Automatic region-growing system for the segmentation of large point clouds

Recognizing geometric primitives in 3D point clouds of mechanical CAD objects

“…Being the method able to handle only two primitive types, its applicability is however restricted. To handle the increasing availability of acquired data, [1] introduces a region-growing-based system for the segmentation of large point clouds in planar regions. Other approaches, devised to detect only specific types of primitives, are: [21], which deals with quadric surfaces; [22] which fits surfaces of revolution; and [23], which extracts cylindrical shapes from non-oriented point clouds.…”

“…To give an example, it is highly convenient to recognise and reconstruct a digital model so that it can be interpreted in terms of some basic components and easily manipulated by CAD systems. A large number of points, the presence of noise and outliers, the occurrence of missing or redundant parts and the non-uniform distribution of the data severely limit the use of tessellations (e.g., meshes) as a means to ease the analysis and reconstruction of shapes; rather, they make it more convenient to analyze the point cloud directly [1].…”

“…In our implementation, peaks that correspond to primitives are automatically recognised by keeping the local maxima having a persistence higher than 10% of the maximum value of H, by using the algorithm for persistent maxima proposed in[39]. The coordinates of the cell centres of the maxima correspond to the parameters of potentially recognised surface primitives 1. The notion of topological persistence was introduced in[37] for encoding and simplifying the points of a filtration f by classifying them as either a feature or noise depending on its lifetime or persistence within the filtration.…”

Breast Ultrasound Image Segmentation Algorithm Using Adaptive Region Growing and Variation Level Sets

“…A hybrid scheme combining region-based and boundary-based techniques for breast ultrasound image segmentation was developed using these methods, in which seed points are automatically generated by an empirical rulebased formula, and boundary points are thereafter determined by region growing (RG) and directional gradient operation, which combines the clustering algorithm used in computed tomography detection [8]. Compared to RG, adaptive region growing (ARG) is more e ective in segmentation [9], improved genetic algorithms [10], enhanced seeded region growing [11], automated inspection [12], magnetic resonance imaging [12], and other types of imaging [13], and it also combines clustering [14], statistical characters, and point cloud images [15]. In addition, Markov random field models have been used for breast ultrasound image segmentation.…”

Digitalization based on high-resolution scanning and HBIM tools for damage assessment of the José de Alencar house

“…Nowadays, approaches reported in the literature are developing studies to automate the processes of segmentation, classi cation, and modeling of point clouds resulting in more accurate three-dimensional surfaces of building elements [30,31] and to support an approach of damage interpretation and structural assessment based on a machine applying Semantic Web technologies [32], respectively. Works involving capturing large-scale data from facades using TLS [33] are also being monitored, as the comparison of precision between low-cost and easy-to-use scanners, and state-of-the-art laser equipment, in the acquisition of point clouds of buildings.…”