相似场景下基于局部地图的激光SLAM前端算法改进

Improvement of LiDAR SLAM Front-end Algorithm Based on Local Map in Similar Scenes

在走廊、隧道等相似场景下,传统激光SLAM(同步定位与地图创建)算法由于观测数据的相似性,算法性能将严重劣化,甚至完全失效。为解决该问题,本文在hdl graph slam算法的基础上,首先基于匀速运动假设改进了运动预测模型,获得了更准确的初始位姿估计;然后通过引入局部地图概念实现点云的稠密化,改善了相似场景下前端里程计的性能。在室内实验中,场景的还原度达到了99.54%,较改进前提高了57.25%;在室外实验中,里程计漂移由原先的111.62 m降至7.65 m。实验结果表明,提出的算法在室内和室外的相似场景中均能带来显著的性能提升。

In the corridor,tunnel and other similar scenes,performances of traditional LiDAR SLAM(simultaneous localization and mapping)algorithms will seriously degrade,and the algorithms might even be completely invalid due to the similarity of observation data.To solve this problem,the motion prediction model is improved firstly with the hdl graph slam algorithm based on the assumption of uniform motion to obtain a more accurate initial pose estimation.Then,the concept of local map is introduced to densify the point cloud,and the performances of the front-end odometer are improved in the similar scenes.In the indoor experiment,the average restoration rate of the scene reaches 99.54%,which is 57.25%higher than that before improvement.In the outdoor experiment,the odometer drift is reduced from 111.62 m to 7.65 m by the improved algorithm.The experimental results show that the proposed algorithm can bring a significant performance improvement in both indoor and outdoor similar scenes.