单 位：
重庆市农业科学院农业机械研究所;西南大学工程技术学院
关键词：
树干检测;激光雷达;相机;深度图;传感器融合
摘 要：
针对传统传感器在树干检测中的局限性和单一性,提出一种基于激光雷达与相机融合的树干检测方法.首先,利用深度图对激光雷达点云进行处理,实现地面点云去除以及树干点云聚类,并在聚类中设置横、纵向自适应阈值,去除聚类中墙体、杂草、树叶等多余信息;然后,利用YOLOv3算法对相机图像进行分析,基于树干特征实现目标识别并返回检测框与类别信息;最后,基于交并比方法(IoU)对2种传感器的检测结果进行融合,识别树干并返回其三维信息与位置信息.以无人割草机为载体开展场地测试,实验结果表明:融合算法的树干检测准确率在93.1%左右,树干定位横、纵向平均误差分别为0.075 m和0.078 m,能够满足无人割草机的树干检测要求,为智能农机的环境感知提供了一种新的方法.
译 名：
Trunk Detection Method Based on Fusion of Lidar and Camera Data
关键词：
trunk detection%lidar%camera%depth map%sensors fusion
摘 要：
Aiming at the limitations and singleness of traditional sensors in tree trunk detection,a new tree trunk detection method based on data fusion between lidar and camera is proposed.Firstly,the depth map was used to process lidar point cloud to remove ground point cloud and cluster the tree trunks point cloud,and set the horizontal and vertical adaptive thresholds in the clustering process to remove excess informa-tion such as walls,weeds and leaves.Then,YOLOv3 algorithm was applied to analyze the camera image,which realized target recognition based on tree trunks feature and returned the detection frame and catego-ry information.Finally,the detection results of two sensors were fused based on IoU to identify the tree trunks and return their three-dimensional information and location information.The field test was carried out with the unmanned lawn mower as the carrier.Experimental results show that tree trunk detection ac-curacy of the proposed data fusion algorithm was about 93.1%,average horizontal and longitudinal errors of the tree trunk positioning were 0.075 m and 0.078 m respectively,which can meet tree trunk detection requirements of the unmanned lawn mower and provide a new method for environment perception of intel-ligent agricultural machinery.
