Author Affiliations
1College of Geodesy and Geomatics, Shandong University of Science and Technology, Qingdao , Shandong 266590, China2School of Civil Engineering, Anhui Jianzhu University, Hefei , Anhui 230601, China3Qingdao Xiushan Mobile Survey Co., Ltd., Qingdao , Shandong 266590, Chinashow less
Fig. 1. Flow chart of proposed algorithm
Fig. 2. Extracting feature points from different number of point clouds
Fig. 3. Local coordinate system
Fig. 4. Calculation principle of PFH
Fig. 5. Calculation principle of FPFH
Fig. 6. Point cloud visualization. (a) Source point cloud; (b) target point cloud; (c) overall point cloud
Fig. 7. Block effect of point cloud. (a) Chunking result of temple source point cloud; (b) chunking result of temple target point cloud
Fig. 8. Extracting feature points from temple point cloud. (a) Feature point extraction result of temple source point cloud; (b) feature point extraction result of temple target point cloud
Fig. 9. Extracting feature points from chunked temple point cloud. (a) Feature point extraction result of temple source point cloud after chunking; (b) feature point extraction result of temple target point cloud after chunking
Fig. 10. Extraction of overlapping point cloud regions
Fig. 11. Accurate alignment of different algorithms. (a) Precision alignment result of conventional ICP algorithm; (b) fine alignment result without extraction of overlapping areas; (c) fine alignment result after extraction of overlapping regions
Point cloud model | Number of point clouds | Time spent extracting feature points /s | Point cloud model | Number of point clouds | Time spent extracting feature points /s |
---|
Temple 1 | 133201 | 77.005 | Moto 1 | 104907 | 78.435 | Temple 2 | 92116 | 52.265 | Moto 2 | 59023 | 43.847 | Temple 3 | 66472 | 38.461 | Moto 3 | 45884 | 33.744 | Temple 4 | 51577 | 29.383 | Moto 4 | 37025 | 27.726 | Temple 5 | 41022 | 24.343 | Moto 5 | 21998 | 15.804 | Temple 6 | 31250 | 17.541 | Moto 6 | 14179 | 10.055 | Temple 7 | 21073 | 12.601 | Moto 7 | 5674 | 4.008 |
|
Table 1. Time taken to extract feature points with SIFT algorithm for different number of point clouds
Parameter | Overlap rate 100% | Overlap rate 80% | Overlap rate 50% | Overlap rate 30% |
---|
Point cloud alignment visualization | | | | | Registration error /m | 2.7×10-3 | 3.2×10-3 | 3.9×10-3 | Matching error |
|
Table 2. Point cloud matching accuracy with different overlap rates
Parameter | Overlapping areas extracted at 80% overlap | Overlapping areas extracted at 50% overlap | Overlapping areas extracted at 30% overlap |
---|
Point cloud alignment visualization | | | | Registration error /m | 3.0×10-3 | 2.7×10-3 | 2.5×10-3 |
|
Table 3. Matching accuracy after extracting overlapping regions from point clouds with different overlapping rates
Time taken to extract feature points /s | | Fine registration time /s | | Fine registration error /m |
---|
Point clouds are unpartitioned | Point clouds are partitioned | No overlapping area extracted | Extracted overlapping areas | No overlapping area extracted | Extracted overlapping areas | 356.166 | 44.801 | 4.523 | 3.159 | 2.569×10-5 | 1.497×10-5 |
|
Table 4. Comparison of running time and registration accuracy of each registration algorithm