多传感器融合的割草机器人障碍物检测方法  

Obstacle detection method of lawn mowing robot based on multi-sensor fusion

作　　者：李忠利[1] 马理想 韩冲 王帅 LI Zhongli;MA Lixiang;HAN Chong;WANG Shuai(Henan Provincial Key Laboratory of Automobile Energy Saving and New Energy,Luoyang,Henan 471003,China;College of Vehicle and Traffic Engineering,Henan University of Science and Technology,Luoyang,Henan 471003,China)

机构地区：[1]河南科技大学河南省汽车节能与新能源重点实验室,河南洛阳471003 [2]河南科技大学车辆与交通工程学院,河南洛阳471003

出　　处：《江苏大学学报（自然科学版）》2024年第2期160-166,共7页Journal of Jiangsu University：Natural Science Edition

基　　金：拖拉机动力系统国家重点实验室开放课题(SKT2021005)。

摘　　要：为了提高割草机器人自主作业时的环境感知能力,提出一种基于相机和低成本固态激光雷达融合的障碍物检测方法.首先,基于改进的DBSCAN聚类算法,得到一种可自适应确定聚类参数的KANN-DBSCAN算法,利用该算法对固态激光雷达采集到的三维点云聚类分析,得到障碍物点云并通过相机和固态激光雷达联合标定结果投影到二维图像上;其次,基于SSD目标检测网络完成障碍物样本训练,并对图像信息进行检测识别;最后,为了避免因光线不足或远距离雷达点云稀疏聚类困难,导致视觉或雷达检测性能受限,提出一种优势互补的目标级信息融合策略.试验结果表明:所提信息融合策略在融合两传感器检测结果的基础上,有效避免了环境条件改变单一传感器检测性能受限时,环境感知出现的漏检和误检,经过信息融合后的障碍物综合检出率约为87.5%,相较于单一传感器具有很大的提高,使环境感知信息更加全面可靠.To improve the environmental perception ability of lawn mowing robot during autonomous operation,the obstacle detection method based on the fusion of cameras and low-cost solid-state lidar was proposed.Based on the improved DBSCAN clustering algorithm,the KANN-DBSCAN algorithm was proposed to adaptively determine the clustering parameters.By the algorithm,the 3D point cloud collected by the solid-state lidar was clustered and analyzed,and the obstacle point cloud was obtained and passed through the camera.The results of joint calibration with solid-state lidar were projected onto 2D.The obstacle sample training was completed based on the single short multibox detector(SSD)target detection network,and the image information was detected and recognized to complete the camera-based obstacle detection.To avoid the limited visual or radar detection performance due to the insufficient light or the difficulty of sparse clustering of long-distance radar point clouds,the target-level information fusion strategy with complementary advantages was proposed.The experimental results show that based on the fusion of the detection results of the two sensors,the proposed information fusion strategy can be used under the change of environmental conditions.When the detection performance of single sensor is limited,the missed detection and false detection of environmental perception can be effectively avoided,and the comprehensive detection rate of obstacles after information fusion is about 87.5%,which is significantly improved compared to single sensor and makes the environmental perception information more comprehensive and reliable.

关 键 词：割草机器人 固态激光雷达 KANN-DBSCAN聚类算法 SSD目标检测 目标级融合 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]