【作者】 李航宇； 李海； 张宪民； 林智鑫；

【Author】 Hangyu Li;Hai Li;Xianmin Zhang;Zhixin Lin;South China University of Technology,Guangdong Province Key Laboratory of Precision Equipment and Manufacturing Technique;

【机构】 华南理工大学,广东省精密装备与制造技术重点实验室；

【摘要】 针对传统激光建图定位方案在含动态目标场景下建图精度下降,且动态点云干扰帧间匹配准确性的问题,本文提出了一种基于运动遮挡的激光-imu建图定位方法。该方法首先使用同心圆区域分割出非地面点,通过误差状态卡尔曼滤波(IESKF)提供先验位姿,引入基于密度的聚类方法将非地面点聚为点云簇,利用运动遮挡法判断点云簇的属性,剔除属性为动态的点云簇,接着利用静态场景点云构建局部地图并提取描述子构建回环纠正定位偏差。本文将该算法在实际环境下与公开数据集下进行测试,实验结果表明,该算法能够成功识别并滤除点云数据中多个动态物体,提高了建图的质量,并减少了定位过程中产生的累积误差。更多还原

【Abstract】 Aiming at the problem that the traditional LIDAR SLAM scheme has decreased the accuracy of map building in scenes containing dynamic targets, and the dynamic point cloud interferes with the accuracy of inter-frame matching, this paper proposes a LIDAR-IMU SLAM method based on motion occlusion. The method firstly uses concentric circle region to segment the non-ground points, provides a priori bit position through error state Kalman filter(IESKF), introduces density-based clustering method to cluster the non-ground points into point cloud clusters, uses motion occlusion method to judge the attributes of point cloud clusters, excludes the clusters whose attributes are dynamic, and then constructs the local map using the static field point cloud and extracts descriptors to construct the loopback to correct the localization deviation. In this paper, the algorithm is tested under real environment and public dataset, and the experimental results show that the algorithm can successfully identify and filter out multiple dynamic objects in the point cloud data, which improves the quality of the map construction and reduces the cumulative error generated in the localization process.更多还原