Chinese Journal of Lasers, Volume. 49, Issue 18, 1810002(2022)
Road Scene Laser Point Cloud Registration Method Based on Geographical Object Features
Figures & Tables(16)
Fig. 3. Schematics of 4PCS registration. (a) Correct registration; (b) meet LCP constraint; (c) 4PCS of key points constraint
Fig. 4. Experimental data. (a) Instrument and equipment; (b) profile measurement area (A and B areas); (c) MLS point cloud; (d) TLS point cloud; (e) relative position of different data
Fig. 7. Key points distribution. (a) Key points distribution extracted with 3D-SIFT algorithm; (b) key points distribution extracted with normal vector angle algorithm combining with LSP; (c) key points distribution extracted with ISS algorithm; (d) key points distribution extracted with our algorithm
Fig. 8. MLS point cloud registration results for region A. (a) Original point clouds; (b) registration results
Fig. 9. Curbstone distribution in segmented point cloud. (a) 50 m segmented point cloud; (b) 100 m segmented point cloud
Fig. 10. TLS and MLS point cloud registration results. (a) Original point cloud; (b) point cloud after registration; (c) local amplification of registration point cloud
Fig. 11. Registration results of different algorithms. (a) 4PCS algorithm; (b) Key-4PCS algorithm; (c) ICP algorithm; (d) Key-KD-ICP algorithm; (e) Key4PCS-KeyICP algorithm
Table 1. Basic information of experimental data
View tableTable 1. Basic information of experimental data
Experimental area Acquisition system Road environment Length of road /km Positional deviation /m Quantity of scanning points /106 A MLS Residential area 1.6 6 160 B FARO Major area 0.15 43 18
Table 2. Quantity of key points extracted with each algorithm and algorithm efficiency
View tableTable 2. Quantity of key points extracted with each algorithm and algorithm efficiency
Key point extraction algorithm Time /s Quantity of key points 3D-SIFT 12.87 850 Normal vector angle algorithm combining with LSP 10.18 1164 ISS 3.11 1018 Ours 6.22 708
Table 3. Experimental results of segmented point cloud with different sections
View tableTable 3. Experimental results of segmented point cloud with different sections
Fractional length /m Time of key points extraction /s Time of registration /s RMSE /m 30 10.62 2.42 0.022 40 8.47 3.20 0.026 50 10.30 5.04 0.038 60 12.40 2.62 0.036 80 12.69 3.97 0.042 100 28.37 3.85 0.043
Table 4. Evaluation of TLS and MLS point cloud registration results
View tableTable 4. Evaluation of TLS and MLS point cloud registration results
Station Time of key point extraction /s Time of registration /s RMSE /m 1 3.73 2.99 0.046 2 4.44 17.84 0.042 3 4.80 5.04 0.042 Overall 12.97 25.87 0.044
Table 5. Comparison on time consuming and precision of different algorithms
View tableTable 5. Comparison on time consuming and precision of different algorithms
Algorithm Time /s RMSE /m 4PCS 12.59 0.380 Key-4PCS 2.06 0.172 ICP 7.81 0.230 Key-KD-ICP 0.82 0.049 Key4PCS-KeyICP 2.30 0.038
Tools
Get Citation
Copy Citation Text
Rufei Liu, Fei Wang, Hongwei Ren, Minye Wang, Jiben Yang. Road Scene Laser Point Cloud Registration Method Based on Geographical Object Features[J]. Chinese Journal of Lasers, 2022, 49(18): 1810002
Paper Information
Category: remote sensing and sensor
Received: Dec. 10, 2021
Accepted: Jan. 17, 2022
Published Online: Jul. 28, 2022
The Author Email: Fei Wang (WangFei202123@163.com)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20