基于图优化的单目线特征SLAM算法

发布时间：2018-04-04 09:32

本文选题：单目SLAM　切入点：线特征　出处：《东南大学学报(自然科学版)》2017年06期

【摘要】：提出了基于图优化的单目线特征同时定位和地图构建(SLAM)的方法.首先,针对主流视觉SLAM算法因采用点作为特征而导致构建的点云地图稀疏、难以准确表达环境结构信息等缺点,采用直线作为特征来构建地图.然后,根据现有线特征的SLAM算法都是基于滤波器的SLAM框架、存在线性化及更新效率的问题,采用基于图优化的SLAM解决方案以提高定位精度及地图构建的一致性和准确性.将线特征的Plücker坐标和Cayley参数化方式相结合,一方面采用Plücker坐标便于线性投影计算,另一方面采用Cayley参数化方式有利于线特征参数的非线性优化.仿真实验结果显示:所提出算法的位姿估计误差平方和与均方根误差分别是里程计位姿估计的2.5%和10.5%,是基于EKF线特征SLAM算法估计位姿误差的22.4%和33%,重投影误差仅为45.5像素;实际图像实验中的位姿估计误差平方和为958 cm~2,均方根误差为3.9413 cm,从而证明了所提出算法的有效性和准确性.
[Abstract]:A method of simultaneous location and map construction of monocular line features based on graph optimization is proposed.Firstly, aiming at the shortcomings of the mainstream visual SLAM algorithm, such as sparse point cloud map constructed by using points as features, it is difficult to express the environment structure information accurately, so straight line is used as feature to construct map.Then, according to the existing line feature SLAM algorithm is based on the filter SLAM framework, there are problems of linearization and update efficiency. The SLAM solution based on graph optimization is adopted to improve the location accuracy and the consistency and accuracy of map construction.Combining the Pl 眉 cker coordinates of line features with Cayley parameterization, on the one hand, using Pl 眉 cker coordinates is convenient for linear projection calculation, and on the other hand, adopting Cayley parameterization is beneficial to the nonlinear optimization of line feature parameters.The simulation results show that the sum of squared error and root mean square error of the proposed algorithm are 2.5% and 10.5% of that of the odometer position and attitude estimation, respectively. The proposed algorithm is based on EKF line feature SLAM algorithm to estimate the pose error of 22.4% and 33%, and the reprojection error is only 45.5 pixels.In the actual image experiment, the sum of square error and root mean square error are 958 cm ~ 2 and 3.9413 cm respectively, which proves the validity and accuracy of the proposed algorithm.
【作者单位】：华北电力大学控制与计算机工程学院;
【基金】：国家自然科学基金资助项目(61105083) 中央高校基本科研业务费专项资金资助项目(2015XS63)
【分类号】：TP242;TP391.41

【相似文献】