单向结构信息辅助的无人机视觉惯性里程计OA
Structure Information Aided Visual/Inertial Odometry of UAVs for Unidirectional Structural Environment
针对地下综合管廊、隧道及长距离走廊等单向结构(Unidirectional structural,UDS)环境中光照不足、纹理特征稀疏等因素导致视觉惯性里程计特征提取不稳定、姿态误差持续累积和无人机(Unmanned aerial vehicles,UAVs)定位精度下降的问题,提出一种单向结构信息辅助的无人机视觉惯性里程计方法.该方法充分利用UDS环境沿纵向延伸、侧壁近似平行以及结构方向长期稳定的特点,将环境结构信息作为无人机姿态估计的外部约束,实现对传统视觉惯性里程计累计误差的持续修正.该方法通过采用自适应中值滤波与 BM3D(Adap-tive median filtering and BM3D,AMF-BM3D)去噪方法来提高特征提取精度,并引入一种惯性容错消失点技术,以实现高精度、无累积误差的无人机偏航角估计;同时利用重力约束来消除滚转角与俯仰角中的累积误差.最后,采用扩展卡尔曼滤波器(Extended Kalman filter,EKF)对传统视觉-惯性里程计(Visual-inertial odometry,VIO)获取的位姿数据进行融合,从而获得精细化的位姿估计结果.在UDS环境中的实验验证表明,与传统VIO方法相比,所提出的算法显著提升了UDS环境下的位姿估计精度及无人机飞行安全性.
The integrated utility system is indispensable for maintaining urban stability and services.Regular unmanned aerial vehicles(UAVs)inspections of these facilities are critical for ensuring operational urban systems.Such scenarios,characterized by spatially confined layouts and unidirectional extension,are defined as unidirectional structural(UDS)environment.Given the sparse textural features inherent and darker light to UDS environment,conventional methods fail to ensure safe UAV navigation.Therefore,this paper proposes a structure-assisted visual-inertial odometry algorithm for UAVs in dark UDS environment.The approach enhances the accuracy of camera feature extraction through an adaptive median filtering and BM 3D(AMF-BM3D)denoising method,and incorporates an inertial fault-tolerant vanishing point technique to achieve high-precision,cumulative-error-free UAV yaw estimation,while utilizing gravity constraints to eliminate cumulative errors in roll and pitch angles.Finally,the extended Kalman filter(EKF)fuses pose data from conventional visual-inertial odometry(VIO)to obtain refined pose estimation.Experimental validation in UDS environment demonstrates that compared to traditional VIO,the algorithm proposed significantly enhances the accuracy of pose estimation and the safety of UAV flight in the UDS environment.
谢盛弘;温烨贝;王洪枫;黄郑;季田阳;袁诚;吕品
南京航空航天大学自动化学院,南京 210016,中国南京航空航天大学自动化学院,南京 210016,中国北京自动化控制设备研究所,北京 100074,中国南京航空航天大学自动化学院,南京 210016,中国南京航空航天大学自动化学院,南京 210016,中国南京航空航天大学自动化学院,南京 210016,中国南京航空航天大学自动化学院,南京 210016,中国
航空航天
结构信息视觉惯性里程计扩展卡尔曼滤波
structure informationvisual-inertial odometry(VIO)extended kalman filter(EKF)
《南京航空航天大学学报(英文版)》 2026 (4)
546-559,14
This work was supported by the Basic Research Program of Jiangsu Province(No.BK20250175)and the National Natural Science Foundation of China(No.62273178).
评论