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.
A new line based 6-DOF monocular algorithm for using graph simultaneous localization and mapping(SLAM) algoritm was proposed.First,the straight line were applied as a feature instead of points,due to a map consisting of a sparse set of 3 D points is unable to describe the structure of the surrounding world.Secondly,most of previous line-based SLAM algorithms were focused on filtering-based solutions suffering from the inconsistent when applied to the inherently non-linear SLAM problem,in contrast,the graph-based solution was used to improve the accuracy of the localization and the consistency of mapping.Thirdly,a special line representation was exploited for combining the Plücker coordinates with the Cayley representation.The Plücker coordinates were used for the 3 D line projection function,and the Cayley representation helps to update the line parameters during the non-linear optimization process.Finally,the simulation experiment shows that the proposed algorithm outperforms odometry and EKF-based SLAM in terms of the pose estimation,while the sum of the squared errors(SSE) and root-mean-square error(RMSE) of proposed method are2.5% and 10.5% of odometry,and 22.4% and 33% of EKF-based SLAM.The reprojection error is only 45.5 pixels.The real image experiment shows that the proposed algorithm obtains only958 cm2 and 3.941 3 cm the SSE and RMSE of pose estimation.Therefore,it can be concluded that the proposed algorithm is effective and accuracy.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2017年第6期1094-1100,共7页
Journal of Southeast University：Natural Science Edition
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)
