一种融合视觉与IMU的车载激光雷达建图与定位方法  

A method of map construction and location using lidar and combining vision and IMU information

作　　者：林晨浩 彭育辉[1] LIN Chenhao;PENG Yuhui(School of Mechanical Engineering and Automation,Fuzhou University,Fuzhou,Fujian 350108,China)

机构地区：[1]福州大学机械工程及自动化学院,福建福州350108

出　　处：《福州大学学报（自然科学版）》2022年第1期82-88,共7页Journal of Fuzhou University(Natural Science Edition)

基　　金：福建省科技厅产学合作重大项目(2017H6007)。

摘　　要：针对激光雷达非匀速运动畸变问题,提出一种融合视觉惯性里程计和激光雷达里程计,进行三维地图构建与定位(simultaneous localization and mapping,SLAM)方法.经预处理和时间戳对齐后的数据,应用视觉估计和惯性测量单元(inertial measurement unit,IMU)预积分对视觉进行初始化,通过约束的滑窗优化和视觉里程计的高频位姿,将传统雷达匀速运动模型改进为多阶段匀加速模型,从而降低点云畸变.同时,利用列文伯格-马夸尔特(Levenberg-Marquardt,LM)方法优化激光里程计,提出一种融合词袋模型的回环检测方法,最终实现三维地图构建.基于实车试验数据,通过与LEGO-LOAM(lightweight and ground-optimized lidar odometry and mapping on variable terrain)方法的结果对比,本文方法在平均误差和误差中位数上分别提升了16%和23%.Aiming at the problem of non-uniform motion distortion of lidar,this paper proposes a method of combining visual-inertial-odometer and laser-odometer to simultaneous localization and mapping.After preprocessing and timestamp aligned data,vision estimation and IMU pre-integration are applied to initialize the vision.Through constrained slide-window optimization and high frequency pose of visual odometer,the traditional lidar uniform motion model is improved to multi-stage uniform acceleration model,so as to reduce point cloud distortion.At the same time,the Levenberg Marquardt method is used to optimize the laser-odometer,and a loop detection method based on the bag of words model is proposed to realize the construction of 3D map.Based on the real vehicle test data,the average error and median error are increased by 16%and 23%respectively by comparing the results of the proposed method and LEGO-LOAM method.

关 键 词：激光雷达 融合视觉 IMU预积分 同时定位与建图 滑窗优化 

分 类 号：TN958.98[电子电信—信号与信息处理]