Fig. 1. Point cloud classification process diagram. (a) Original point clouds; (b) ground filtering; (c) segmentation of non-ground points; (d) final classification result

Fig. 2. CSF filtering effect. (a) Original point clouds 1; (b) filtering result 1; (c) original point clouds 2; (d) filtering result 2

Fig. 3. Sketch maps of calculating normal vector difference. (a) Cloud point normal vector 1; (b) cloud point normal vector 2; (c) difference of cloud point normal vector 1 and cloud point normal vector 2

Fig. 4. Results of precise extraction of different ground point clouds. (a) Differences of normal vector of ground point clouds after CSF; (b) ground point clouds after filtering the difference of normal vector

Fig. 5. Over-segmentation processing. (a) Before processing; (b) after processing

Fig. 6. Histogram of vertical slice sampling

Fig. 7. Sketch maps of features. (a) Homocentric sphere; (b) centroid distance histogram

Fig. 8. Ground filtering and segmentation results with our algorithm and algorithm in Ref.[15]. (a) Overall result of ground filtering with our algorithm; (b) overall result of ground filtering with algorithm in Ref.[15]; (c) details of ground filtering with our algorithm; (d) details of ground filtering with algorithm in Ref.[15]; (e) point cloud segmentation with our algorithm; (f) point cloud segmentation with algorithm in Ref.[15]

Fig. 9. Classification results of the first fusion strategy and the third fusion strategy. (a) Original point clouds; (b) segmentation result of non-ground point cloud; (c) classification result of non-ground point clouds based on the first fusion strategy; (d) classification result of non-ground point clouds based on the third fusion strategy; (e) final classification result based on the first fusion strategy; (f) final classification result based on the third fusion strategy