给每个特征点计算特征向量r
描述子。这样一方面算法的时间得到了降低，而另一方面算法的鲁棒r
性也得到了提高。r
针对求解基本矩阵方面，利用5一点算法求解出本质矩阵，然后利r
用基本矩阵与本质矩阵的标定关系推导出基本矩阵，同时对5一点算法r
中的所有解引入了反投影残差最小约束法，排除错误解，提高了5r
点算法的正确率。r
关键词基本矩阵，5一点算法，本质矩阵，特征点提取算法ABSTRACTr
Patter
Matchi
gi
comPutervisio
i
theresumPtio
ofther
corresPo
di
grelatio
shiPbetwee
twoimagesPixel，hasbee
thefocusr
ofthestudyCollectedfromdiffere
tviewPoi
tsofthetwosameobjectr
sPaceimagePoi
tcorresPo
de
eebetwee
certai
co
strai
ts，This15ther
ePIPolargeometricco
strai
tbasiemathematicalexPressio
ofmatrix15r
thiski
dofgeometricco
strai
tFu
dame
talmatrixeo
tai
sadiffere
tr
vieWPoi
taretheimPorta
tgeometricalrelatio
betwee
imagea
dr
describedthecorresPo
di
gPoi
tsshouldmeettheePIPolarco
strai
tr
co
ditio
sco
tai
stwoeamerasorbetwee
asi
glemotio
Parametersr
oftheeameraA
aeeurateestimateofthefu
dar
e
talmatrix15a
r
imPorta
ttaski
eomPutervisio
ThisArticleresPeetivelyo
comPuterr
imageProeessi
gofimagematchi
gi
volvedi
tech
ology，Featurer
Poi
tdetectio
teeh
ologya
dsolutio
offu
dame
talmatrixofthreer
co
ductmorei
一dePthstudies，comParedtoallki
dsofalgorithms，saidr
theiradva
tagesa
ddisadva
tages，a
dPutforwarda
imProvedr
algoritllmO
featurePoi
tsextraetio
，asSIFTalgorithmforealeulati
gr
lo
gtime，a
dHarrisalgorithma
dgraphrotatio
oisedoes
otfit，Putr
forwardacombi
atio
ofadva
tagesofthetwoHarris一SIFTalgorithm，r
that15，thefirstestablishascalesPaee，useHarrisalgorithmofFeaturer
extraetio
a
dthe
toeachfeaturePoi
tfeatureveetordescriPtorr
Algorltllmoftimehasbee
reducedo
theo
eha
d，a
dalgorithmr
robust
esshasbee
imProvi
gO
thesolutio
ofthefu
dame
talr
matrix，Putforwardas一Poi
talgorithx
forsolvi
gtheesse
tialmatrix，r
a
dthe
usethebasic
atureofmatricesa
dmatrixcalibratio
r
relatio
shiPderivedthefu
dame
talmatrixa
do
misu
dersta
di
gr
Poi
tsi
as一Poi
talgorith
lfordeteetio
ofPolardista
eemi
imumr
Pri
eiPle，imProvetheaceuraeyrateofthes一Poi
talgorithmr
Keywordsfu
dame
talmatrix，5一Poi
talgorithlll，esse
tialmatrixr
featureextractio
algorithl
目录r
摘要……Ir
ABSTRCA丁r