• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2021, Vol. 57 ›› Issue (20): 206-214.doi: 10.3901/JME.2021.20.206

• 运载工程 • 上一篇    下一篇

扫码分享

基于传感器融合里程计的相机与激光雷达自动重标定方法

彭湃, 耿可可, 殷国栋, 庄伟超, 刘帅鹏, 徐利伟   

  1. 东南大学机械工程学院 南京 211189
  • 收稿日期:2021-01-21 修回日期:2021-04-25 出版日期:2021-10-20 发布日期:2021-12-15
  • 通讯作者: 殷国栋(通信作者),男,1976年出生,博士,教授,博士研究生导师。主要研究方向为先进电动汽车、车辆动力学与控制、智能无人汽车和车辆主动安全控制。E-mail:ygd@seu.edu.cn
  • 作者简介:彭湃,男,1993年出生,博士研究生。主要研究方向为智能车辆多模态融合感知。E-mail:13002573282@163.com
  • 基金资助:
    国家自然科学基金资助项目(51975118,52025121)。

Automatic Recalibration of Camera and LiDAR Using Sensor Fusion Odometry

PENG Pai, GENG Keke, YIN Guodong, ZHUANG Weichao, LIU Shuaipeng, XU Liwei   

  1. School of Mechanical Engineering, Southeast University, Nanjing 211189
  • Received:2021-01-21 Revised:2021-04-25 Online:2021-10-20 Published:2021-12-15

摘要: 智能驾驶车辆行驶环境复杂多样,不可避免地导致传感器相对位姿发生变化,此时需要进行重新标定。针对智能驾驶车辆的相机和激光雷达发生漂移后的重标定问题,提出一种基于传感器融合里程计的自动重标定方法。基于点云投影和图像配准原理建立基准点云和观测图像之间的3D-2D点对,利用N点透视投影得到平移尺度不准的相机运动;通过融合估计的激光雷达运动来恢复准确尺度的相机运动,并将基准点云根据相机运动转换到观测位置下,与观测点云通过点云配准求解变换矩阵,使用时域均值滤波得到最终的外参矩阵。基于智能驾驶车辆试验平台进行室内外实车试验。结果表明所提出的基于传感器融合里程计的方法无需设置标定板,不受环境和数据特征限制,能够实现相机和激光雷达的精确重标定,对传感器漂移具有较高的鲁棒性。

关键词: 智能驾驶车辆, 相机和激光雷达, 传感器漂移, 传感器融合里程计, 重标定

Abstract: The driving environment of the intelligent vehicle is complex and diverse, which inevitably leads to the change of the relative pose of the sensors. Aiming to solving the problem of recalibration after the drift of camera or LiDAR equipped on the intelligent vehicle, the automatic recalibration method using sensor fusion odometry is proposed. First, based on the principle of point cloud projection and image registration, the 3D-2D point correspondences between the reference point cloud and the observed image are established, and the camera motion with inaccurate translation scale is estimated by the Perspective-n-Point method; Then, the accurate camera motion is recovered by fusing the estimated LiDAR motion, the reference point cloud is transformed to the observation position using the camera motion, and registrated with the observation point cloud, the final extrinsic parameter is obtained using the average filtering method. Finally, the real vehicle experiments are implemented in indoor and outdoor scenario on an intelligent vehicle. The results show that the proposed sensor fusion odometry based method does not need to set up the calibration board, and is unrestricted with the environment and data features. It is able to achieve accurate recalibration of camera and LiDAR, and has high robustness to sensor drift.

Key words: intelligent vehicle, camera and LiDAR, sensor drift, sensor fusion odometry, recalibration

中图分类号: