Author Affiliations
1 School of Mechatronic Engineering, Nanchang University, Nanchang, Jiangxi 330031, China2 School of Architectural and Mechanical Engineering, Chifeng University, Chifeng, Inner Mongolia 0 24000, Chinashow less
Fig. 1. Overall procedure of the proposed method. (a) Original point cloud; (b) region clustering segmentation result; (c) feature point recognition result; (d) generation of the feature lines
Fig. 2. Flow chart of the feature point recognition stage
Fig. 3. Feature line extraction from Model 1. (a) Original point cloud of Model 1; (b) region clustering segmentation; (c) set of candidate feature points; (d) set of feature points; (e) feature line extraction results
Fig. 4. Feature line extraction from Model 2. (a) Triangulated lighting model of Model 2; (b) region clustering segmentation; (c) set of candidate feature points; (d) set of feature points; (e) feature line extraction results
Fig. 5. Feature line extraction from Model 3. (a) Triangulated lighting model of Model 3; (b) region clustering segmentation; (c) set of candidate feature points; (d) set of feature points; (e) feature line extraction results
Fig. 6. Feature line extraction from Model 4. (a) Triangulated lighting model of Model 4; (b) region clustering segmentation; (c) set of candidate feature points; (d) set of feature points; (e) feature line extraction results
Fig. 7. Comparison of the feature extraction methods used in Model 4. (a) Ref. [13] method; (b) proposed method
Fig. 8. Comparison of the feature extraction methods used in Model 5. (a) Triangulated lighting model of Model 5; (b) Ref. [13] method; (c) proposed method
Fig. 9. Feature line extracted by the proposed method using different neighborhood scales. (a1) Original point cloud and its partial enlarged detail in Model 3; extraction effects with (a2) k=10, (a3)k=16, (a4)k=25; (a5) feature line of Model 3; (b1) original point cloud and its partial enlarged detail in Model 4; extraction effects with (b2) k=10, (b3) k=16, (b4) k=25; (b5) feature line of Model 4
Fig. 10. Results of the proposed method for noisy datasets. (a)-(c) Add noise of 15 dB, 20 dB and 30 dB in Model 2, respectively; (d)-(f) add noise of 45 dB, 50 dB and 60 dB in Model 4, respectively
Model | Number of original point clouds | Number of candidate feature points | Number of feature points |
---|
Model 1 | 6177 | 1729 | 803 | Model 2 | 13294 | 2924 | 1728 | Model 3 | 40000 | 8799 | 5199 | Model 4 | 54465 | 13071 | 7080 | Model 5 | 25901 | 5957 | 3626 |
|
Table 1. Complexity in different stages
Model | Candidate feature point extraction time /s | Feature point extraction time /s | Feature line generation time /s |
---|
Model 1 | 0.040 | 0.039 | 0.241 | Model 2 | 0.090 | 0.099 | 0.379 | Model 3 | 0.346 | 0.286 | 1.763 | Model 4 | 0.501 | 0.412 | 1.837 | Model 5 | 0.190 | 0.164 | 0.796 |
|
Table 2. Duration of the feature extraction pipeline in seconds