Abstract: In order to solve the problem of under-segmentation and over-segmentation such as missed detection and false detection of target objects under multi-size and multi-distance conditions in the process of Lidar detection under hilly and mountainous orchard conditions, a target object detection method based on adaptive target clustering radius of Lidar is proposed. Firstly, by using Lidar to sense the three-dimensional point cloud of the surrounding environment, the ground point cloud is removed and the preprocessing of down sample is performed by voxel filter. The amount of data is reduced and the noise points in the point cloud is removed. Secondly, the K-d tree model is established and the nearest neighbor search is carried out to accelerate the process of Euclidean clustering. By adaptively determining the clustering radius of each crown, the Euclidean clustering can get better clustering results. Finally, in order to verify the accuracy and practicability of the algorithm, based on the orchard tracked vehicle platform, a 32-line Lidar is used to test the algorithm. The results show that the algorithm can accurately cluster the canopy point cloud of fruit trees in hilly and mountainous orchards, and the field target detection rate is 94.41%.

Key words: Lidar, crown detection, K-dimensional tree model, adaptive clustering

摘要： 为解决丘陵山地果园条件下激光雷达检测过程中面对多尺寸、多距离条件下出现的目标物体漏检、误检等欠分割和过分割问题,提出一种基于激光雷达的自适应目标聚类半径目标物体检测方法。首先,在使用激光雷达感知到周围环境的三维点云后,去除地面点云并且使用体素滤波进行降采样的预处理,在减少数据量的同时去除点云中的噪声点。其次,建立K-d tree模型进行最近邻搜索,以加速欧式聚类的进程,通过自适应确定每颗树冠的聚类半径,使欧式聚类能够得到更好的聚类效果。最后为验证算法准确性和实用性,基于果园履带车平台,采用32线激光雷达对所提算法进行实车测试。结果表明:在丘陵山地果园中该算法可准确聚类果树树冠点云,且实地目标正检率为94.41%。

关键词: 激光雷达, 树冠检测, K-d tree模型, 自适应聚类