Author Affiliations
1College of Information Technology, Hebei University of Economics and Business, Shijiazhuang 050061, Hebei , China2Faculty of Information Technology, Beijing University of Technology, Beijing 100124, Chinashow less
Fig. 1. Image matching at distance ratio threshold of 0.8
[27]. (a) 1m2; (b) 1m3; (c) 1m4; (d) 1m5
Fig. 2. Histograms of distance ratio of matching images. (a) 1m2; (b) 1m3; (c) 1m4; (d) 1m5
Fig. 3. Distribution of inliers and outliers in the matched images with zoom change
[28]. (a) Correct matched pairs; (b) distribution of inliers and outliers (red points are outliers, yellow points are inliers)
Fig. 4. Secondary removal of outliers
[29]. (a) Matched pairs obtained by existing method; (b) vectors of position change between inliers, and the included outliers
Fig. 5. Flow chart of the proposed method
Fig. 6. Schematic of conscient distribution of neighbor points
Fig. 7. Algorithm of inlier ratio promotion based on near neighbor inlier distribution consensus
Fig. 8. Histograms of inlier distance errors. (a) Boston; (b) BruggeTower; (c) ExtremeZoom; (d) Graf; (e) Effel
Fig. 9. Multi-model estimation algorithm based on distance error marginalization
Fig. 10. Vector of position change between inliers. (a) Leafs
[28]; (b) Toy and Bread
[29]; (c) Booksh
[28]; (d) ExtremeZoom
[28] Fig. 11. Schematic of image gridding and local area with its neighbor areas
Fig. 12. Secondary removal algorithm of outlier
Fig. 13. Experimental results on homogr dataset. (a) Inlier ratio of different image pairs; (b) inlier ratio after outlier filtering out; (c) recall of inliers; (d) original number of matched pairs; (e) number of matched pairs after outlier filtering out
Fig. 14. Inlier distance error obtained by different methods
Fig. 15. Comparison of average performance of different algorithms on different criteria. (a) Undetected outlier ratio; (b) number of inliers; (c) consumption time
Fig. 16. Multi-plane estimation obtained by the proposed method under Adelaidermf data set. (a) ladysymon; (b) neem; (c) nese; (d) johnsona; (e) elderhallb; (f) unihouse; (g) bonhall; (h) napiera; (i) oldclassicswing; (j) library
Parameter | Indoor | | Outdoor |
---|
Wash | Scene 0722 | Scene 0758 | Scene 0726 | Toys and Breads | | Graf | Scared Heart | Saint Peter’s Basilica | Kremlin | Potala |
---|
Size/(pixel×pixel) | 768×576 | 1296×968 | 1296×968 | 1296×968 | 480×640 | | 800×640 | 1065×693/1039×687 | 1039×688/1032×771 | 800×500/800×541 | 1023×682/1400×808 | Number of correspondences | 813 | 3197 | 2909 | 930 | 1172 | | 1233 | 1292 | 1440 | 1729 | 2280 |
|
Table 1. Information of image pairs
Image | Indicator | RANSAC | PROSAC | NAPSAC | P-NAPSAC | SC-RANSAC | Adalam | OANET | SuperGlue | Proposed method |
---|
Wash | noi | 14 | 24 | 7 | 17 | 34 | 67 | 41 | 63 | 64 | nor | 5 | 3 | 2 | 5 | 2 | 6 | 0 | 3 | 3 | t /s | 0.63 | 0.128 | 0.44 | 0.216 | 0.223 | 0.26 | 0.321 | 0.176 | 0.135 | Scene 0722 | noi | 5 | 8 | 4 | 16 | 27 | 10 | 76 | 92 | 123 | nor | 5 | 4 | 1 | 2 | 4 | 0 | 48 | 2 | 1 | t /s | 0.57 | 0.97 | 0.288 | 0.35 | 0.92 | 0.671 | 0.77 | 0.61 | 0.56 | Scene 0758 | noi | 7 | 59 | 8 | 32 | 58 | 59 | 121 | 95 | 127 | nor | 4 | 6 | 3 | 6 | 7 | 1 | 68 | 5 | 2 | t /s | 0.78 | 0.49 | 0.62 | 0.73 | 0.75 | 0.64 | 0.76 | 0.52 | 0.63 | Scene 0726 | noi | 8 | 6 | 3 | 10 | 19 | 16 | 120 | 38 | 47 | nor | 1 | 1 | 3 | 1 | 3 | 0 | 60 | 2 | 0 | t /s | 0.93 | 0.183 | 0.419 | 0.299 | 0.314 | 0.301 | 0.351 | 0.233 | 0.213 | Toys and Breads | noi | 189 | 184 | 88 | 142 | 193 | 451 | 275 | 439 | 460 | nor | 2 | 4 | 7 | 3 | 5 | 1 | 59 | 4 | 1 | t /s | 0.53 | 0.165 | 0.42 | 0.36 | 0.39 | 0.43 | 0.411 | 0.308 | 0.253 | Graf | noi | 8 | 28 | 7 | 29 | 60 | 43 | 69 | 104 | 83 | nor | 8 | 13 | 5 | 16 | 6 | 1 | 3 | 3 | 1 | t /s | 0.62 | 0.21 | 0.392 | 0.411 | 0.43 | 0.37 | 0.458 | 0.419 | 0.352 | Scared Heart | noi | 15 | 77 | 5 | 51 | 85 | 112 | 205 | 135 | 185 | nor | 2 | 5 | 5 | 8 | 13 | 3 | 55 | 4 | 3 | t /s | 0.61 | 0.211 | 0.356 | 0.412 | 0.39 | 0.383 | 0.416 | 0.294 | 0.314 | Saint Peter’s Basilica | noi | 7 | 12 | 5 | 35 | 73 | 47 | 105 | 100 | 116 | nor | 0 | 9 | 5 | 6 | 5 | 0 | 20 | 3 | 2 | t /s | 0.71 | 0.291 | 0.652 | 0.489 | 0.521 | 0.477 | 0.503 | 0.391 | 0.401 | Kremlin | noi | 8 | 14 | 9 | 28 | 40 | 15 | 57 | 46 | 45 | nor | 0 | 8 | 7 | 3 | 6 | 0 | 7 | 8 | 1 | t /s | 0.71 | 0.341 | 0.482 | 0.532 | 0.591 | 0.52 | 0.521 | 0.587 | 0.512 | Potala | noi | 7 | 8 | 6 | 12 | 11 | 6 | 76 | 51 | 56 | nor | 2 | 2 | 4 | 4 | 8 | 0 | 8 | 1 | 1 | t /s | 0.8 | 0.261 | 0.611 | 0.542 | 0.665 | 0.651 | 0.639 | 0.579 | 0.599 |
|
Table 2. Comparison results
Image | Number of planes | PEARL | Multi-X | Multi-H | CONSAC | MCT | Sequential RANSAC | Proposed |
---|
ladysymon | 2 | 8.91 | 5.31 | 4.49 | 2.95 | 3.80 | 3.80 | 1.43 | neem | 3 | 4.21 | 0.00 | 0.00 | 2.74 | 14.44 | 14.44 | 1.88 | nese | 2 | 5.33 | 0.00 | 0.00 | 0.00 | 12.83 | 0.47 | 0.83 | johnsona | 2 | 9.21 | 3.75 | 2.47 | 14.48 | 18.77 | 28.04 | 3.7 | elderhallb | 5 | 10.33 | 6.45 | 5.31 | 11.69 | 20.31 | 18.67 | 5.28 | unihouse | 5 | 9.91 | 6.39 | 7.21 | 8.84 | 10.69 | 10.69 | 2.99 | bonhall | 6 | 15.63 | 7.91 | 8.22 | 16.93 | 29.29 | 20.43 | 8.19 | napiera | 3 | 11.99 | 3.12 | 3.44 | 2.72 | 21.32 | 11.66 | 2.53 | oldclassicswing | 2 | 6.11 | 0.00 | 0.00 | 1.69 | 15.2 | 1.32 | 0.02 | library | 3 | 6.71 | 0.96 | 1.43 | 1.21 | 14.79 | 11.35 | 0.66 | Mean | | 8.834 | 3.389 | 3.257 | 6.325 | 16.144 | 12.09 | 2.72 | Average standard deviation | | 2.56 | 2.57 | 2.48 | 5.33 | 5.02 | 6.65 | 1.83 |
|
Table 3. Plane error rate of different methods