复杂环境中的无人车多传感器紧耦合SLAM方法

The SLAM Method of Multi-Sensor Tight Coupling for UGV in Complex Situations

提出了一种多传感器信息紧耦合的无人驾驶车辆SLAM方法—Ligom,此方法在复杂环境中,可以实现无人驾驶车辆状态的实时估计与环境地图的构建。Ligom基于迭代扩展卡尔曼滤波(iEKF)理论设计,融合了惯性测量单元(IMU)、全球导航卫星系统(GNSS)、激光雷达(Lidar)与相机(Camera)等不同传感器。首先利用IMU预测车辆的先验状态与消除点云畸变,并使用滑动窗口方法保存点云特征以提高点云匹配精度;然后,通过对GNSS信息作初始化与坐标系变换,分别完成滤波更新得到后验状态;最后在后端环节引入关键帧进行状态优化与反馈更新,构建全局环境点云地图。此外,利用Camera-Lidar联合检测的多目标检测与跟踪算法,完成全局桩桶地图的构建。Ligom在四个不同平台与环境采集的数据集上进行了充分验证。

A SLAM method for Unmanned Ground Vehicle with Multi-Sensor tightly coupled,i.e.,Ligom,is presented,which achieves real-time UGV state estimation and map-building in complex situations.Ligom is designed based on Iterative Extended Kalman Filter(iEKF),combining different sensors,including IMU,GNSS,Lidar and Camera.Firstly,IMU is used to predict the prior state and de-skews point clouds,and sliding window approach is used to save the point cloud features to improve the matching accuracy.Then,through the initialization of GNSS and transformation of coordinate system,filter update is fultilled and the posterior state is obtained respectively.Finally,keyframes is introducted to optimize the state update feedback in back-end and build Global map.Moreover,the Multi target detection and tracking algorithm based on Camera-Lidar is used to build the map of cones.Ligom is extensively evaluated on datasets gathered from four platforms and environments.