动态环境下无人地面车辆点云地图快速重定位方法

The use of on-board sensors to determine the position of unmanned ground vehicle in a map without global positioning system (GPS) signals is the key to realize the autonomous driving. The existing methods are mainly based on 2-D grid maps, and are not suitable for complex environment. A fast localization algorithm, which is based on 3D point cloud map in dynamic environment and uses a clustered bag-of-words model, is proposed. The dynamic obstacles are removed by backtracking the continuous multi-frame lidar point cloud data and using Bayesian formula. The word list and database are constructed by clustering the point clouds and extracting the viewpoint feature histogram (VFH) descriptors, which enables the unmanned ground vehicle to quickly find the starting position in the point cloud map. Then the subsequent precise positioning is achieved by using the lidar odometry and mapping (LOAM) real-time algorithm. The experimental results show that the proposed algorithm can be used to accurately localize an unmanned ground vehicle in the 3D point cloud map in dynamic environment and meet the real-time requirements.