreter amounts of those parameters, Ylhich<br />

cannot be determined \vith sufficient reliability a sultabl€ strateqy mdy consist<br />

in introd!cing<br />

small keights. This has been done in Case B _ for each pdrameter<br />

a vreight \ras introduced whjch js eq!ivalent<br />

to dn i'nage point displdcenrent of<br />

15 rrn near the imaqe corners ( l*l = lyl<br />

= 100 mm). This proceedinq leads to m!ch<br />

better resLrlts (accuräcy improvement of factor 1.7'in planimetry and factor 4.3<br />

in heiqht) and cecreases the dangerouS parameter afiounts toqether wr'th thei r root<br />

fiean square errors considerably. The critical correlations stil1 remain large,<br />

t,hereas they are decreasing sli!htlyr<br />

= 0.981<br />

"uralr,<br />

,<br />

Secause of these high correlatiors the 1-djnensjonal t-test nust be reDlaced by<br />

the 2-dimensionai F-test<br />

(33). Applyinq this test to a14/b23 and to d32lb41 leads<br />

to the statenent that bath critical sets are siqnificant even on the 99: d_level.<br />

flence one pardmeter of each par r must remain in the adjustfient. It was decided to<br />

reiect at4 and bat and the computatjon of Case C \,as performed. T!re accuracy in_<br />

provenent against case ^ is of factor 1.6 in planimetry and of factor 9.5 in<br />

rei!ht. The hidhest correlation no!r appearing is bet\,/een a2l dnd b." : r. ,<br />

= -0./t llhrch is not be regerded as inconvenient. "' "21")2<br />

The correlation matrjces of Case A and Case C are shown in Figure 5.<br />

Case A<br />

Fi!Lrre 5l Correl äti on matri ces nonorthogonal Darameter set (34)


19<br />

Co'nparing the signifjcant pdraneters received from t-test in Case A and Case C,<br />

a reirarkable different beha!iour of third degree parameters is noticed, what is<br />

dLre to the fact thal t-lest application in Case A leads to wronE conclusions<br />

concernjn! the paranreters al4,a3?, b23, O,tt.<br />

This yields the requirement of mostly independent oarairreters, ühich can be satisfied<br />

by an orlhoqondl set as described above, as practicall,v applied to o!r Cenorstration<br />

nrojecL drd prese.ted ir Tdble:-.<br />

3LL lhere is arotl'er ohe.c enor recessdr/ to Dut'rto con-tde.dtior, up Lo ror'<br />

Ne had restrjcted oLrr inverstigations to correlations bet'Jeen additional parameters<br />

and between additional parametert and orientation ele'nents, !rhpreby Ädmissible<br />

statistical tests of uni- or nultivariate tyDe cdn be performed vrith<br />

"^<br />

^ _oTeters a^. dro a-r on tre qaourd poifrs one frrds<br />

out that a., leads to a Dldnimer.r. ^. -ach strip in strip<br />

direction, only to be determined r'/ith at least one additional control point in<br />


2A<br />

iä5le 5: Prrahe!?r aho!i:5 iid<br />

):r:matar ser (1ll<br />

-l<br />

rrn j<br />

iLr I<br />

r 0.6 I<br />

la-7<br />

a<br />

!a L-.1a-,1_ L:- . !:l_<br />

L.i<br />

^ltl.<br />

i:_1<br />

ß T- L.s<br />

ro rits 1pl 'iinetrv<br />

helgIi)<br />

Fig!re 6: Correlation matrix of<br />

Case A of the ortho!onal<br />

l2 pa rapie te r ser ,I3).<br />

0nly if the critical parameters b4, b5 are reiected, vlhich corresponds !lith the<br />

reiection<br />

of a22, a1\ in the set (34) ar accuracy i$provement appears (factors<br />

1.5 in planimetry and height) !rhereas a bad behavioLrr of the critical paraPreters<br />

is not visi5le if only the correlations betl/een additional paraneters and betv/een<br />

additional parameters and orientation elerlents are checked'<br />

Tnis leads to the irportant conclusion that high correlations between additional<br />

paraneters anJ grcund point coordinates can occur, so that an exanination of<br />

correlations nust incluCe the ground points to get nost reliable results As it<br />

is seen from the correlation ratrices publi5hed ir flcueLshagen /17/, these 'orre'<br />

latjons firay jncrease up to a value of 0.96 or even more' provid€d that an !nproper<br />

choice of ndditional parameters \'/as flrade !lith regafd to some of the !eometrical<br />

conditions (image overlap, control point distribution).<br />

So a drastic protection a!ar'nst h/orse parameter effects must incl!de the examination<br />

of the correlations betreen additional parämeters and ground point coordrnates.<br />

Jo avoid an excessive computational expense one has to restrict on a fe'{<br />

ground goints at most. Probably the conputation anC exanination of the c0rrelations<br />

of ä single, vrell_locat€d groLrnd point will be sufficient For that a<br />

perineter point should be selected !lhich has to be as far a'tay irom the aciacent<br />

control points as possible. lle!ertheless, some further investiqations are<br />

necessary to obtain a better insight in the problems concerninq the correlations<br />

bet"reen additional parameters and !round points, especially \.lith respect to the<br />

!eneration of an operational procedure for detection of those !ndesired effects


2l<br />

6. Concl udinq remarls<br />

Through the conpensation of systenatic image errors the photo!rdmmetric methcds<br />

of point determination, especially the <strong>bundle</strong> adj!stment |1ith s c I f - c a 1 i b r a t i o n ,<br />

have achieved such a hiCh accuracy level, that new oossibjljties of apoljcation<br />

arise, e.g. in control network denslficätion and cadastrdl surveyinq.<br />

l,{owadays the s e I f - c a I j b r a t i o n te€hnique offers the poss i bi 1i ty te meet the theoreticdl<br />

accuracy expectdtjons. As it wds shown in this paper by several test<br />

projects, thes€ expectations have not been met conpletely. 8ut it has to be<br />

pointed o!t that on those high accuracy levels (up to l-2cn) as presented in<br />

chapter 4, an inrportant restricticn consists in the insufficient accuracy of geodetically<br />

deteririned contro I pojnts.<br />

The superiority of s e I f - c a I i b r a t I o n technjqLre co mpa re d !'/i th test field cal j bration<br />

|'ith respect to accuracy and reliability hds been sho\,ln in chaDter 3, so<br />

that no\.,/adays s e I f - c a I i b r a t i o n technjque m!st be re!arded as an essent.ial part<br />

in a modern b!ndle adjustnent program.<br />

lhereby the whole accul.acy potential of photogrammetry js to be utjlized only<br />

jf a suitable selection of additional parafi€ters js nade. This requires a higirly<br />

developed, flexr'ble parameter set 1/hrclr äl lovrs for the totat co'npensation of the<br />

systeoratic inage errors and which avoids the danger of o v e r p a r a n e t r i z a t i o n . An<br />

overpararnetrization is iniicated by the appearance of hieh correlations. Ceneral<br />

ly those may occuf ainong additional<br />

paraneters and bet|leen additionat param€ters,<br />

€xterjor crientation.elenents and ground point coordinates. Then the 1-djßensional<br />

t-test appljcation yields vrong concl!sions and th!s \,rrono rejection decisions<br />

are fidde.<br />

Startin! from a general linear hypothesjs it has been sho!,/n in which wa_v stätistical<br />

test criterion5 can be Cerived providing for sophisticated testjng.<br />

Ey applying orthogonal parameter sets tre correlations among additional paräfteters<br />

and betl.reen alditiönal oarat,reters and orientation eleirents can be reduc-.d<br />

to a certain minimun. This is one reason for the proposal of the concept oi bi_<br />

varrä!e ortho!onal polynomials for addjtional<br />

parafneter sets.<br />

From an operational ooint of vietr an inportant restriction consists if the<br />

necessity of cornD!tätion of the correlations betr/een adCjtjonal parameters and<br />

Sround point coordinates. Though further investigations have to be performed in<br />

thjs field, a possjble strategy may consist in coflputin! the correlations to one<br />

specially selected qround ooint only.<br />

The results of !he practjcal examFles oresented jn thjs paper have been comruted<br />

by usinq blockinvarr'ant paranreter sets exclusively. Sonre previoLrs tests dith<br />

stripinvariant (project "l.tcosach,') and even iirageinvariant (project Rheidt 71 )<br />

0!Tarneters indicated no further iilr!rover,€nt. This wculd allo|, the conclusion thai<br />

the systenaijc indge errors renajn largely stable durinq a fliqft performance,<br />

!/hat h01/ever slror.rld be confirmed by additional tests.<br />

llhile the achievenrent of hiah photoqranf0etric accuracy levels (3-5!nr) is no lon!er<br />

problenratic, ricre attention must be pajd !pon the stabjlity<br />

of rhotoqrannretrlc<br />

systems <strong>with</strong> reso€ct to reliability problefis. In the fjeld of !eodetic net!lork<br />

adj!strnent several investigations have been perforfied dealinq rith those


22<br />

llok the time has cone to utilize lhe !urpcaed statisticdl test rnethods for photogrammetric<br />

systems, This is vdlid |!ith resoect to investiqations<br />

nLrmber ani distributl-on of points and the flight<br />

concerning the<br />

arran!ement to obtain systems as<br />

.Fl1:hla ä< n.q


23<br />

/12/<br />

/13/<br />

/14/<br />

/15/<br />

/16/<br />

/17/<br />

/18/<br />

/1-a/<br />

/2C/<br />

/ 21/<br />

/22/<br />

/23/<br />

/ 21/<br />

Cra!!bi1i, I.,{.r An introduction to ljnear statisticat models. Voturie<br />

Ic Graw-Hill Book Compdny, Inc., tiew york, Toronto. London 1961.<br />

-lhe<br />

C.üa, A.: sinultaneoLls conoensation of systelratic<br />

llunich <strong>bundle</strong> proqram 11B0P. Ir cerman. prese;teC paoer to the XllIth<br />

Congress of the ISP, CommissioF III, Helsinkj l_076.<br />

i(aeh, K.R- i DistribLrtions of probabiljti€s for statistical reviews<br />

of <strong>adjustment</strong> results. In ce ian. r,litteitun!en aus den InstitLrt für<br />

Theoretische Geodäsie der Universität Bonn,'rir. 39, Bonn 1975.<br />

Xoch, it.F. i ilodel formation for parame<br />

ter estimaiion. In Gernan.<br />

Allg. Vermessungsnachrichten, Heft 7, Juli 1977, Dt. 272 - 277.<br />

Xupie!, c... Impl.ovemext of analytica I aerial triangulation by field<br />

calibration. Proceedi nqs of the ASP, Phoenix, 26.-31.10.1975,<br />

pp. 51 - 67.<br />

l,laue l s hagea, r.I Partly caljbration of a Dhotograrfretric system <strong>with</strong><br />

variable control point distribution and different deteririnistic irodels.<br />

ln Gernan. DGK, Reihe C, Heft llr. 236, '1ünchen 1977.<br />

i,!ol,etaa!, 1. r Error detection in planinretric block adjüstment. Presented<br />

Paper to the XIIIth Conoress of the I5P, Conmission ill, felsirki 1975.<br />

Pape, A.J-: The statistjcs of residuals and the detecti on of out<br />

PaDer presented to the XYIth General Assefibly of the IAG, Crenöb<br />

Pabe"ts, F.G. r -lests of <strong>bundle</strong> block <strong>adjustment</strong> for s Lrrvey coord<br />

Presented PaDer to the XIIIth Congress of the I5P, Commissjon II<br />

Helsinki 1976.<br />

Salnenpe!ö, t:., Ande!Eos, .t., SaraLd":ner, ..1. j Efficiencv of |re<br />

extended mathematical nodel in <strong>bundle</strong> <strong>adjustment</strong>. prese'nted paper<br />

to the Symposium of Commission III of the ISp, Stuttgart 1974.<br />

3chut, c.il. r 0n correction terfis for systenatic errors jn <strong>bundle</strong><br />

adjustn'ent. Presented Paper to the Symoosium of Comrjssion llI<br />

of the I5P, St!ttgart 1974.<br />

lhönds, J. i Variation of fli,,rht arrangement. Its effect on <strong>bundle</strong><br />

<strong>adjustment</strong> <strong>with</strong> respect to randoF errors and systenatic errors. In<br />

Gernan. DGK, Reihe C, Heft llr. 233, i.jnchen 1977.<br />

i ers.<br />

e 1975.<br />

UctiLa, U.A. i Statistical tests as q!idelines in anat,vses of äoJUs!nenr<br />

of control nets, Surveying and |tapping, l,tarch 1975, pi.47 - a2.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!