Traffic sign defogging based on conditional adversarial neural network pix2pixHD

A foggy environment has a great impact on the recognition of traffic signs, and it is very difficult to recognize traffic signs in a foggy environment directly. In this paper, we debug and train the conditional generative adversarial network pix2pixHD to generate a fog removal model consistent with an advanced vision task (traffic sign recognition). First, a foggy traffic sign dataset is constructed based on the atmospheric scattering model. Secondly, the idea of the conditional adversarial neural network is introduced to complete the construction and training of the pix2pixHD model. Finally, two image quality evaluation indexes, Peak Signal to Noise Ratio (PSNR) and Structural similarity (SSIM), and the Yolov5 detection algorithm are used to complete the pix2pixHD defogging model to compare with the common defogging models respectively. The experiments prove that the model trained in this paper can effectively improve the quality of foggy images and generate high-resolution images that are more compatible with the recognition of traffic signs.