在冠脉造影的三维重建过程中,由于在成像过程中不同方向上的对应图像形状差别很大,通过常规的特征点匹配方法得不到匹配点,因此只能手动选取血管中轴的分叉点作为特征点,这样得到的特征点的数目有限,一般为十个左右,这样对于基础矩阵的精确计算是一个难点。为了解决这一难题,本文利用冠状动脉造影序列信息提取大样本匹配点对,将景物重建中的常用的随机采样一致性(RANSAC)方法应用到冠状动脉造影三维重建中来。试验表明本方法对冠脉造影的对极几何估计具有很强的适应性和鲁棒性。

Since the images vary much with different view points in the 3D reconstruction of coronary artery,it is impossible to get matching points using traditional methods.Therefore,we only manually got bifurcate points as characteristic points,and the number of points was few,thus the estimation of fundamental matrix was difficult.To overcome this problem,a method was introduced to enhance the geometrical accuracy from two noncalibrated serial two-dimensional angiographic projections.Using serial coronary artery angiograms,we could get many matching points.Thus we could calculate the epipolar geometry using Random Sample Consensus Paradigm (RANSAC) which was often used in the reconstruction of nature scene.The results show that the method proposed is accurate and robust for estimating the fundamental matrix in angiograms.