首页|Feature points extraction of defocused images using deep learning for camera calibration
Feature points extraction of defocused images using deep learning for camera calibration
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NSTL
Elsevier
Camera calibration is difficult when the focus plane is in a dangerous place or where people are not easy to reach. Therefore, this paper proposes a calibration method which can be used in the out-of-focus area of the camera. Firstly, a circular calibration pattern based on phase coding is made. Next, a deep learning-based phase recovery network (Phase-Net) is built, and then the recovered phase diagram is corrected for ellipse eccentricity to obtain the feature points needed for camera calibration. The proposed method only needs one shot to recover the phase, which overcomes the problem that the defocusing calibration method based on the phase shift principle needs multiple patterns to recover the phase. Simulation and experiments demonstrate that the maximum mean reprojection error is 0.11 pixels, and the relative error between the calibration results of this method and phase shifting method is 1.54%. The obtained results validate the effectiveness of Phase-Net in engineering applications.
DefocusCamera calibrationFeature point extractionDeep learningRecovery networkPhase shiftingFRINGE PROJECTION PROFILOMETRYPARAMETERS
Huo, Junzhou、Meng, Zhichao、Zhang, Haidong、Chen, Shangqi、Yang, Fan
NSTL
Elsevier
