High-Precision Binocular Camera Calibration Based on Super-Resolution Corner Point Detection Algorithm

Ting Sun; Na Chen; Ran Meng; Qian Long; Qiwei Xie

doi:10.3788/LOP223305

Journals >Laser & Optoelectronics Progress >Volume 60 >Issue 8 >Page 0811029 > Article

Laser & Optoelectronics Progress
Vol. 60, Issue 8, 0811029 (2023)

High-Precision Binocular Camera Calibration Based on Super-Resolution Corner Point Detection Algorithm

Ting Sun^1,2, Na Chen¹, Ran Meng², Qian Long^3,*, and Qiwei Xie⁴

Author Affiliations

¹School of Mathematics and Statistics, Hubei University, Wuhan 430062, Hubei, China

²Beijing Smarter Eye Technology Co., Ltd., Beijing 100023, China

³Yunnan Observatories, Chinese Academy of Sciences, Kunming 650000, Yunnan, China

⁴Research Base of Beijing Modern Manufacturing Development, Beijing University of Technology, Beijing 100124, China

show less

DOI: 10.3788/LOP223305 Cite this Article Set citation alerts

Ting Sun, Na Chen, Ran Meng, Qian Long, Qiwei Xie. High-Precision Binocular Camera Calibration Based on Super-Resolution Corner Point Detection Algorithm[J]. Laser & Optoelectronics Progress, 2023, 60(8): 0811029 Copy Citation Text

show less

Fig. 1. Principles of binocular stereo vision imaging

Download full size

Fig. 2. MASP algorithm framework structure diagram

Download full size

Fig. 3. MANet overall framework

Download full size

Fig. 4. MAConv framework

Download full size

Fig. 5. Reconstruct image frame

Download full size

Fig. 6. Corner detection frame

Download full size

Fig. 7. Visualization of sub-pixel corner detection at different resolutions. (a) 2× resolution; (b) 3× resolution; (c) 4× resolution

Download full size

Fig. 8. Low-resolution checkerboard calibration board left and right camera images. (a)(c) Left images; (b)(d) right images

Download full size

Fig. 9. Calibration plate diagram at a distance of 3-14 m

Download full size

Scale factor	$R_{d e}$					$R_{d}$
Scale factor	$E_{e r r o r}$ =0-0.2 pixel	$E_{e r r o r}$ =0.2-0.4 pixel	$E_{e r r o r}$ =0.4-0.6 pixel	$E_{e r r o r}$ =0.6-0.8 pixel	$E_{e r r o r}$ =0.8-0.9 pixel	$R_{d}$
$\times$ 2	0.4982	0.4175	0.0788	0.0053	0.0000	0.9738
$\times$ 3	0.4937	0.4165	0.0789	0.0089	0.0017	0.9720
$\times$ 4	0.4046	0.4334	0.1312	0.0251	0.0053	0.9703

Table 1. Comparison of accuracy ratio of corner detection under different resolution

Algorithm	$R_{d e}$
Algorithm	$E_{e r r o r}$ =0-0.2 pixel	$E_{e r r o r}$ =0.2-0.4 pixel	$E_{e r r o r}$ =0.4-0.6 pixel	$E_{e r r o r}$ =0.6-0.8 pixel	$E_{e r r o r}$ =0.8-0.9 pixel	$E_{e r r o r}$ =0.9-2 pixel	$R_{d}$	$E_{m d}$ /pixel
SuperPoint	0.0960	0.3352	0.3333	0.1920	0.0433	0.0000	0.9267	0.34
Harris	0.0059	0.0633	0.1920	0.3485	0.1485	0.2415	0.8813	0.64
MASP	0.4982	0.4175	0.0788	0.0053	0.0000	0.0000	0.9738	0.21

Table 2. Comparison of different corner detection algorithms

Binocular camera parameter	Left camera parameter	Right camera parameter
$R_{x}$ =0.0011	$k_{1}$ =-0.51	$k_{1}$ =-0.52
$R_{y}$ =-0.0006	$k_{2}$ =-0.68	$k_{2}$ =-0.22
$R_{z}$ =0.0149	$f_{x}$ =967.70	$f_{x}$ =964.79
$T_{x}$ =-120.5738	$f_{y}$ =965.78	$f_{y}$ =962.77
$T_{y}$ =-0.9858	$C_{x}$ =318.27	$C_{x}$ =317.93
$T_{z}$ =-8.5164	$C_{y}$ =178.92	$C_{y}$ =180.30

Table 3. Calibration parameters obtained by Harris corner point detection

Binocular camera parameter	Left camera parameter	Right camera parameter
$R_{x}$ =0.0010	$k_{1}$ =-0.55	$k_{1}$ =-0.51
$R_{y}$ =0.0002	$k_{2}$ =0.42	$k_{2}$ =-0.40
$R_{z}$ =0.0149	$f_{x}$ =1932.87	$f_{x}$ =1928.53
$T_{x}$ =-120.4570	$f_{y}$ =1929.41	$f_{y}$ =1925.31
$T_{y}$ =-1.1657	$C_{x}$ =637.05	$C_{x}$ =637.41
$T_{z}$ =-4.4702	$C_{y}$ =356.87	$C_{y}$ =359.43

Table 4. Calibration parameters obtained by 2× resolution MASP corner point detection

Binocular camera parameter	Left camera parameter	Right camera parameter
$R_{x}$ =0.0012	$k_{1}$ =-0.56	$k_{1}$ =-0.52
$R_{y}$ =0.0003	$k_{2}$ =0.43	$k_{2}$ =-0.21
$R_{z}$ =0.01486	$f_{x}$ =1934.35	$f_{x}$ =1932.74
$T_{x}$ =-120.6524	$f_{y}$ =1932.60	$f_{y}$ =1925.31
$T_{y}$ =-1.0536	$C_{x}$ =634.11	$C_{x}$ =631.59
$T_{z}$ =-4.3653	$C_{y}$ =358.47	$C_{y}$ =361.42

Table 5. Calibration parameters obtained by 3× resolution MASP corner point detection

Binocular camera parameter	Left camera parameter	Right camera parameter
$R_{x}$ =0.0014	$k_{1}$ =-0.54	$k_{1}$ =-0.51
$R_{y}$ =0.0004	$k_{2}$ =0.00	$k_{2}$ =-0.53
$R_{z}$ =0.01485	$f_{x}$ =1932.99	$f_{x}$ =1922.54
$T_{x}$ =-120.4545	$f_{y}$ =1930.37	$f_{y}$ =1919.69
$T_{y}$ =-1.3821	$C_{x}$ =636.61	$C_{x}$ =633.80
$T_{z}$ =-12.1748	$C_{y}$ =358.22	$C_{y}$ =361.84

Table 6. Calibration parameters obtained by 4× resolution MASP corner point detection

Method	$R_{e r r o r}$
Method	$M c_{M}$ $M r_{H}$	$M c_{M}$ $M r_{M}$	$M c_{H}$ $M r_{H}$	$M c_{H}$ $M r_{M}$
Average error rate	0.0340	0.0193	0.0308	0.0297
D3	0.0076	0.0073	0.0113	0.0112
D4	0.0083	0.0074	0.0100	0.0100
D5	0.0114	0.0099	0.0149	0.0154
D6	0.0190	0.0139	0.0217	0.0211
D7	0.0169	0.0149	0.0248	0.0244
D8	0.0266	0.0153	0.0377	0.0341
D9	0.0273	0.0297	0.0339	0.0312
D10	0.0341	0.0199	0.0390	0.0360
D11	0.0359	0.0146	0.0531	0.0557
D12	0.0384	0.0293	0.0544	0.0396
D13	0.0583	0.0299	0.0384	0.0367
D14	0.1250	0.0399		0.0413