基于改进Vins-Mono的INS/视觉组合定位方法研究

Research on INS/Vision Combination Localization Method Based on Improved Vins-Mono

摘要: 针对现有Vins-Mono算法在惯性导航系统(Inertial Navigation System, INS)与视觉组合定位中存在图像特征点提取分布不均匀、特征点匹配准确率低和系统定位精度不高的问题，提出一种改进的Vins-Mono算法。首先，采用四叉树算法划分图像特征点疏密不同的区域，在各个区域内同时进行特征点提取，加快了提取速率；然后，采用RANSAC算法进行特征点匹配，提高了匹配准确率，并结合汉明距离加快了匹配时间；最后，采用LM算法取代高斯牛顿算法，添加信赖区域控制更新步长，提高了系统定位精度。相比于传统Vins-Mono算法，本文算法在特征点提取分布均匀度上提高了54%，在特征点匹配准确率上提高了28.4%，在不同难度数据集上的平均定位精度提高了34%。

Abstract: Aiming at the problems of the existing Vins-Mono algorithm in combined positioning of Inertial Navigation System (INS) and vision, such as uneven distribution of image feature points extraction, low matching accuracy, and low positioning accuracy, an improved Vins-Mono algorithm is proposed. Firstly, the quadtree algorithm is used to divide regions with different densities of feature points and extract feature points simultaneously in each region to speed up the extraction rate. Then, the RANSAC algorithm is used to match the feature points, which improves the matching accuracy and speeds up the matching time by combining the Hamming distance. Finally, the LM algorithm is used to replace Gauss Newton algorithm, and trust region control is added to update the step size, which improves the positioning accuracy of the system. Compared with the traditional Vins-Mono algorithm, the algorithm in this paper improves the distribution uniformity of feature point extraction by 54%, the accuracy of feature point matching by 28.4%, and the average positioning accuracy on different difficulty datasets by 34%.