Fig 1.
The general architecture of the point cloud data segmentation.
Fig 2.
Framework of our proposed neighborhood selection method.
Fig 3.
(A) Curvature, (B) Omnivariance, and (C) Verticallity features of individual points from a portion of an urban area.
Fig 4.
An example of adding points to the neighborhood in the low omnivariance region.
(A) Fold points, (B) Edge points between planes. Here, red points represent Point Pi, and cyan points indicate the selected neighborhood.
Fig 5.
Simplicial complex-based neighborhood selection.
(A) Point Cloud Data, (B) Choosing the neighborhoods.
Table 1.
Eigenvalue-based features.
Table 2.
Elevation-based features.
Fig 6.
The LiDAR point cloud of ISPRS Vaihingen 3D benchmark dataset.
(A) Site 1 for training, (B) Site 2 and 3 for testing. The legend at the bottom indicates the segmentation labels rendered in colors.
Fig 7.
The Toronto-3D benchmark LiDAR point cloud dataset.
(A) Site L001 for training, (B) Site L002 for training, (C) Site L003 for training, (D) Site L004 for testing. The legend at the bottom indicates the segmentation labels rendered in colors.
Table 3.
Number of points per category in training and test sets of the Vaihingen area of ISPRS benchmark dataset.
Table 4.
Number of points (thousand) per category in training and test sets of the Toronto-3D dataset.
Fig 8.
A portion of Vaihingen point cloud data.
(A) Labeled ground truth, (B) Four distinct regions of the portion.
Fig 9.
Neighborhood selection from distinct region.
(A) Planar Region, (B) Vertical Region, (C) Low Omnivariance Region, and (D) High Omnivariance Region. The red point indicates any point Pi, and the cyan color indicates the corresponding selected neighborhood.
Table 5.
The accuracy, precision values, recall values, and F1-score values according to different neighborhood selection methods of the Vaihingen dataset.
Fig 10.
(A) Prediction map and (B) error map of the proposed method on the Vaihingen dataset.
Fig 11.
Confusion matrix for the machine learning classifier using proposed neighborhood approach on Vaihingen test dataset.
Fig 12.
Visualization of segmentation outcomes for site 2 and site 3 within the Vaihingen dataset employing various neighborhood retrieval approaches.
(A) Ground Truth, (B) Proposed Method, (C) Nong et al. (D) Xue et al. (E) He et al., (F) Günen, (G) Weinmann et al., (H) k=100, (I) k=50.
Fig 13.
(A) Prediction map, and (B) error map of our proposed method on the Toronto-3D dataset test area.
Fig 14.
Confusion matrix of the machine learning classifier using the proposed approach on Toronto-3D dataset.
Table 6.
The accuracy, precision values, recall values, and F1-score values comparison according to different state-of-the-art methods of the Toronto-3D dataset.