基于3D LiDAR的矿山无人驾驶车行驶边界检测  

Boundary detection of mine drivable area based on 3D LiDAR

作　　者：陈龙 司译文[1] 田滨 檀祖冰[1] 王永涛 CHEN Long;SI Yiwen;TIAN Bin;TAN Zubing;WANG Yongtao(Vehicle Intelligence Pioneers Inc.,Qingdao 266109,China;School of Data and Computer Science,Sun Yat-sen University,Guangzhou 510006,China;The State Key Laboratory of Management and Control for Complex Systems,Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China)

机构地区：[1]青岛慧拓智能机器有限公司,山东青岛266109 [2]中山大学数据科学与计算机学院,广东广州510006 [3]中国科学院自动化研究所,复杂系统管理与控制国家重点实验室,北京100190

出　　处：《煤炭学报》2020年第6期2140-2146,共7页Journal of China Coal Society

基　　金：国家自然科学基金青年基金资助项目(61503380)。

摘　　要：路沿检测是矿区无人驾驶的关键技术之一,所得路沿信息可用于辅助无人矿车的感知、规划和定位。准确的路沿检测与路沿图构建同时也是高精地图构建中的首要步骤,通过机器学的方法可以在现有路沿点的基础上初步构建道路结构拓扑。与传统的卡尔曼框架下的路沿跟踪所不同的是,笔者提出了基于占用栅格思想的路沿跟踪方案。所谓占用栅格,即根据传感器感知到的障碍物信息来对全局范围内的场景进行构建的一种方法。由于矿区环境下的路沿具有高度不确定性,所以需要用一种由包含概率信息的模型找出最高可能性是路沿的位置,而基于八叉树的结构正好可以满足这一需求。笔者通过将每一帧点云实时栅格化来筛选路面候选栅格,然后以环压缩为主进行多特征的路沿检测,其中栅格是一种应用已经非常广泛的点云结构,根据矿山道路本身具有的粗粒度特性,将特征检测扩展到以每一个扇形单元格为不可分割的基本单元,在提高了检测效率的同时,也避免了很多路面凹凸所带来的干扰。路沿的跟踪策略采用基于八叉树的三维占用栅格,融合多帧检测结果,建立全局路沿地图,将传统的二维的占用栅格图的概念扩展到了三维的仅包含路沿属性的全局障碍物图。实际矿区道路检测结果表明,笔者提出的针对矿山非结构化道路的路沿检测算法可以准确地实现对矿山道路边界的检测,并可以满足无人矿车驾驶的实时性。Road boundary detection is one of the key technologies for unmanned driving in mining areas.The obtained road boundary information can be used to assist the perception,planning and positioning of unmanned mining vehicles.Accurate road boundary detection and road boundary map construction are also the first step in the construction of high-precision maps.The road structure topology can be preliminarily constructed on the basis of existing road boundary points through machine learning methods.Different from the traditional Kalman frame of road boundary tracking,this paper proposes a road boundary tracking scheme based on the idea of occupied grid.The so-called occupancy grid refers to a method for constructing scenes in the global scope based on obstacle information sensed by sensors.Due to the high uncertainty of road boundarys in mining areas,a model containing probability information is needed to find the location of the road boundary.The structure based on the octree can meet this demand.This paper screens the candidate road grids by rasterizing each frame of the real-time point cloud,and then performs a multi-feature road boundary detection based on ring compression.Among them,the grid is a point cloud structure that has been widely used.The coarse-grained nature of the mine road itself extends the feature detection to each fan-shaped cell as an inseparable basic unit,which improves the detection efficiency and avoids the interference caused by many road bumps.The road boundary tracking strategy uses a three-dimensional occupancy grid based on an octree,merges the multi-frame detection results,and establishes a global road boundary map,extending the concept of the traditional two-dimensional occupancy grid map to a three-dimensional road boundary attribute global obstacle map.The actual mine road detection results show that the proposed road edge detection algorithm for mine unstructured roads can accurately detect the mine road boundaries and can meet the requirements on the real-time driving of unmanned mine

关 键 词：矿山 3D LIDAR 路沿检测 建图 无人驾驶 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TD178[自动化与计算机技术—控制科学与工程]