Lidar-based Simultaneous Localization and Mapping in Dynamic Environments

In this paper, we propose a method of simultaneous localization and mapping (SLAM) based on Lidar, which can improve the accuracy of vehicle pose estimation in a dynamic environment. This method is composed of three modules. The first module is a Lidar odometry with static weight, namely Static Weight Normal Distribution Transform (SW-NDT). Static weight describes the probability that a point cloud belongs to a static object. To reduce the adverse effects of point clouds generated by dynamic objects on pose estimation, static weights are added to Normal Distribution Transform (NDT). The second module is back-end optimization. Scan Context is applied to detect whether a closed loop is formed between the current and historical frames. If a closed loop is detected, pose graph optimization is performed to optimize the poses of all key frames in the closed loop. The third module joins point clouds of the key frames to form a global map according to the optimized poses. For validation of the method proposed in this paper, KITTI dataset is utilized. The results show that the method proposed herein outperforms the other three methods in positioning accuracy.