Author Affiliations
1School of Automation, Beijing University of Posts and Telecommunications, Beijing 100876, China2School of Instrumentation Science and Opto-electronics Engineering, Beihang University, Beijing 100191, Chinashow less
Fig. 1. Geometric indication of multi-view triangulation.
Fig. 2. Four types of synthetic triangulation instances. (a) Type A: cameras and 3D points are randomly distributed; (b) Type B: the camera moves along a curved trajectory around 3D points; (c) Type C: the camera moves on a circle while 3D points are located at the center; (d) Type D: the camera moves along a curved trajectory towards the 3D scene.
Fig. 3. Time and accuracy analysis with different noise levels on Type D synthetic data. (a), (b), and (c) are the results of the time, 3D error, and 2D error when the Gaussian covariance is set as 2 pixels; (d), (e), and (f) are the results of the time, 3D error, and 2D error when the Gaussian covariance is set as 5 pixels; (g), (h), and (i) are the results of the time, 3D error, and 2D error when the Gaussian covariance is set as 8 pixels.
Fig. 4. Overall convergence of INT with different datasets. (a) Convergence curve of Type A dataset; (b) convergence curve of Type B dataset; (c) convergence curve of Type C dataset; (d) convergence curve of Type D dataset.
Fig. 5. INT based on real datasets. (a) Lund Cathedral, (b) Orebro Castle, (c) Ystad Monestary, and (d) Skansen Kronan.
Method | Type A | Type B | Type C |
---|
Time (s) | 3D Error (m) | 2D Error (pixels) | Time (s) | 3D Error (m) | 2D Error (pixels) | Time (s) | 3D Error (m) | 2D Error (pixels) |
---|
MVMP | 1.06 | 0.0192 | 5.1115 | 2.45 | 0.0352 | 5.0493 | 2.14 | 0.0292 | 4.9511 | IRMP | 4.01 | 0.0139 | 4.7698 | 11.76 | 0.0303 | 4.9635 | 8.41 | 0.0251 | 4.8888 | INT | 0.68 | 0.0146 | 4.7955 | 0.91 | 0.0312 | 4.9649 | 1.08 | 0.0258 | 4.8901 | ININT | 0.98 | 0.0145 | 4.7936 | 1.45 | 0.0307 | 4.9537 | 1.62 | 0.0252 | 4.8895 | NN | 6.51 | 0.0139 | 4.7698 | 22.60 | 0.0303 | 4.9635 | 14.07 | 0.0251 | 4.8888 | GMRE | 7.08 | 0.0124 | 4.7059 | 22.10 | 0.0272 | 4.9621 | 15.61 | 0.0237 | 4.8680 |
|
Table 1. Comparisons between Different Incremental Triangulation Performances
Data | Total Runtime (s) | Last Mean 3D Error |
---|
MVMP | IRMP | INT | ININT | NN | GMRE | MVMP | IRMP | INT | ININT | NN | GMRE |
---|
E | 113.40 | 941.01 | 125.16 | 204.96 | 1477.41 | 1386.46 | 0.0148 | 0.0131 | 0.0139 | 0.0134 | 0.0131 | 0.0124 | F | 110.15 | 749.16 | 63.84 | 110.93 | 1187.04 | 1195.77 | 0.0020 | 0.0010 | 0.0014 | 0.0011 | 0.0010 | 0.0002 | G | 97.99 | 469.11 | 83.55 | 138.63 | 797.93 | 665.38 | 0.0077 | 0.0072 | 0.0075 | 0.0072 | 0.0072 | 0.0067 | H | 23.29 | 206.40 | 21.24 | 35.97 | 295.81 | 246.67 | 0.0018 | 0.0009 | 0.0012 | 0.0010 | 0.0009 | 0.0004 |
|
Table 2. Runtime Comparisons of Different Methods with Real Datasetsa