Ϣ
Information and Control
1999ꡡ285ڡVol.28No.51999



2DӦи˶Ƶ³㷨

º塡ά

ժҪһָݸ˶ǰɵ2D ӦݼϹ˶ķǵдڲƥĶӦ㣬 Сƽ͹ƺ͵ݹвķ޳ЩݣӶøռ˶ȷ ҲҪκ֪ʶʵ㷨Чԣ 
ؼʣ˶ƣСƽͣŴ㷨³ԣ 㣮ݹв
ͼţTP13ױʶ룺B

ROBUST ESTIMATION OF RIGID BODIES' MOTION PARAMETERS
FROM 2-D POINT CORRESPONDENCE
CHEN Hai-feng1 HE Tie-jun2 CHEN Wei-nan1
(1. Institute of Automation Southeast University 210096; 2. Research Institute of Intelligent Transportation System, Southeast University)
Abstract: Given the 2-D point correspondence set of a scene containing moving rigid object, this article proposes a method to estimate its m otion parameters. Because there exists several incorrect correspondence in the s et, the least trimmed square estimator and recursive residuals analysis is adopt ed to deal with the outliers and get the correct motion parameters. It does not need any prior knowledge. Computer simulation results demonstrate the effective ness and efficiency of the algorithm.
Key words: rigid motion, parameter estimation, least trimmed sq uares, genetic algorithm, robust, outliers, recursive residuals

1ǰ
˶Ӿоδӱ仯һϵвͬʱ̵ͼУȡйس״λú˶ϢǻӾһҪооϿɷΪķͻڹķ࣮оǻ˶Ʒ Ϊͬʱt1t2ʱͼP1(x,y)P2(x,y)Ҫͼи ˶(R,t)RǸתtƽ
кܶĳ㷨Щ㷨޵źûͼӦƥȻʵڸżȻأţʹ˶ǰͼĶӦ㷢ƥһЩǸϵĵ㣮Դˣ˸ָ³㷨HaralickMƷ[7]÷ڳ ٷֱȴ20ʱʧЧׯ»һģƥĹ㷨[3]ѧ ʼֵ³㷨[6]ǶҪһЩĳʼֵա µСƽֵĹ㷨[5]ܺܺõشеĳ,÷ ɳʼݵʱϳⳢСƽ͹(Least Trimmed Squ ares LTS)͵ݹв(Recursive Residuals)ķ޳ݣȷĲ ƣΪ㷨ЧʣвŴ㷨Ѱŵֶݣǵʵ пø÷Ƹ˶ܹЧĿ˷ڳӶõ˶ ȶĹƶκ֪ʶ
2˶Ƶ㷨
nӦ󵥸˶Ѿкܶ㷨нӰTai & Huang㷨[1]͸ͶӰģͣ轹f=1ôϵ(X,Y,Z)ͼϵ(x,y)֮ĹϵΪx=X/Z,y=Y/Z,(X,Y,Z)(Xt,Yt, Zt)ֱΪ˶ǰϵ㣬˶33תRƽt ʾ
(1)
һԳƾ  Ȼ T.t=0uuTTu=0ڷʽ(1)ͬʱ(X,Y, Z).T
(x,y,l).E.(xyl)T=0(2)
Է(2)εģֻ8ﲻ裬h9=-1,ô(2)ʽԱʾɣ
T=1Ц=(xx,yx,xxy,yy,yx,y)T=(h1,h2, h3,hs)T.(3)
ɼÿһ(xi,yi)(xi,yi)ӦһʸiԷ̦Ti=1Ҫδ֪£8Ӧ㣮С˷º󣬾ͺø˶(R,t)[2]
㷨õ˶ǰͼһ㶼ģЩΪ࣬һ㱾λһȽС̬ʾͨС˷ ܵõЧƣһ㷨ļӰȽϴڳּٵ ֮ƥɵģǲֲ̬ͳгΪ㡱ĴڶԻ̬ģ͵С˷˺ܴ⣮Ϊˣͳѧо³һ⣬в˶Ӿõ˶³Է[3-7]ǲһСƽ(LTS)ĹƷ Էϣκ֪ʶģܺܺôгݵ˶ ³Ʒ
3˶Ƶ³㷨
3.1 LTSû
ڻعģ
yi=xi11+xi22++xipp+i(4)
вri()=yi-ri11-ripp,r2i()0(r 2())1n(r2())2nܡ(r2())n: n(r2())k:nʾĵkвƽôСƽ(LT S)=(1,2,,p)TʹĺС
(5)
LTS(5)ʽֻȡвƽǰ֮ͣ迼ϴĲвֻҪ ѡ۲гӰ죮LTSһнǿ³ ԵĹӣh=nʱLTS͵ڳС˷ڸ˶ģ Уǽ㼯QΪQ=AW,Aʾ˶㼯ǶӦ ʸ{i}Է̦Ti=1Wʾ㼯[12]ָ ʹڳWռQıӽ50ʱLTSЧعƳ ʱhȡ,[ ]ʾȡ㣮
(5)ʽкg()Ƿ͹ģóŻֲС[13] pӼķʹg()СĦ£ڼQȡpԣǶӦʸ pԷ(3)һ£p̹ɵķΨһ£ô nʸļϣCpnܵĽ⣬ɼϦJJеÿһ ¡ӦɼAеpԣpӼڼϦJʹg()СĲ¡ƽʹ(5)ʽСģpάѰſռΪCpnռн 
(6)
ڲJg()СҲ£n=100,p=8ôռСΪ1.91011ٷЧ̫ͣ³ԲǿŴ㷨һֻ˼Ż㷨ָһĲռиЧнǿȫܣǲŴ㷨(6)ʽ
Ŵ㷨Holland70۵ŻⷽģȻʤľƣȶȾɫӦȣͨ棬ͻȻŴŻʹȾɫƵõֱۣĳһвӦ㹻ߵ⣮
Ϊ˽Ŵ㷨Сƽ͹⣬ӦȾɫʾʽӦȺѡȡµŴӣ
(1) ȾɫıʾHollandԭʼ[9]УȾɫ嶼ʹö01 ɵĴʾΪŻĽ⣬ڽʹöʹżֱʹ ֵʾڸ˶УǽQеnԴ1nбţڼ ʱQѡpԣpԵıŴ㷨һȾɫS=(i 1,i2,,ip)ȾɫSУЩŵĴ޹ؽҪģ
(2) ȾɫӦȵȡ
a)ȾɫS=(i1,i2,,ip)еÿһl,l(i1,i2,,ip)Ӧ һ(xl,yl)(xl,yl)ʸlԼaT l¡=1SбŶӦԼ̣ʾΪ¡䣮
b)вr 2i(¡)
0(r2(¡))1:n(r2(¡))2:nܡ(r2( ¡))n:n
ǰh֮ͣΪȾɫSӦȣ
(3) ֻӣȾɫʾʽص㣬ǲһµĽ ȾɫS1S2ڽӵ·ֳӴC1C2ʱC1C 2жS1S2ֵıţS1S2Ļ첿֣C1C2 ѡͬĿıţֱɺpŵļϣΪֻ ӡ
(4) 滻ͻӣӶһpŵȾɫ壬ѡȡĿıŽɾȥٴӱ1nгȡͬĿвȾɫеıУǱĿӦʹȱԭеƣַ˱죮 
̣ͨǿԻ㼯Ap(ӦڳӦߵȾɫp)ӦĲ£p˶ЧGaussӰ죬Ϊ˳ֵQеݣǽŲõݹвֶΣӼẠӶС˷ŵСƽ͵ŵһ
3.2 ӼA
Ŵ㷨㼯QpԣӼAӦʸ{i, i=1,2,,p}Է飺
Xp(p)=1(7)
Ц(p)ʾpӼAʸɵķĽ⣮Ӽ̵Ļ˼ǽQеn-pʣʸ(7)УȻԵĲвСĳһֵʱΪӼA޳
ڻعģ͵ĲвУ׼вѧв[15]ȶϸز ģڳУϣҵһֲصĲвܹͨƹʽм 㣮Brownĵݹв[10]ǵҪҵj-1ӼA ʸSj-1={ai|aiA,i=1,2,,j-1;j-1p},ôͬ ģai(j-1)=1 j-1̲ùƲj-1Ϊ ۲ʸajǷͬSj-1ʸͬģͣ塰ݹв[10]ָaj,aj+1, ,an㷽(7) ôwj,wj+1,wnΪֵΪ2ΪǿԸݵw jֵɶAӹ̣wjСֵͽʸajӦӼAУ(7)е(j-1)и£ջAȫݼ Ӧ˶ʸ
[14]е¹ʽǽӼAʵ£
for (j=p+1;j<n;j++)



Add aj into A;}
else {(j)=(j-1);
(XTjXj)-1=(XTj-1Xj-1) -1}
}

4ʵ
Ϊ֤㷨Чԣʵ飮øΧX,Y,ZתǦ, ,ʾתR[5].ȡ(,,)=(6,9,12)ƽt=(6,9,3)T.
ɸ˶ǰݼϣ
B1={(X,Y,Z)T|Xi(-20,20),Yi,Zi(10,20), i=1,2,,N1},
˶ά(1)ʽ˶ǰά͸ͶӰõάͶӰ꣮ǻ˶ǰ2DӦAСN1=40ڼAе˶άͶӰмӾֵΪ0GaussSNR=40
ֲ֤ͬ㷨³ԣ
Q=AWвκγ㣬W=ֱȽ㷨³㷨׼ȷԣ㷨(1,1,1)=(6.03,9.09,11.94)ƽ1=(5.97,8.91,3.13)T, ³㷨(2,2,2 )=(6.08,8.95,11.91)ƽ2=(6.04,8.93,3.07)Tɼ޳£㷨³㷨GaussƵģ
ڼQм㣬ɵ2D꼯W={(xi,yi)(xi,yi)|i=1,2,,N2}ֱWQռΪ10%, 15%, 20%, 25%, 30%,35%, 40%±Ƚ㷨³㷨Ľͼ1ʾͼ1(a)ʾƵƽ ʵƽ֮ļнų仯Ľͼ1(b),1(c),1(d)ֱ ʾת,,׵Ĺų仯Ľͼ1(a)(d)У ߶Ӧ㷨Ľ߶ӦСƽ㷨Ľԣڴ ڳ£㷨ʧЧ³㷨׼ȷعƳ˶

(a)ƽ (b) ת

(c)תȡ (d) ת
ͼ1ʵ

ʵŴѰŵСƽ͹㷨ڹ˶ʱ߽ǿ³ԣҲκ֪ʶһȽĹ㷨

߼飺º壬1975ʿоΪͼͻӾ
1974˶ʿоΪģʽʶϵͳеӦã
άϣ1937ʿʦоΪܿƺͻӾ
ߵλº塢άϣϴѧԶϾ210096;
ϴѧܽͨо

ο
1Tai R Y,Huang T S. Uniqueness and Estimation of 3D Motion Parameters of Rigid Bodies with Curved Surfaces, IEEE trans. PAMI, 1984,6(1): 1327
2 Weng J, Huang T S, Ajuja N. Motion and Structure from Two Perspective Views: Algorithms, Error Analysis, and Error Estimation. IEEE trans. PAMI, 1989, 11(5):451476
3 Zhuang X, Wang T,Zhang P. A Highly Robust Estimator Through Partially Likel ihood Function Modeling and its Application in Computer Vision. IEEE Trans. PAMI , 1992, 14(1):1935
4 ѧ. άӦݷָ͹ƶ˶Զѧ,1998, 24(2):186191
5 գ. 2Dƥָ˶³ƣѧ,1994, 17(5):388392
6 ѧ. һά˶³Ʒ. ѧ,1990,13(7): 481488
7 Haralick R M, Jee H, Lee CN, Zhuang X et al. Pose Estimation from Corres ponding Point Data, IEEE trans. PAMI, 1989, 19(6):14261446
8 Ruddph G. Convergence Analysis of Canonical Genetic Algorithms. IEEE trans.N eural Network, 1994, 5(1): 96101
9 Holland J H. Adaption in Natural and Artificial Systems. Ann Arbor:The Univ. of Michigan Press.1977
10 Brown R L, Durbin J, Erans J M. Techniques for Testing the Constancy of Regr ession Relationships Over Time. J. Roy. Stastic. Soc. B, 1975,37:149 192
11 ¹. Ŵ㷨Ӧ. ʵ. 1996
12 Rousseeuw P J, Levoy A. Robust Regression and Outlier Detection. New York:Jo hn Wiley,1987
13 Rousseeuw P J,Gilbert W, Bassett J R. Robustness of the P-subset Algor ithm for Regression with High Breakdown point. Directions in Robust Stastics a nd Diagnostics. Part 2. 1991:185194
14 KianiFard F, Swallow W H. A Review of th Development and Application of Recu rsive Residuals in Linear Models. JASA. 1996,91:391400
15 Τɵ. ͳ. ϴѧ,1991
ոڣ1998-08-25
