多传感器融合的SLAM算法在林业智能化的应用与改进  

Application and Improvement of Multisensor Fasion SLAM Algorithm in Forestry Intelligence

作　　者：柳运昌 韩军宇 冯灵霄 刘远航 董蕊芳 Liu Yunchang;Han Junyu;Feng Lingxiao;Liu Yuanhang;Dong Ruifang(Beijing Forestry University,Beijing 100083,P.R.China)

机构地区：[1]北京林业大学,北京100083

出　　处：《东北林业大学学报》2025年第1期97-104,共8页Journal of Northeast Forestry University

基　　金：国家自然科学基金青年基金项目(62203059)。

摘　　要：同步定位与建图(SLAM)算法可实现机器人在未知环境下的定位,并对周围环境构建增量式地图,是机器人实现自主导航的基础。针对单一传感器(全球导航卫星系统(GNSS)、激光、相机等)的SLAM算法在林区环境存在定位精度低、构建的地图一致性较差的问题,提出了多传感器融合的SLAM算法在林业智能化的应用与改进。首先,联合使用激光雷达、单目相机、惯导对林区环境构建三维点云地图的同时对林业机器人进行实时定位;其次,提出以时间为约束的RS-T回环帧搜索法,进行候选回环帧搜索。结果表明:针对M2DGR数据集,RS-LVI-SAM算法与当前经典的SLAM算法对比,RS-LVI-SAM算法生成轨迹的均方根误差与标准差分别为0.09和0.04 m,定位精度最高;在林区场景中,RS-LVI-SAM算法建立的林区点云地图中侧柏的胸径与真实胸径的均方根误差与绝对平均误差分别为1.27、1.16 cm,建图效果最佳;生成的轨迹与真实轨迹的均方根误差与绝对平均误差分别为0.46、0.34 m,定位精度最高。因此,RS-LVI-SAM算法可以实现林区环境下的精准定位与林区点云地图构建,为机器人林区作业提供了技术支撑。The Simultaneous Localization and Mapping (SLAM) algorithm enables robots to locate themselves in unknown environments and to build incremental maps of their surroundings, forming the foundation for autonomous navigation. Due to issues such as low localization accuracy and poor map consistency associated with SLAM algorithms that rely on single sensors (such as GNSS, LiDAR, and cameras) in forest environments, a multi-sensor fusion SLAM algorithm, referred to as RS-LVI-SAM, is proposed. First, this approach combines LiDAR, a monocular camera, and inertial navigation to construct a 3D point cloud map of the forest environment while simultaneously providing real-time localization for forestry robots. Second, the RS-T loop closure frame search method, which is time-constrained, is introduced to perform candidate loop closure frame searches. The results showed that, for the M2DGR dataset, the RS-LVI-SAM algorithm achieves a root mean square error of 0.09 m and a standard deviation of 0.04 m for the generated trajectory, demonstrating the highest localization accuracy compared to current classical SLAM algorithms. In a forest scenario, the chest diameter of Platycladus orientalis in the point cloud map constructed by the RS-LVI-SAM algorithm shows an root-mean-square error and absolute mean error of 1.27 cm and 1.16 cm, respectively, indicating optimal mapping performance. Furthermore, the root-mean-square error and absolute mean error between the generated trajectory and the ground truth trajectory are 0.46 m and 0.34 m, respectively, reflecting the highest localization accuracy. Therefore, the RS-LVI-SAM algorithm enables precise localization and forest point cloud map construction in forest environments, providing technological support for robotic operations in forestry.

关 键 词：林业机器人 同步定位与建图 传感器 激光雷达 

分 类 号：S771.8[农业科学—森林工程]