【摘要】 针对仿射尺度不变变换提取(ASIFT)算法计算效率低的问题,提出了一种大倾角航空倾斜影像自动匹配方法 H-SIFT。该方法利用影像粗略外方位元素计算两幅待匹配影像之间的单应变换矩阵,对左影像进行二维射影变换得到其纠正影像以消除两幅影像之间的几何变形、尺度和旋转问题,再对左影像的纠正影像和右影像进行尺度不变特征变换(SIFT)。匹配时,为了适当提高正确匹配点对的数量,利用不严格的比值提纯法和左右一致性检验得到粗匹配点对,并利用随机一致性检验剔除误匹配。最后将左影像其纠正影像上的匹配点反算到左影像上。通过对国产五倾斜相机平台(SWDC-5)获取的三组典型城区航空倾斜影像数据进行实验,对于三组数据,该算法获得的正确匹配点对数量分别为ASIFT算法的2.18、1.31、1.70倍,该算法匹配耗时分别为ASIFT算法的0.93%、0.88%、0.97%。实验结果表明,与ASIFT算法相比,该算法获得的匹配点对在计算效率、数量和分布情况上都得到了显著提高。更多还原

【Abstract】 In order to reduce the high computing complexity of Affine Scale Invariant Feature Transform( ASIFT)algorithm, a robust and rapid matching method for large angle aviation oblique images based on homography transformation was proposed, which was named H-SIFT. Firstly, the homography matrix between the two oblique images was calculated by making full use of the rough Exterior Orientation( EO) elements of the images, then a homography transformation was made to the left image to get its rectified image for eliminating geometric distortion, scale and rotation. Secondly, the matches between the rectified image and the right image were got by using Scale Invariant Feature Transform( SIFT) algorithm. During the matching process, the coarse matches were got by using two matching constraints, Nearest Neighbor Distance Ratio( NNDR)and consistency checking, then the false matches in them were eliminated by using the RANdom SAmple Consensus( RANSAC) algorithm. Finally, as the matching points on the rectified image were got, the corresponding matching points on the left image were calculated by using the homography matrix. The experiments on three pairs of typical oblique images obtained by Si Wei Digital Camera 5( SWDC-5) demonstrate that the matching points obtained by the proposed algorithm are significantly improved in the computation efficiency, quantity and distribution than ASIFT algorithm, as the proposed algorithm not only takes just about 0. 93%, 0. 88%, 0. 97% time of ASIFT algorithm, but also gets the correct matching points about 2. 18,1. 31,1. 70 times of ASIFT algorithm.更多还原