结合超体素与PFCM的点云分割方法

Point cloud segmentation method combining supervoxels and PFCM

为了实现点云数据的区域划分,提出了一种结合超体素与粒子群优化模糊C均值(PFCM)的聚类分割算法(SPFCM)。用随机采样一致性算法去除点云平面,根据3-D点云的空间位置、曲率以及快速直方图特征,利用八叉树体素化点云得到超体素。采用PFCM算法对超体初步划分,并对粘连的点云再划分,克服了PFCM算法对于堆叠物体无法分割及较大物体过分割的缺点,并在OSD-v0.2数据集上对SPFCM算法进行了性能测试。结果表明,相较于PFCM算法,SPFCM不仅保留了其参量少、操作简单等优点,而且指标得到了较大提升,准确率达到86%,查全率达到83%。该研究对3-D点云复杂场景的准确分割提供了帮助与参考。

In order to realize the area division of point cloud data,a segmentation algorithm(SPFCM)combining supervoxels and particle swarm optimization fuzzy C-means(PFCM)was proposed.A random sampling consensus algorithm was used to remove the point cloud plane.According to the spatial position,curvature and fast point feature histogram characteristics of the 3-D point cloud,the octree voxelization point cloud was used to obtain the supervoxel.The PFCM algorithm was used to preliminarily divide the superbody and subdivide the connected point cloud,which overcomes the shortcomings of the PFCM algorithm for stacking objects and over-segmentation of larger objects.The performance test of the SPFCM algorithm was performed on the OSD-v0.2 data set.The experimental results show that compared with the PFCM algorithm,it not only retains its advantages such as fewer parameters and simple operation,but also the index has been greatly improved,and the accuracy is up to 86%,while the recall rate reaches 83%.This research provides help and reference for the accurate segmentation of complex scenes in 3-D point clouds.