Chinese Journal of Lasers, Volume. 47, Issue 12, 1204007(2020)
Inverse Depth Adaptive Weighting Based Multi-View Triangulation Method
Figures & Tables(9)
Fig. 3. Synthetic datasets for multi-view triangulation. (a) Type A; (b) type B; (c) type C
Fig. 5. Multi-view triangulation results of our method under the public datasets. (a) Lund Cathedral; (b) Aos Hus; (c) San Marco; (d) Orebro Castle; (e) Buddah Statue; (f) East Indiaman Goteborg; (g) Ystad Monestary; (h) Round Church; (i) Skansen Kronan; (j) Skansen Lejonet
Fig. 6. Number of iterations of our method for multi-view triangulation in public datasets
Table 1. Results obtained by the multi-view triangulation method based on simulation data
View tableTable 1. Results obtained by the multi-view triangulation method based on simulation data
σ /pixel Method Type A Type B Type C Time /s 3D error / m 2D error / pixel Time /s 3D error / m 2D error / pixel Time /s 3D error / m 2D error / pixel Midpoint 0.648 0.023 6.107 2.521 0.021 5.945 0.853 0.017 5.825 6 Ours 1.761 0.017 5.728 5.971 0.018 5.938 2.251 0.016 5.801 L2Rep 2.996 0.015 5.704 11.668 0.017 5.936 4.005 0.016 5.797 Midpoint 0.617 0.045 12.244 2.462 0.059 11.941 0.810 0.036 11.652 12 Ours 1.802 0.031 11.669 5.675 0.037 11.883 2.068 0.031 11.595 L2Rep 2.841 0.030 11.620 11.678 0.035 11.879 3.933 0.029 11.582 Midpoint 0.603 0.094 25.953 2.578 0.208 24.278 0.849 0.078 23.498 24 Ours 1.914 0.063 23.009 7.191 0.074 23.781 2.382 0.064 23.216 L2Rep 3.224 0.059 22.905 14.965 0.071 23.773 4.017 0.059 23.180
Table 2. Results of different methods based on Lund's public dataset
View tableTable 2. Results of different methods based on Lund's public dataset
Dataset Time /s Mean 2D reprojection errors /pixel ID View Point Midpoint Ours L2Rep Mean Variance Midpoint Ours L2Rep Midpoint Ours L2Rep 1 1208 159055 23.082 43.014 91.910 1.088 1.078 1.077 0.217 0.206 0.205 2 800 354134 22.303 45.273 93.063 0.816 0.805 0.805 0.302 0.289 0.287 3 1498 231507 33.786 66.944 134.205 0.807 0.799 0.798 0.331 0.316 0.315 4 761 53857 10.998 20.410 40.532 0.942 0.936 0.936 0.196 0.191 0.190 5 322 156356 9.440 15.233 31.873 0.651 0.649 0.649 0.270 0.266 0.265 6 179 25655 2.723 5.736 11.040 1.127 1.122 1.121 0.386 0.377 0.376 7 290 139951 14.520 26.328 53.460 0.970 0.968 0.967 0.170 0.169 0.168 8 92 84643 6.366 11.699 22.445 0.387 0.385 0.385 0.121 0.119 0.119 9 131 28371 3.812 7.254 14.793 0.807 0.802 0.802 0.175 0.170 0.169 10 368 74423 8.370 15.871 34.350 1.032 1.023 1.022 0.209 0.202 0.200
Table 3. Experimental results of different methods on the Lund public dataset (σ=5 pixel)
View tableTable 3. Experimental results of different methods on the Lund public dataset (σ=5 pixel)
ID Time /s Mean 2D reprojection error /pixel Midpoint Ours L2Rep Midpoint Ours L2Rep 1 20.508 44.275 90.556 2.136 2.105 2.103 2 23.829 46.673 96.919 2.964 2.934 2.930 3 34.913 63.022 139.873 2.663 2.627 2.624 4 11.380 20.585 46.909 1.677 1.664 1.662 5 9.691 17.316 38.984 3.050 3.046 3.044 6 2.724 5.838 11.809 2.356 2.350 2.351 7 13.903 27.213 58.175 2.257 2.241 2.236 8 6.289 12.397 26.147 2.829 2.813 2.811 9 3.859 6.827 15.900 1.917 1.871 1.870 10 7.786 15.550 34.184 2.139 2.109 2.105
Tools
Get Citation
Copy Citation Text
Fang Wei, Yang Kui. Inverse Depth Adaptive Weighting Based Multi-View Triangulation Method[J]. Chinese Journal of Lasers, 2020, 47(12): 1204007
Paper Information
Category: Measurement and metrology
Received: Jun. 18, 2020
Accepted: --
Published Online: Nov. 17, 2020
The Author Email: Wei Fang (fangwei@bupt.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20