一种新的估计基础矩阵的高精度鲁棒算法

A New Fundamental Matrix Estimation Algorithm of High Accuracy and Robustness

针对基础矩阵估计过程中因异常数据的影响导致精度和稳定性不好等问题,提出一种新的基础矩阵的估计算法.该算法首先对匹配点采用了新的分组抽样策略,突出了随机性和均匀性.然后以对极距离作为准则,通过二次中值的方法获得新的内点集和基础矩阵初值.最后引入动态惩罚加权的思想,对新的匹配点集进行非线性优化,恢复了精确的对极几何关系.大量的模拟数据和真实图像实验数据结果表明,在误匹配和噪声存在的情况下,提出的算法切实可行,能够提高基础矩阵估计的精度和鲁棒性.

In view of the dissatisfactory accuracy and stability caused by outliers in the process of estimating fundamental matrix, a novel fundamental matrix estimation algorithm was presented in this paper. First a new packet sampling strategy was adopted to the matched points, which emphasized the randomness and uniformity. Under rules of the distance between the point matches and the corresponding epipolar lines, a new inlier set of matched points and original fundamental matrix were obtained via the twice-mid-value method. Finally, the thought of dynamic punishment weighing was introduced to nonlinearly optimize the new point set, which gained the accurate epipolar geometry. Results of a mass of experiments on simulated data and real images indicate that the proposed approach is a practical way and yields the precise and robust fundamental matrix in the case of mismatch and Gaussian noise.