基于GPS的先验地图辅助视觉惯性定位方法

Visual-inertial positioning method based on GPS prior map assisted

为提高视觉惯性导航系统定位精度,解决其航向缺失且无法提供绝对地理位置的问题,提出一种利用GPS生成先验地图来辅助视觉惯性同时定位与地图构建(SLAM)方法。将每一帧图像采集点的GPS地理坐标投影成绝对高斯坐标,与视觉关键帧信息同步,构建先验地图。利用词袋模型进行闭环检测,由视觉重投影误差因子、IMU残差因子、先验图像重投影误差因子共同构建目标优化函数,基于最小二乘法进行位姿估计与优化。通过地面车载和低空机载实验对该方法进行了验证,实验结果表明:相比于无先验地图的定位方法,x、y、z方向的平均定位误差分别降低了68.2%、75.1%、46.4%以上;相比于有先验地图(未融合GPS)的定位方法,x、y、z方向的平均定位误差分别降低34.9%、51.5%、19.4%以上,有效提高了视觉惯性导航系统在大范围环境中的定位精度。

In order to improve the localization accuracy of the visual-inertial navigation system,and solve the problem of missing heading information and unable to provide absolute geographic location,a method of visual-inertial simultaneous localization and mapping assisted by prior map generated from GPS is proposed.The GPS geographic coordinates of each frame of image acquisition point are projected into absolute Gaussian coordinates and synchronized with the visual key frame information to build a prior map.The closed-loop detection is carried out by using the bag of words model.The objective optimization function is constructed by the visual reprojection error factor,IMU residual error factor and prior image reprojection error factor,and the least square method is used for the pose estimation and optimization.The proposed method is verified by ground vehicle and low-altitude airborne experiments.The results show that compared with the localization method without prior map,the average location errors in x,y and z directions are reduced by more than 68.2%,75.1%and 46.4%respectively.And compared with the localization method with prior map(without GPS),the average location errors in x,y and z directions are reduced by more than 34.9%,51.5%and 19.4%respectively,which improves the localization accuracy of the visual-inertial navigation system in a large-scale environment effectively.