Using Motion Deblurring Algorithm to Improve Vehicle Recognition via DeblurGAN
Zhiyong Liang, Boxiong Yang, Heng Xiao · 2020
In order to overcome the effect of motion blur on automatic vehicle detection, recognition and tracking in traditional vehicle identification field, we propose an algorithm via Generative Adversarial Networks (GAN) for de-blurring motion blurred vehicle images. The algorithm uses Generator and Discriminator in GAN to game each other, making the resulting image closer to the target image. Based on the above, the two networks use the VGG19 network in the process of playing with each other, which makes the image de-motion blurring better, thus effectively improving the vehicle recognition rate. The algorithm is therefore significant for target detection systems in computer vision systems. From the final experimental evaluation, the images generated via this algorithm have greatly improved the recognition appearance, PSNR and SSIM, and the recognition rate of the vehicles can be increased by up to 34%.