基于集成式因子图优化的煤矿巷道移动机器人三维地图构建

3D map construction of coal mine roadway mobile robot based on integrated factor graph optimization

  • 摘要: 煤矿井下移动机器人作业精度严重依赖于同步定位与建图 (SLAM) 技术的准确性。井下长直巷道存在特征缺失、光照条件差等问题,导致激光里程计和视觉里程计易失效,因而限制了传统SLAM方法在煤矿巷道的有效应用,且目前SLAM方法的研究主要聚焦于多传感融合建图方法,较少关注激光 SLAM 方法建图精度的提升。针对上述问题,面向移动机器人在煤矿巷道的建图需求,提出了一种基于集成式因子图优化的煤矿巷道移动机器人三维地图构建方法,采用前端构建和后端优化的策略,设计了前端点云配准模块和基于滤波、图优化的后端构建方法,使建图结果更准确、适应性更强。针对煤矿长直巷道环境退化导致三维激光点云配准精度低的问题,融合迭代最近点 (ICP) 和正态分布变换 (NDT)算法,兼顾点云几何特征和概率分布特征,设计了集成式前端点云配准模块,实现了点云的精确配准。针对三维激光 SLAM 后端优化问题,研究了基于位姿图和因子图优化的后端构建方法,构建了集成 ICP和NDT 相对位姿因子的因子图优化模型,以准确估计移动机器人位姿。分别利用公开数据集 KITTI和模拟巷道点云数据集对三维地图构建方法在不同工况下的性能进行了实验验证。公开数据集 KITTI上的实验结果表明:在全局一致性上,该方法与传统基于特征点匹配的A−LOAM方法和基于平面分割及特征点提取的LeGO−LOAM方法具有相似的性能,在建图局部精度上优于其他2种方法。模拟巷道点云数据集上的实验结果表明:该方法具有显著优势,通过因子图优化,可得到一致性较高的三维地图,提升了煤矿巷道三维地图构建的精度及鲁棒性,解决了井下长直巷道特征点缺失、激光里程计失效的难题。

     

    Abstract: The working precision of mobile robots in coal mines seriously depends on the accuracy of simultaneous localization and mapping (SLAM) technology. There are some problems such as feature missing and poor lighting conditions in long and straight underground roadway. The problems lead to the failure of the laser odometer and visual odometer. The result limits the effective application of traditional SLAM method in coal mine roadway. At present, the research of the SLAM method mainly focuses on the multi-sensor fusion mapping method. There is a lack of research on the improvement of the mapping precision of the laser SLAM method. In order to solve the above problems, facing the mapping requirements of mobile robot in coal mine roadway, a 3D map construction method of coal mine roadway mobile robot based on integrated factor graph optimization is proposed. The method adopts the strategy of front-end construction and back-end optimization. The method designs a front-end point cloud registration module and a back-end construction method based on filtering and graph optimization. Therefore, the mapping result is more accurate and adaptable. The environmental degradation in coal mine long and straight roadway leads to the low registration precision of 3D laser point cloud. In order to solve the above problem, integrating iterative closest point (ICP) and normal-distributions transform (NDT) algorithms, taking into account the geometric characteristics and probability distribution characteristics of point clouds, an integrated front-end point cloud registration module is designed, which realizes the accurate registration of point clouds. Inview of the back-end optimization problem of 3D laser SLAM, the back-end construction method based on pose map and factor map optimization is studied. The factor map optimization model integrating ICP and NDT relative pose factors is constructed to accurately estimate the pose of the mobile robot. The performance of the proposed method of 3D map construction under different working conditions is verified by using the open dataset KITTI and the simulated roadway point cloud dataset. The experimental results on the open dataset KITTI show the following points. In terms of global consistency, this method has similar performance with the traditional A-LOAM method based on feature point matching and the LeGO-LOAM method based on plane segmentation and feature point extraction. It is superior to the other two methods in the local precision of mapping. The experimental results on the simulated roadway point cloud dataset show the following points. This method has significant advantages, through factor map optimization, a 3D map with high consistency can be obtained. The precision and robustness of 3D map construction of coal mine roadway are improved. The problems of the feature point missing and laser odometer failure in long straight underground roadway are solved.

     

/

返回文章
返回