Dynamic scene deblurring and image de-raining based on generative adversarial networks and transfer learning for Internet of vehicle

Abstract Extracting traffic information from images plays an increasingly significant role in Internet of vehicle. However, due to the high-speed movement and bumps vehicle, image will be blurred during acquisition. In addition, rainy days, as a result rain attached lens, target blocked by rain, distorted. These problems have caused great obstacles for extracting key transportation images, which affect real-time judgment vehicle control system on road conditions, further cause decision-making errors even bearing accidents. this paper, we propose motion-blurred restoration removal algorithm IoV based generative adversarial network transfer learning. Dynamic scene deblurring de-raining are both among challenging classical research directions low-level vision tasks. For tasks, firstly, instead using ReLU conventional residual block, designed block containing three 256-channel convolutional layers, used Leaky-ReLU activation function. Secondly, networks task with our Resblocks, well task. Thirdly, experimental results synthetic blur dataset GOPRO real RealBlur confirm effectiveness model deblurring. Finally, learning, can fine-tune pre-trained less training data show good several datasets removal.