融合车辆动力学的双目视觉惯性SLAM研究  

Study on stereo visual inertial SLAM combining vehicle dynamics

作　　者：余卓平[1,3] 鞠然 韩燕群 赵君峤 YU Zhuoping;JU Ran;HAN Yanqun;ZHAO Junqiao(School of Automotive Studies,Tongji University,Shanghai 201804,China;School of Electronics and Information Engineering,Tongji University,Shanghai 201804,China;Nanchang Automotive Innovation Institute,Nanchang 330052,China)

机构地区：[1]同济大学汽车学院,上海201804 [2]同济大学电子与信息工程学院,上海201804 [3]南昌汽车创新研究院,江西南昌330052

出　　处：《华中科技大学学报（自然科学版）》2022年第11期85-89,95,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(41801335,41871370)。

摘　　要：针对地下车库环境中无人驾驶汽车视觉同时定位和建图(SLAM)定位精度低的问题,提出一种融合惯性测量单元(IMU)角速度信息和车辆动力学信息的预积分方法.以IMU频率进行旋转预积分,以车辆动力学频率进行平移预积分.在平移预积分的计算中引入角速度信息,使其可以表达非平面运动.首先使用李代数和旋转群推导了相关的预积分公式、雅可比及噪声状态转移方程;然后以此预积分为基础将车辆动力学信息融合到双目视觉惯性SLAM中,以提高定位精度.地下车库实车实验表明:该方法将双目视觉惯性ORB-SLAM3的平均定位精度提高了32%.Aiming at the problem of low accuracy of unmanned vehicle visual simultaneous land and mapping(SLAM) in underground garage,a pre-integration method combining inertial measurement unit(IMU) gyroscope measurement and vehicle dynamics measurement was proposed. The rotation pre-integration was performed at IMU frequency,and the translation preintegration was performed at vehicle dynamics measurement frequency. The gyroscope measurement was introduced into the calculation of translation pre-integration,so that it could express non planar motion. First,the pre-integration equation,Jacobian matrix and noise state transition equation were derived by Lie algebra and rotation group. Then,based on this pre-integration,the vehicle dynamics information was incorporated into the stereo visual inertial SLAM,so as to improve the accuracy.Experiment of underground garage show that the proposed method improves the average accuracy of the stereo visual inertial ORB-SLAM3 by 32%.

关 键 词：无人驾驶汽车定位 同时定位和建图(SLAM) 多传感器融合 车辆动力学 预积分 

分 类 号：P283[天文地球—地图制图学与地理信息工程] TP242[天文地球—测绘科学与技术]