单目视觉里程计/惯性组合导航算法(英文)  

Algorithm for monocular visual Odometry/SINS integrated navigation

作　　者：冯国虎[1] 吴文启[1] 曹聚亮[1] 宋敏[1] 

机构地区：[1]国防科技大学机电工程与自动化学院,长沙410073

出　　处：《中国惯性技术学报》2011年第3期302-306,共5页Journal of Chinese Inertial Technology

基　　金：高等学校博士学科点专项科研基金(编号20069998009)

摘　　要：提出一种单目视觉里程计/捷联惯性组合导航定位算法。与视觉里程计估计相机姿态不同,惯导系统连续提供相机拍摄时刻对应的三维姿态,克服了单纯由视觉估计相机姿态精度低造成的长距离导航误差大的问题。通过配准和时间同步,用惯导系统解算的速度和视觉里程计计算的速度之差作为组合导航的观测量,采用Kalman滤波修正组合导航系统的误差,同时估计视觉里程计标度因数误差。分别在室内外不同环境下进行了22 m的推车实验和1412 m的跑车实验,定位误差分别为3.2%和4.0%。与Clark采用姿态传感器定期更新相机姿态估计结果的方法相比,单目视觉里程计/惯性组合导航定位精度更高,定位误差随距离增长率低,适合步行机器人或轮式移动机器人在复杂地形环境下车轮严重打滑时的自主定位导航。A new algorithm is presented for monocular visual odometry/SINS integrated navigation,in which the camera attitude is provided by the SINS other than visual estimation.The low accuracy of visual attitude estimation which would cause much error for long-range navigation is avoided.After registration and time synchronization,the velocity computation difference between SINS and visual odometry is chosen as observation of integrated navigation.A Kalman filter is used to correct the integrated navigation error including the visual odometry scale factor error.The 22 m indoor and 1412 m outdoor navigation experiments show that the position errors are 3.2% and 4.0% respectively.It can be seen that,compared with Clark＇s method in which camera attitude estimation is updated periodically with attitude sensors,the proposed algorithm is more accurate and robust with low rate of error growth.The proposed algorithm can be applied to the autonomous navigation of walking robots or wheeled mobile robots in case of serious wheel slip in complex terrain.

关 键 词：单目视觉里程计 捷联惯性组合导航系统 组合导航 标度因数 

分 类 号：U666.1[交通运输工程—船舶及航道工程]