Segmentation of Large Unstructured Point Clouds Using Octree-Based Region Growing and Conditional Random Fields

Abstract

Point cloud segmentation is a crucial step in scene understanding and interpretation. The goal is to decompose the initial data into sets of workable clusters with similar properties. Additionally, it is a key aspect in the automated procedure from point cloud data to BIM. Current approaches typically only segment a single type of primitive such as planes or cylinders. Also, current algorithms suffer from oversegmenting the data and are often sensor or scene dependent.<br><br> In this work, a method is presented to automatically segment large unstructured point clouds of buildings. More specifically, the segmentation is formulated as a graph optimisation problem. First, the data is oversegmented with a greedy octree-based region growing method. The growing is conditioned on the segmentation of planes as well as smooth surfaces. Next, the candidate clusters are represented by a Conditional Random Field after which the most likely configuration of candidate clusters is computed given a set of local and contextual features. The experiments prove that the used method is a fast and reliable framework for unstructured point cloud segmentation. Processing speeds up to 40,000 points per second are recorded for the region growing. Additionally, the recall and precision of the graph clustering is approximately 80%. Overall, nearly 22% of oversegmentation is reduced by clustering the data. These clusters will be classified and used as a basis for the reconstruction of BIM models.

DPRNet: Deep 3D Point Based Residual Network for Semantic Segmentation and Classification of 3D Point Clouds

“…It is necessary because raw point cloud of a scene containing multiple objects contains no structural information that yields the neighbourhood relationship, i.e., points stored in memory next to each other may belong to completely different and spatially distinct objects. Typically, such point ordering/structuring may be partially achieved by employing hierarchical data structures such as kd-tree [29], octree [53] etc., that build efficient computational graphs to explicitly capture the neighbourhood relationships. Without such explicit relationship (which exists in 2-D images or voxel representation), directly feeding the unstructured point clouds to the deep network would require the network to be invariant to N!…”

A Service-Oriented Indoor Point Cloud Processing Pipeline

“…Such methods are suitable for segmentation of both indoor and outdoor point clouds, and rely on either the spatial coordinates, pre-computed point normals or RGB color values of a given point set in order to calculate distinguishable regions of homogeneous point clusters. A notable method for segmentation that we make use of is called Region Growing, and relies on iterative sampling of neighbouring points for determining the segmented point clusters, making use of either pre-computed point normals or colors properties, which are iteratively searched within a voxelized data structure (Bassier et al, 2017). The success of region-growing segmentation approaches is largely dependent on the initial point sampling size, thresholds for color variance, and the accuracy of the normals estimation required to compare the regions.…”

Clustering-Based Plane Refitting of Non-planar Patches for Voxel-Based 3D Point Cloud Segmentation Using K-Means Clustering

“…Vo et al [14] improved a voxel-based region growing algorithm by using the surface normal in the weight measurements of the connections between adjacent voxels. Bassier et al [18] proposed another voxel-based region growing algorithm by using the surface normal and the RGB color values in two separate weight measurements. Xu et al [19] improved edge weight measurements composed with the surface normals, the vector between the barycenters, and the spatial distances of the voxels in their algorithm "Voxel and Graph-based Segmentation" (VGS).…”