Line Laser Measurement Error Compensation Under Vibration Conditions With Improved DeblurGAN

In the process of line laser measurement, the vibration of the sensor or the workpiece will cause the light stripe image to be blurred, resulting in unstable extraction of the laser light stripe center and reduced measurement accuracy. Aiming at the measurement error problem caused by blurring of line laser stripe image under vibration, this article studies the restoration method of blurred light stripe image based on an improved deblur generative adversarial network (DeblurGAN). The improved DeblurGAN includes adding hybrid dilated convolution (HDC), replacing the original nine residual blocks with a residual-in-residual dense block (RRDB), and adding the skip connection and
$L1$
loss. The HDC eliminates the local checkerboard phenomenon. RRDB obtains more abundant local features and makes the training more stable. The skip connection alleviates the gradient disappearance. With the improved DeblurGAN, the center of the laser stripe is extracted to obtain the true contour of the surface. The simulation shows that the proposed method has excellent performance, in which PSNR is up to 29.3486 and SSIM is up to 0.9621. The measure experiments under different vibration conditions show that the mean squared error of the fit light-stripe line was reduced by 23.89% for the flat workpiece and by 73.42% for the standard ball radius by the deblurred light stripe image, which verified the effectiveness of the proposed method.