Fig 1.
GeoSLAM ZEB horizon multi-function LiDAR scanner.
Fig 2.
Local feature integration.
Table 1.
Pseudo-code for BN layer implementation.
Fig 3.
Building a semantic segmentation network architecture based on RandLA-Net (FC is Fully Connected Layer, LFA is Local Feature Aggregation, RS is Random Sampling, MLP is Shared Multi-Layer Perceptron, US is Upsampling, DP is Dropout Layer).
Fig 4.
Visual interface of the system.
Fig 5.
Point cloud data before and after preprocessing.
(a) Pre-processed point cloud data. (b) Point cloud data after pre-processing.
Fig 6.
Before and after comparison of the final visualization results of the semantic segmentation module.
(a) Trimble Realworks manual semantic annotationTrimble Realworks. (b) Semantic segmentation results.
Table 2.
Running effect of point cloud semantic segmentation module.
Fig 7.
Original point cloud model.
Fig 8.
Preprocessed point cloud model.
Fig 9.
Point cloud semantic segmentation process.
Table 3.
Statistics of semantic segmentation results (Note: types of recognition errors in parentheses).
Table 4.
Confusion matrix for semantic segmentation results.
Fig 10.
Display of point cloud semantic segmentation results.