基于 RANSAC 的视觉里程计优化方法研究

针对图像特征产生误匹配影响基础矩阵计算，导致同步定位与地图构建(SLAM)视觉里程计估计精度差的问题，提出一种基于随机抽样一致(RANSAC)的视觉里程计优化方法。该方法首先通过最小距离阈值法对初始匹配集粗滤除，再采用RANSAC计算图像间相对变换关系，若符合变换关系为内点，内点数最多的迭代结果为正确匹配结果；然后计算图像间单应变换并利用其计算基础矩阵，采用对极几何约束确定内点，得到具有最多内点的基础矩阵；最后采用TUM数据集从特征匹配与基础矩阵计算两方面进行优化算法效果验证。结果表明，该算法可提高运行效率且有效去除误匹配特征点，使匹配正确率提高7.7%,基础矩阵估计算法在提高基础矩阵计算精度的同时，内点率也提高了3%,算法为提高视觉里程计精度估计精度提供了理论基础。

Study on optimization method of visual odometry based on RANSAC

The mismatch of image features affects the basic matrix calculation and leads to poor estimation accuracy of SLAM visual odometry. To address this issue, an optimization method of visual odometry based on RANSAC is proposed. First, the initial matching is roughly filtered by the minimum distance threshold method with an appropriate threshold, and the relative transformation relationship between images is then calculated by RANSAC. The result that conforms to the transformation relationship is considered an interior point. The iteration result with most interior points is the correct matching result. Then, the homographic transformation between images is calculated, and the basic matrix is derived from the calculated results. The interior points are determined by epipolar geometric constraints and the fundamental matrix with most interior points is obtained. Finally, the TUM data set is used to validate the performance of the Visual Odometry optimization algorithm from characteristic matching and basic matrix calculation. The experiment results show that the optimized RANSAC algorithm not only effectively improves operation efficiency and removes the mismatched feature points, but also improves the accuracy of image feature point matching by 7. 7% . Meanwhile, the interior-point rate of the basic matrix estimation algorithm in this paper is increased by 3% while improving the basic matrix calculation accuracy. This algorithm provides the theoretical basis for improving the accuracy of visual odometer estimation.