Optics and Precision Engineering, Volume. 32, Issue 11, 1759(2024)
Multi-level filter network for low-overlap point cloud registration
Figures & Tables(15)
Fig. 7. Registration error chromatogram in single scanning of flywheel shell
Fig. 8. Difference of evaluation index between nearest correspondence and positive correspondence
Fig. 9. Registration results of car body dataset by 3D Match with different overlap ratios
Fig. 10. Consistency voting of multi-scale features in flywheel shell
Table 1. Experimental results of different methods on indoor 3DMatch and 3DLoMatch datasets
View tableView in ArticleTable 1. Experimental results of different methods on indoor 3DMatch and 3DLoMatch datasets
Method 3DMatch FMR 3DLoMatch FMR 5000 2500 1000 500 250 5000 2500 1000 500 250 FCGF［ 20 ］97.4 97.3 97 96.7 96.6 76.6 75.4 74.2 71.7 67.3 PREDATOR［ 18 ］96.6 96.6 96.5 96.3 96.5 78.6 77.4 76.3 75.7 75.3 CoFiNet［ 31 ］98.1 98.3 98.1 98.2 98.3 83.1 83.5 83.3 83.1 82.6 GeoTransformer［ 22 ］97.9 97.9 97.9 97.9 97.6 88.3 88.6 88.8 88.6 88.3 MulFNet（Ours） 97.6 97.4 97.1 96.7 96.8 85.7 84.6 84.7 82.4 79.4 Method 3DMatch IR 3DLoMatch IR 5000 2500 1000 500 250 5000 2500 1000 500 250 FCGF［ 20 ］56.8 54.1 48.7 42.5 34.1 21.4 20 17.2 14.8 11.6 PREDATOR［ 18 ］58 58.4 57.1 54.1 49.3 26.7 28.1 28.3 27.5 25.8 CoFiNet［ 31 ］49.8 51.2 51.9 52.2 52.2 24.4 25.9 26.7 26.8 26.9 GeoTransformer［ 22 ］71.9 75.2 76 82.2 85.1 43.5 45.3 46.2 52.9 57.7 MulFNet（Ours） 66.5 63.6 70.8 67.2 65.3 38.1 38.6 35.2 36.1 30.8 Method 3DMatch RR 3DLoMatch RR 5000 2500 1000 500 250 5000 2500 1000 500 250 FCGF［ 20 ］85.1 84.7 83.3 81.6 71.4 40.1 41.7 38.2 35.4 26.8 PREDATOR［ 18 ］89 89.9 90.6 88.5 86.6 59.8 61.2 62.4 60.8 58.1 CoFiNet［ 31 ］89.3 88.9 88.4 87.4 87 67.5 66.2 64.2 63.1 61 GeoTransformer［ 22 ］92 91.8 91.8 91.4 91.2 75 74.8 74.2 74.1 73.5 MulFNet（Ours） 89.4 91.2 90.4 88.2 89.6 68.5 69.7 66.3 67.6 66.8
Table 2. Running time of different methods on indoor 3DMatch and 3DLoMatch datasets
View tableView in ArticleTable 2. Running time of different methods on indoor 3DMatch and 3DLoMatch datasets
Method 3DMatch 3DLoMatch Feature Pose Total Feature Pose Total FCGF［ 20 ］0.055 3.492 3.547 0.054 3.404 3.458 PREDATOR［ 18 ］0.034 5.376 5.410 0.033 5.207 5.239 CoFiNet［ 31 ］0.121 1.897 2.018 0.117 1.820 1.936 GeoTransformer［ 22 ］0.079 1.636 1.715 0.076 1.598 1.674 MulFNet（Ours） 0.146 1.857 2.003 0.138 1.796 1.915
Table 3. Experimental results of flywheel shell dataset （overlap ratio of 10.49%）
View tableView in ArticleTable 3. Experimental results of flywheel shell dataset （overlap ratio of 10.49%）
Method Coarse Fine Valid /% RMSE
/mm
RMSE
/mm
ER
/（°）
ET
/mm
Valid
/%
RMSE
/mm
RMSE*
/mm
ER
/（°）
ET
/mm
4PCS［ 12 ］29 21.003 26.655 2.248 6.597 36 4.868 27.808 2.972 0.617 Super4PCS［ 35 ］100 20.882 26.466 2.902 6.347 100 8.149 27.176 2.620 0.829 FGR［ 17 ］100 9.510 13.901 0.790 3.368 100 3.390 9.047 0.346 0.409 FPFH［ 15 ］21 26.064 31.703 2.978 7.538 21 7.736 23.031 2.960 0.925 ICP［ 3 ］- - - - - 100 1.749 3.816 0.169 0.147 DPWVM［ 9 ］- - - - - 100 1.601 4.690 0.312 0.134 FCGF［ 20 ］14 7.781 31.886 3.613 0.610 9 2.082 3.399 0.377 0.154 PREDATOR［ 18 ］99 19.283 23.571 2.424 5.737 100 6.041 25.949 2.421 0.820 CoFiNet［ 31 ］100 9.202 11.804 1.035 2.846 100 1.501 7.704 0.789 0.189 GeoTransformer［ 22 ］100 7.587 31.794 3.163 1.130 100 7.071 28.133 2.775 1.009 MulFNet（Ours） 78 1.385 5.419 1.097 0.248 65 1.032 1.649 0.151 0.078
Table 4. Experimental results of car body dataset （overlap ratio of 14.56%）
View tableView in ArticleTable 4. Experimental results of car body dataset （overlap ratio of 14.56%）
Method Coarse Fine Valid
/%
RMSE
/mm
RMSE*
/mm
ER
/（°）
ET
/mm
Valid
/%
RMSE
/mm
RMSE*
/mm
ER
/（°）
ET
/mm
4PCS［ 12 ］20 25.719 27.682 3.573 6.684 20 7.689 33.365 4.863 0.843 Super4PCS［ 35 ］100 30.458 32.661 2.828 6.597 98 10.555 25.859 2.899 0.931 FGR［ 17 ］100 13.942 18.226 2.034 4.583 100 7.687 13.409 1.233 0.688 FPFH［ 15 ］58 24.045 27.767 3.427 6.859 53 8.200 16.515 1.982 0.959 ICP［ 3 ］- - - - - 90 1.837 3.352 0.143 0.123 DPWVM［ 9 ］- - - - - 100 1.043 2.605 0.157 0.137 FCGF［ 20 ］78 6.322 24.845 3.296 1.009 80 1.183 1.703 0.169 0.093 PREDATOR［ 18 ］98 8.263 10.051 1.002 2.526 100 4.300 9.788 0.748 0.215 CoFiNet［ 31 ］100 6.894 9.194 0.900 2.223 100 3.333 7.433 0.581 0.161 GeoTransformer［ 22 ］100 7.672 33.404 2.651 1.083 100 6.811 28.089 3.103 1.059 MulFNet（Ours） 67 3.171 12.464 1.121 0.343 82 0.958 1.379 0.274 0.151
Table 5. Ablation experiment result of MulFNet architecture
View tableView in ArticleTable 5. Ablation experiment result of MulFNet architecture
Multi-level Info-Interaction Filter Valid/% RMSE/mm RMSE*/mm ER/mm ET/（°） × √ √ 38 1.706 2 5.946 2 0.354 9 0.237 2 √ × √ 47 2.336 8 7.286 4 0.601 1 0.247 9 √ √ × 83 2.521 4 11.113 4 1.107 0 0.371 2 √ √ √ 65 1.031 5 1.649 0 0.150 6 0.077 8
Tools
Get Citation
Copy Citation Text
Minqi HE, Li LIU, Shang LI, Hao WU, Dahu ZHU. Multi-level filter network for low-overlap point cloud registration[J]. Optics and Precision Engineering, 2024, 32(11): 1759
Paper Information
Category:
Received: Nov. 6, 2023
Accepted: --
Published Online: Aug. 8, 2024
The Author Email: Hao WU (wuhao2023@whut.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20