基于对偶四元数的单目视觉/惯性组合导航算法

摘要

An algorithm of monocular camera/SINS integrated navigation is presented for structural road. In order to overcome the difficulties of point features matching and tracking, which are restricted by carrier velocity and camera view, line features are extracted from the lane marks on the road. the line features are represented by dual quaternion. Compared with the relative pose estimation algorithm based on dual quaternion, the proposed algorithm deduced the formula of image feature increment and camera position increment. After registration and rime synchronization, the velocity computation difference between the SINS and the camera was chosen as observation of integrated navigation. A Kalman filter was used to correct the integrated navigation error including the camera scale factor error of velocity measurement. The experiment results show that the proposed algorithm is accurate for structural road.%提出一种适用于结构化道路的单目视觉/惯性组合导航定位算法.针对点特征匹配和连续多帧追踪受车速和相机视野制约的不足,提取道路上车道线的直线特征,引入对偶四元数描述直线特征.在基于对偶四元数的相对位姿估计算法的基础上,推导了图像特征增量与相机位姿增量的表达式.通过配准和时间同步,用惯导系统和相机分别解算的载体速度之差作为组合导航的观测量,建立kalman滤波修正组合导航系统的误差,包括相机测速标度因数误差.车载实验结果表明在结构化道路上算法是有效的.