【摘要】 计算机科学中研究的图像是真实世界(即二维、三维欧式空间)到像平面的射影变换。平面射影变换(单应)估计是特征目标检测、注册、识别、三维重建等方面的关键步骤,但是如何鲁棒、精确地估计单应矩阵是一个没有很好解决的问题。在研究中发现,基于点与直线的直接的单应矩阵估计方法会导致出现较大误差的情况。针对这一情况,文中介绍了一种基于统计优化的单应矩阵估计方法,这种方法是通过单应矩阵的协方差张量的计算和优化来估计单应矩阵的。最后进行了简单的实验,比较了统计优化方法与进行归一化处理后的直接线性方法的估计结果,证明了基于优化统计的估计方法更加有效。更多还原

【Abstract】 The study of the images in computational science is projective transformation between the real world(the two-dimentional and three-dimentional Euclidean geometry) and the plane.Plane projective transformation(homography) estimation is a necessary step in many feature-based object detection,registration,recognition,three-dimention reconstruction and some other aspects,however,how to robustly and accurately estimate homography from images is a difficult problem.In this paper found that usual method of homography estimation based on points and lines may result in relatively big errors.Under such configurations,a new estimation method which is based on statistical optimization is proposed in this paper,this method estimates homography matrix through covariance tensor of homography matrix.Have conduced some experiments and compared the estimation results between the method of statistical optimization and the method of normalized DLT,thus proved that the estimation method based on statistical optimization is more effective.更多还原