基于SIFT的宽基线立体影像最小二乘匹配方法.pdf

39 2 Vol.39, No.2 2010 4 Acta Geodaetica et Cartographica Sinica Apr.,2010 :1001-1595(2010)02-0187-08 SIFT 1, 2 1, 2 2 杨化超 , 张书毕 , 张秋昭 1. , 221116; 2. , 221116 Least Squares Matching Methods for ide Base-line Stereo Images Based on SIFT Features 1, 2 1, 2 2 YANG Huachao , ZHANG Shubi , ZHANGQiuzhao 1. Jiangsu Key Laboratory of Resourcesand EnvironmentalInformation Engineering, Xuzhou221008, China; 2. School of Environ- ment Spatial Informatics, China University of Mining andTechnology, Xuzhou221008, China Abstract:Some newfeature extractionand matching algorithms(especiallythe Scale Invariant FeatureTransfor- mationalgorithm, SIFT) that are invariant totranslation, scaleand rotationchangeshavebeenwidely used in dig- ital photogrammetryand computervisionfields. However, SIFT features are not invariant to affine deformation, and the location precision may be lowbecause the feature extractionand matching are independent. In practical applications, thisdisadvantage makesit not appropriate forwide baselinestereo matching. Inorderto solve this problemand to make it invariant to affine transformation, a newSIFTfeatures. matching algorithmbased ondupli- cate constraintsand least squaresimage matching methodwasproposed inthispaper. Inthisalgorithm, the opti- malSIFT features withgood spatial distributionand large information content wasfirst selected, thenthese SIFT featureswasmatched by usingSVD-SIFT algorithmand adaptive NCCmatching method based onscale and orien- tation informationof SIFT features. The fundamentalandhomography matrix canbeestimated by usingtheseinitial correspondences. Other SIFT features were matched by using duplicate geometric constraint and weighted least squares matching with multiple scale templatewindow. At thesametime, the least squar