基于GC-RANSAC算法的单目视觉同时定位与地图构建

A Monocular Visual Simultaneous Localization and Mapping Based on GC-RANSAC Algorithm

移动机器人单目视觉同时定位与地图构建(Simultaneous Localization and Mapping,SLAM)技术在应用过程中会获取大量数据,针对其带来的异常值干扰问题,提出一种基于新提出的随机抽样一致性(Random Sample Consensus,RANSAC)算法改进的半直接单目视觉里程计(Semi-direct Monocular Visual Odometry,SVO)算法。算法分为两个线程:建图线程提取点特征与边缘特征,并采用了图割RANSAC(Graph-Cut RANSAC,GC-RANSAC)算法进行异常值剔除,通过计算特征点的深度值,来构建的环境特征地图;位姿估计线程通过最小化局部地图点和边缘线的重投影误差以及帧与帧、特征与特征之间的约束关系优化,得到位姿信息,实现定位功能。通过Euroc公开数据集上得到的仿真实验结果可见,该算法剔除异常值效果明显,平均定位精度相比SVO算法提高了15.6%。

An improved SVO(semi-direct monocular visual odometry)algorithm based on the new RANSAC(Random Sample Consensus)algorithm is proposed to solve the problem about the interference of outliers in excessive data interpretation existing in the SLAM(simultaneous localization and mapping)system.The proposed algorithm is divided into two threads.In the mapping thread,an environmental-feature map is constructed by extracting the point-features and line-features,adding the rejection of outliers which based on the GC-RANSAC(Graph-Cut RANSAC)and calculating the depth of the features.In the motion estimation thread,the visual pose is estimated by minimizing reprojection errors of points and line segments in the local map and the constraint relationships between frame and frame,feature and feature.The simulation results on the public datasets show that this algorithm is efficient and reliable;the effect of the rejection of outliers is improved well.The average positioning accuracy of the proposed algorithm increases by 15.6%on the Euroc dataset compared with that of the SVO algorithm.