基于数字视差模型和改进SIFT特征的数字近景立体影像匹配

Digital Close-range Stereo Image Matching Based on Digital Parallax Model and Improved SIFT Feature

针对数字近景立体摄影测量中立体影像的高效自动匹配的需要,提出一种基于数字视差模型(digital parallax model,DPM)和改进SIFT(scale invariable feature transformation)特征的立体影像匹配算法。该算法首先根据近景立体摄影系统中相机基距和相互关系固定的特点,利用已知的标定数据,构建立体影像的数字视差模型,并计算立体影像中的同名核线参数,再在DPM和核线的约束下,确定右影像上同名像点的搜索范围,以提高匹配速度和准确性;其次,利用SIFT特征和灰度特征,设计结合局部结构信息和全局灰度信息的匹配相似性测度计算模型,以增强匹配的稳健性。利用该算法对实际立体影像进行试验,结果表明本文方法可大大消减错误匹配,提高匹配的可靠性和精度。

For the purpose of automatic and effective stereo image matching in the field of digital close-range stereo photogrammetry,a new stereo image matching algorithm based on digital parallax model（DPM） and improved SIFT feature is proposed.In the new algorithm,considering characteristics that baseline and relation of two came-ras is fixed in digital close-range photogrammetry system,digital parallax model of stereo image is constructed firstly,and computation parameters for confirming conjugate epipolar line are work out with known calibration data.Then searching scope of conjugate points on the right image is confirmed under constraints of DPM and epipolar line to improve matching speed and accuracy.Secondly,to increase matching robustness,the computation model of matching similarity measure combined with local structure information and global gray-value information is designed by SIFT feature and gray-value feature.The new algorithm is applied to actual stereo images taken by digital close-range photogrammetric system.Experimental results show that the proposed algorithm can decrease matching error greatly,and enhance the reliability and precision of stereo image matching.