Dense Feature Matching Based on Improved DFM Algorithm

Table 1. Analysis of matching accuracy in outdoor large scenes
View table
Table 1. Analysis of matching accuracy in outdoor large scenes
Algorithm Rotation error /（°） Translation error /m Inlier matches
DFM 20.57 13.24 99
DFM 18.58 9.27 71
DFM 1.78 2.80 9
Improved DFM 0.65 2.70 2566
Improved DFM 1.63 0.99 1709
Improved DFM 0.52 2.23 487

Table 2. Feature matching accuracy evaluation

View table

Table 2. Feature matching accuracy evaluation

Method	$f_{M M A}$			Number of feature matches /10³
Method	3 pixel	5 pixel	10 pixel	3 pixel	5 pixel	10 pixel
SIFT	67.8	69.4	70.3	0.40	0.41	0.41
SuperGlue	71.3	79.9	88.6	0.42	0.46	0.49
NCNet	60.4	69.3	86.5	0.54	0.60	0.86
LoFTR	89.1	90.6	92.4	2.30	2.30	2.40
Patch2Pix	82.4	87.6	91.0	0.62	0.64	0.66
DFM	89.4	92.5	93.9	3.10	3.20	3.30
Improved DFM（r=0.9）	89.7	91.1	91.6	18.20	18.60	18.60
Improved DFM（r=0.5）	97.4	97.6	97.6	3.10	3.10	3.10

Table 3. Results of homography estimation evaluation

View table

Table 3. Results of homography estimation evaluation

Method	HEA
Method	3 pixel	5 pixel	10 pixel
SIFT	77.2	83.8	85.3
SuperGlue	76.7	82.3	87.0
NCNet	51.0	63.3	76.4
LoFTR	76.7	82.6	85.9
Patch2Pix	76.7	81.0	84.5
DFM	74.5	80.2	85.2
Improved DFM（r=0.9）	79.6	85.5	88.3
Improved DFM（r=0.5）	66.9	74.3	81.6

Tools

Get Citation

Copy Citation Text

Yanhan Zhang, Yinxin Zhang, Zhanhua Huang, Kangnian Wang. Dense Feature Matching Based on Improved DFM Algorithm[J]. Laser & Optoelectronics Progress, 2024, 61(8): 0815001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites