基于激光雷达的果园树干检测

Laser radar based orchard trunk detection

为探索激光雷达在农业机器人环境理解和导航中的应用,研究一种基于改进DBSCAN算法的果园树干检测算法。该算法使用自适应密度阈值和聚类半径对不同距离处数据点进行聚类和整合,以克服DBSCAN算法对全局变量值敏感的缺点。针对激光雷达可能扫到地面造成机器人误检的问题,采用机器人航位推算模型计算当前帧数据中待定类的距离,通过与前一帧数据中对应类距离的比较判定待定类的类别,进而对地面干扰类进行排除。试验结果表明:1)机器人正常行走时本算法能够排除噪声准确识别树干类点;2)存在果树分枝或地面干扰时,有少量漏检,平均误判果树数目为-0.13棵,能够区分出地面类和果树类。该研究可以应用到农业机器人果园环境理解和导航中。

Aiming at exploring the laser radar application in environment understanding and navigation of agricultural robot,an algorithm for orchard trunk detection based on improved DBSCAN algorithm is studied in this research.To overcome the defects that DBSCAN algorithm is sensitive to global variables,the algorithm adopts an adaptive density threshold and clustering radius to cluster and integrate data points at different distances.For avoiding false detection when laser radar sweep over the ground,robot dead reckoning model is used to calculate distance of unidentified class in current frame,and then compare with the distance of corresponding class in previous frame to category the unidentified class,finally to exclude the ground interference.The results show ：1）The algorithm can exclude noise and identify the trunk points when robot is walking smoothly;2）There is certain extent of missed detection of trees since tree branch or ground interference.The average false detection number was-0.13,and the algorithm is able to distinguish trees out of ground objects.The study can be applied in agricultural robot environment understanding and navigation in orchard.