Camera calibration method based on double neural network

Wenyi Chen; Jie Xu; Hui Yang

doi:10.3788/IRLA20210071

Journals >Infrared and Laser Engineering >Volume 50 >Issue 11 >Page 20210071 > Article

Infrared and Laser Engineering
Vol. 50, Issue 11, 20210071 (2021)

Camera calibration method based on double neural network

Wenyi Chen^1、2, Jie Xu^1、*, and Hui Yang¹

Author Affiliations

¹Industry School of Modern Post, Xi’an University of Posts and Telecommunications, Xi’an 710061, China

²Collaborative Innovation Center for Modern Post, Xi’an University of Posts and Telecommunications, Xi’an 710121, China

show less

DOI: 10.3788/IRLA20210071 Cite this Article

Wenyi Chen, Jie Xu, Hui Yang. Camera calibration method based on double neural network[J]. Infrared and Laser Engineering, 2021, 50(11): 20210071 Copy Citation Text

show less

Fig. 1. Relationship between coordinate systems

Download full size | View in the Article

Fig. 2. Camera calibration based on double neural networks

Download full size | View in the Article

Fig. 3. Schematic diagram of BP network model

Download full size | View in the Article

Fig. 4. Flow chart of PSO-BP algorithm

Download full size | View in the Article

Fig. 5. Schematic diagram of optical axis correction

Download full size | View in the Article

Fig. 6. Schematic diagram of calibration plate correction

Download full size | View in the Article

Fig. 7. Experimental platform

Download full size | View in the Article

Fig. 8. Training chart of PSO-BP double neural network algorithm

Download full size | View in the Article

Fig. 9. Training curve of traditional BP neural network

Download full size | View in the Article

Fig. 10. Result of Z-axis output error

Download full size | View in the Article

Fig. 11. Image of 3D reconstruction

Download full size | View in the Article

Fig. 12. Distorted image

Download full size | View in the Article

Number of hidden layer nodes	$error_{xz} /{\rm mm}$	$error_{ {yz} } /{\rm mm}$
6	0.130	0.0987
8	0.164	0.148
10	0.0822	0.0683
12	0.0751	0.0797
14	0.0647	0.0622
16	0.000210	0.000274
18	0.000242	0.000381
20	0.000195	0.00159
22	0.000569	0.000889

Table 1. Influence of hidden layer node number on experimental results

View in the Article

Expert output			Proposed method			Method in Ref. [13]
${X_{\rm{w} } }/{\rm mm}$	${Y_{\rm{w} } }/{\rm mm}$	${Z_{\rm{w} } }/{\rm mm}$	${X_{\rm{w} } }/{\rm mm}$	${Y_{\rm{w} } }/{\rm mm}$	${Z_{\rm{w} } }/{\rm mm}$	${X_{\rm{w} } }/{\rm mm}$	${Y_{\rm{w} } }/{\rm mm}$	${Z_{\rm{w} } }/{\rm mm}$
30	60	3.7650	30.0132	59.9826	3.5177	29.7803	59.7636	3.5139
30	120	18.7900	30.0824	120.103	18.8464	30.1413	119.846	18.4452
60	30	108.9200	59.9819	29.8937	108.9940	59.6575	29.9067	108.3465
90	90	108.9200	89.9033	89.9493	108.8836	90.0887	90.2409	107.9858
120	240	108.9200	119.8934	240.051	108.8500	119.7397	240.082	108.9832
150	30	108.9200	150.1060	29.8590	108.9702	149.7396	29.9120	109.1905
180	60	105.1450	179.8512	59.9526	105.0527	179.7629	59.9960	104.9458
210	180	105.1450	210.1164	179.833	105.0298	210.2637	179.878	105.3433
240	210	105.1450	240.0994	210.101	105.0015	240.2051	210.132	104.9842
330	240	108.9200	330.2025	240.035	108.9867	330.0932	240.1141	108.6544
$E$			0.1786			0.4378

Table 2. Error of partial calibration point

View in the Article

	Proposed method		Method in Ref. [13]		BP method
${\rm Error/mm}$	$X,Z$	$Y,Z$	$X,Y$	$Z$	$X,Y,Z$
$best$	0.000343	0.000414	0.000370	0.000835	0.031687
$avg$	0.3193		0.5016		1.2126

Table 3. Calibration error of different methods under the condition of high lens distortion

Wenyi Chen, Jie Xu, Hui Yang. Camera calibration method based on double neural network[J]. Infrared and Laser Engineering, 2021, 50(11): 20210071

Download Citation

Tools

Save the article for my favorites

Paper Information