面向快速室内视觉定位的ORB-SLAM2 算法

针对随机采样一致性(random sample consensus，RANSAC)算法在快速去除错误匹配时因随机性导致算法 效率较低的问题，提出一种采用顺序采样一致性(progressive sample consensus，PROSAC)算法来改进ORB-SLAM2 框架中的错误匹配删除方法。通过利用特征点的匹配质量对特征点进行预排序，减少图像匹配过程中的迭代次数； 提出基于最大化割归一化割算法(normalized cuts and image segmentation，Ncut)的全局BA 分段优化算法，以降低计 算复杂度。通过数据集验证，结果表明：优化后的即时定位与地图构建(simultaneous localization and mapping，SLAM) 系统在保持绝对轨迹和相对位姿误差同ORB-SLAM2 基本一致的情况下，相同图像的错误匹配去除的效率提升了 50%，证明了该算法的有效性。

ORB-SLAM2 Algorithm for Fast Indoor Visual Localization

To solve the problem of low efficiency of random sample consensus (RANSAC) algorithm in fast removing error matching, a sequential sample consensus (PROSAC) algorithm is proposed to improve the error matching deletion method in orb-slam2 framework. By using the matching quality of feature points to pre sort the feature points, the number of iterations in the process of image matching is reduced. A global BA subsection optimization algorithm based on maximum cut Ncut algorithm is proposed to reduce the computational complexity. Through the data set verification, the results show that the optimized simultaneous localization and mapping (SLAM) system improves the efficiency of error matching removal of the same image by 50% under the condition that the absolute trajectory and relative pose errors are basically consistent with ORB-SLAM2, which proves the effectiveness of the algorithm.