基于改进欧式聚类的点云分割方法

Point Cloud Segmentation Based on Improved Euclidean Clustering

随着3D相机以及点云处理的兴起,点云分割技术已经被广泛应用到工业CAD/CAM、激光遥感等领域中。本文在开源PCL库的基础上,针对3D相机获取的点云信息,依据点云分布情况实现点云分割。该算法首先利用预处理方法对原始点云进行去噪处理,然后对余下点云利用随机采样一致性(RANSAC)算法拟合平面,并去除该平面模型包含的点云,最后利用改进后的欧式聚类分割算法对去除点云模型后的数据构建KD树,利用平滑度重新定义聚类方式,通过迭代得到不同物体的点云子集,在实现点云分割的同时还可以有效去除噪声点。实验通过对多个物体点云数据进行分割,结果表明,该算法不仅可以有效分割场景点云中的平面结构,而且能够准确地分割出不同的物体,满足了工业机器人抓取的实时性要求。

With the rise of 3 D camera and point cloud processing, point cloud segmentation technology has been widely used in industrial CAD/CAM, laser remote sensing and other fields.Based on the open source PCL library, this paper realizes point cloud segmentation according to the distribution of point cloud information obtained by 3 D camera.The algorithm first uses the pre-processing method to de-noise the original point cloud.Then random sample consensus(RANSAC) algorithm was used to fit the plane and remove the point clouds contained in the plane model.Finally, the improved Euclidean Clustering segmentation algorithm is used to construct KD tree for the data after removing the point cloud model, and the clustering method is redefined by using smoothness.Point cloud subsets of different objects are obtained through iteration, which can effectively remove noise points while realizing point cloud segmentation.The experimental results show that the algorithm can not only effectively segment the plane structure in the scene cloud, but also accurately segment different objects, which meets the real-time requirements of industrial robot grasping.