Զѧ
ACTA AUTOMATICA SINICA
1999 25 6 Vol.25 No.6 1999



ڲԱ궨㷨1)
⸣ں鴨ԬΤ
ժҪͨת˶궨ڲ.תʱô˼ڲĹʽ.ùʽ2DͶӰ任ӽֵPʱǷǳʵõ.δ֪תʱӦ2DͶӰ任þ۸ڲͨ⹫ʽ.ͨͬδ֪תڲܱΨһȷ.ЩΪԱ궨㷨ṩۻͬʱҲʵ㷨ģʵʵͼʵĽ㷨һʵüֵ.
ؼʡԱ궨ͶӰ任ڲ.
CAMERA SELF-CALIBRATIONTHEORY AND ALGORITHMS
WU FuchaoYU HongchuanYUAN BoWEI Sui
(Institute of AI, Anhui University, Hefei230039)
AbstractCamera self-calibration is one of the fundamental issues in computer vision. This paper discusses the calibration of camera intrinsic parameters from the rotation of camera. When camera is rotated around its coordinate axis, the formulas of camera intrinsic parameters are obtained by the algebraic method. These formulas are very useful when 2D projective transformation is very close to the theoretic P. When rotational axis are unknown, the general solution of camera intrinsic parameters is obtained by the eigenvectors of 2D projective transformation. The camera intrinsic parameters could be uniquely determined through the method based on two different rotational axes. These results are the theoretic basis of the self-calibration algorithm. Meanwhile a practical algorithm is provided. Tests with synthetic data and real images indicate that the algorithm presented in the paper is robust.
Key wordsSelf-calibration, 2D projective transformation, camera intrinsic parameters.
1
ǼӾеҪ.ͨĶͨṹ֪ͼƽĳλãڲ14.ַʵӦʵ.Faugeras5ݵԱ궨ͨ˶ȷڲʹ궨Ϊ.ĿǰӾԱ궨ɷΪࣺһ̵оԱ7ķͨάռƽ˶ڲһַBasu8ݣDuBrady9ݣHartley10ݵͨתڲķ㷨һľ.ĶHartley㷨ϼƣ㷨ԴﵽʵõĿ.
ķתΪ֪δ֪Σڲۺ㷨.תΪ֪Ǹڲļ㹫ʽùʽ2DͶӰ任ӽֵʱǷǳʵõģͬʱҲ˾кǿ³Եʵ㷨תδ֪Ǹ˱궨̵ͨ⹫ʽӦ2DͶӰ任ȷڲͨ⹫ʽΪԱ궨㷨ṩۻͬʱҲ˾кǿ³Եʵ㷨.

2Hartley㷨
ΪHartley㷨ǸڲҪ˼.p=(u,v,w)TΪƽ꣬x=(x,y,z,1)TΪϵ.3D2D֮ͶӰ任Ϊp=MxMΪ334ͨΪ.MһֽΪM=K(R-Rt)Ϊڲ(R-Rt)Ϊ.Ϊڲ󣬲ϵԭΪͷĹģʱƽʸt=0MΪ33M=KR.ռͶӰ任Ϊp=KRxxΪ.ƹĶͬһתͬǶʱõͼpi=KRix(i=0,1,,n).
δתʱ(i=0)R0Ϊλ.Piǵ0֡ͼi֡ͼ֮ͶӰ任ΪתRiյӦ2DͶӰ任ƳPiK=KRi.һأPK=KRΪı궨󷽳.ƽ꣬ͶӰPһ.תRdetR=1ܼٶdetP=1Ҫʱɳʵ.תΪԿƳP(KKT)=(KKT)P-T.C=KKTCΪʵԳ󣬱궨󷽳̻ΪPC=CP-T˷ҲΪ궨󷽳.
Hartley㷨Ҫ£
1)궨PC=CP-TдԷʽXa=0aΪ궨CӦ6ά(Cһ)
2)Xa=0С˽a(ӦXTXСֵĵλ)ӦľΪC
3)CholeskiʽֽC=VVT
4)VQRֽV=KQ,KΪڲ.
3֪ת
3.1Z-()ת


PK=KRz()ΪϵͳʽԴۿƳ궨󷽳PK=KRz()ݵĳҪΪ

ʱͨΪK=K1(z)diag(1)Ϊ

ڲK0궨PK=KRz()궨ݵģ.
1.Z-()ת(0<<)ǣתΪRz()Ӧͼ֮2DͶӰΪPͨZ-תȽʵ2D任PڲΪK=K1(z)diag(,,1)Ϊ.
3.2Y-ת
2.Y-ת(0<<)ǣתΪRy()Ӧ2DͶӰΪPͨY-תȽʵ2D任PڲK=K1(y)diag(1,1)Ϊ

3.3X-ת
3.X-ת(0<<)ǣתΪRx()Ӧͼ2DͶӰΪPͨX-תȽʵ2D任PڲΪK=K1(x)diag(1,1)Ϊ

3.4ʽڲ
ǽҪ֪תͿPתR().ʵϣκתֵΪ1e-iȣei(Ϊת).KȵģPK=KR()֪PR()ֵͬ.˴Pֵȷ.֪תᣬȷת.
13֪Z-תȷku,kv,sֵܹȷ֮ıֵY-תȷkv,sֵͬȷkvs֮ıֵX-תkuֵȷ.ǿͨڲͬתڲ.磬ǿͨZ-ת֪kukv=£ͨY-ת֪kuľȷֵkukv=¿kvֵsֵҲƵȷ.ˣǿͨԲͬתȷڲ.ַΪʽ.
3.5ʵ㷨
ϸ֪2DͶӰ任Pת()£ڲKļ㹫ʽ.PϾȷʱ㹫ʽǷǳʵõ.ڹʵУѾȷȷ2DͶӰ任Pֵһƫõı궨󷽳пǲݵģҪṩһʵõ㷨.
궨󷽳PK=KRz()ΪԷ顡Ak=hAR95hR5k=(ku,kv,s,pu,pv)TR5ΪڲKӦ.
ڲKҪku>0kv>0ڲɻΪԼС

ӦԼС˵ͨΪk=A+h+(I-A+A)yyR5A+ΪAĹ棬IΪλ.Բʽ

hiΪA+hĵiaiΪI-A+Aĵi(a,b)ΪabڻõһΪy0.ǵõԼСһk0.ӦľK0(z)Ǳ궨󷽳̵һؽ.1ͨΪK0(z)diag(1)>0.
ڶY,XתƵ.
ۣܽɵó⵱ڲʵ㷨.
㷨.
1)Z-תͼ2DͶӰPzͨ

2)Y-(X-)תͼ2DͶӰPyͨ

3)ССֵ(0,0,0)

4)ڲΪ

4תδ֪
4.1궨PC=CP-TʵԳͨ⹫ʽ
1.ti(i=1,2,3)ΪTĵiTΪPT=TRz()ķֱΪPֵe-iȣeiȣ1.
֤.







f=t1+it2t3ֱΪPֵe-iȺ1ĵλ1

СT=(t1t2t3).ԷPC=CP-1ɻΪTRz()T-1C=CT-TRz,()TT

ʽ֪

ǣa,b,cΪⳣ.

CΪʵԳ󣬱b=0,a>0,c>0.һӵ£.
4.PC=CP-TΪC(a)=Tdiag(a,a,1)TTa>0.

