【作者】 左炜； 修春娣；

【Author】 ZUO Wei;XIU Chundi;School of Electronic Information Engineering,Beijing University of Aeronautics and Astronautics;

【机构】 北京航空航天大学电子信息工程学院；

【摘要】 针对低速场景下的自动驾驶与移动机器人高精度实时定位与建图问题,提出了一种基于IESKF(迭代误差状态卡尔曼滤波)和因子图优化的激光/视觉/惯导紧耦合框架。基于IESKF的视觉激光惯性里程计融合了预积分约束、帧间激光约束与视觉重投影约束来形成状态误差的最大后验估计,因此在几何形状限制不足(影响激光雷达),或者是由激进的光照变化造成的无纹理区域(影响视觉)场景下具有一定的鲁棒性。后端采用因子图对视觉地图标志、相机位姿进行优化。此外,建图方法采用了ikd-tree点云存储的数据结构,有效降低了运行耗时。更多还原

【Abstract】 Aiming at the high-precision real-time positioning and mapping of autonomous and mobile robots in low-speed scenes, a tight coupling framework of laser/vision/inertial navigation based on IESKF(iterative error state Kalman filter) and factor graph optimization is proposed. The visual laser inertial odometer based on IESKF combines pre integration constraints, inter frame laser constraints and visual reprojection constraints to form the maximum a posteriori estimation of state error. Therefore, it has certain robustness in the case of insufficient geometric constraints(affecting lidar) or textureless areas(affecting vision) caused by radical lighting changes. The back-end uses factor graph to optimize the visual map signs and camera pose. In addition, the data structure of IKD tree point cloud storage is adopted in this method, which effectively reduces the running time.更多还原