摘要
以提高弱纹理场景下同步定位与建图(SLAM)算法的定位精度、鲁棒性和运行效率为目的,提出一种基于点线特征的快速单目视觉惯性SLAM算法。算法通过融合点线特征以及惯性测量信息估计出SLAM系统初始运行参数,在SLAM后端通过构建先验误差、惯性测量单元(IMU)测量误差、视觉特征重投影误差、回环检测误差函数进行非线性优化,估计更加精确的相机运动、路标点坐标及IMU状态信息,对算法中的点线特征检测基于计算机统一设备架构(CUDA)并行计算模型设计了并行检测方法。实验结果表明,该算法运行效率相对于串行计算模式有了较明显提升,且在某些场景下具有比VINS-Mono算法更高的定位精度和鲁棒性。
Aiming at improving the robustness,positioning accuracy and running efficiency of SLAM algorithm in weak texture scene,a fast monocular visual-inertial SLAM algorithm based on point and line feature is proposed.The initial operating parameters of SLAM system by fusing point-line features with inertial measurement information are estimated.At the back end of slam,a priori error,IMU measurement error,visual feature re projection error and loop detection error functions are constructed for nonlinear optimization to estimate more accurate camera motion,landmark coordinates and IMU state information.A parallel detection method based on CUDA parallel computing model is designed for point and line feature detection in the algorithm.The experimental results show that the algorithm runs more efficiently than serial computing mode and has higher positioning accuracy and robustness than VINS-Mono in some scenarios.
作者
刘建军
卢大威
胡雪花
王丽佳
Liu Jianjun;Lu Dawei;Hu Xuehua;Wang Lijia(College of Intelligent Manufacturing,Hebei Vocational University of Industry and Technology,Shijiazhuang 050091,China;College of Electronic Science,National University of Defense Technology,Changsha 410073,China)
出处
《国外电子测量技术》
北大核心
2022年第3期14-19,共6页
Foreign Electronic Measurement Technology
基金
中央引导地方科技发展资金(216Z1004G)
河北省冶金工业过程数字化控制技术创新中心研究项目(SG2021185)
河北工业职业技术大学博士基金(ZK202102)项目资助
关键词
同步定位与地图构建
单目视觉惯性系统
弱纹理环境
点线特征
并行计算
simultaneous location and mapping
monocular visual-inertial system
weakly textured environment
point and line features
parallel calculation
作者简介
刘建军,副教授,主要研究方向为视觉SLAM、视觉惯性组合导航等。E-mail:550292346@qq.com;卢大威,副教授,主要研究方向为信号处理、视觉SLAM等;通信作者:胡雪花,讲师,主要研究方向为计算机视觉,机器人导航。E-mail:506735160@qq.com;王丽佳,教授,主要研究方向为计算机视觉。
