Refine
Document Type
Language
- English (3)
Keywords
- Plane detection (1)
- Point cloud (1)
- Region growing (1)
- Segmentation (1)
- Unorganized point clouds (1)
- Wall detection (1)
- point clouds (1)
- registration (1)
- segmentation (1)
Institute
Plane detection is an important step in the reconstruction of 3D models of buildings from point clouds. The results of plane detection methods based on the region growing approach mainly depend on the choice of seed points. In this study, we introduce a novel region growing-based method for plane detection in unorganized point clouds. Our method uses the Iterative Closest Point (ICP) algorithm to extract reliable seeds. To enhance the performance and the quality of the results, we used voxel grids representation of the point clouds in the growing process. The classification of the candidate planes is improved by using the number of voxel cells covering accumulated segments. The method is deterministic, runs in O(nlog(n)), and does not take into account the orientation of the point clouds. The results of plane detection using the proposed method on a benchmark consisting of 9 point clouds of buildings show a better precision of extracted planes compared to those obtained with 3-D KHT and PCL-RANSAC. Although slower than 3-D KHT, our method requires an execution time (3 x times) shorter than PCL-RANSAC.
Volumetric wall detection in unorganized indoor point clouds using continuous segments in 2D grids
(2022)
The quality of 3D models of existing buildings reconstructed from point clouds is strongly related to the segmentation process used to detect structural elements. A new wall detection method in the indoor point clouds of buildings is presented in this study. The point clouds are segmented into horizontal layers, and a concept of continuous segments in a 2D grid representation is used to extract the footprints of the wall structures, and 2D blocks are projected into 3D space to obtain the wall segments in the initial 3D point cloud. The results obtained from the execution of the proposed method demonstrate that wall blocks in indoor point clouds are detected independently of their shape. Executing the proposed method on a set of 9 in-door point clouds revealed better performance in terms of result quality and execution time compared to RANSAC. The robustness of the method can be improved by adding a classification step to eliminate non-consistent blocks.
Point cloud registration combines multiple point cloud data sets collected from different positions using the same or different devices to form a single point cloud within a single coordinate system. Point cloud registration is usually achieved through spatial transformations that align and merge multiple point clouds into a single globally consistent model. In this paper, we present a new segmentation-based approach for point cloud registration. Our method consists of extracting plane structures from point clouds and then, using the 4-Point Congruent Sets (4PCS) technique, we estimate transformations that align the plane structures. Instead of a global alignment using all the points in the dataset, our method aligns 2-point clouds using their local plane structures. This considerably reduces the data size, computational workload, and execution time. Unlike conventional methods that seek to align the largest number of common points between entities, the new method aims to align the largest number of planes. Using partial point clouds of multiple real-world scenes, we demonstrate the superiority of our method compared to raw 4PCS in terms of quality of result (QoS) and execution time. Our method requires about half the execution time of 4PCS in all the tested datasets and produces better alignment of the point clouds.