Research On The Restoration Algorithms Of Traffic Blur Images Via Deep Learning

With the increase of vehicles and the complexity of the road environment,intelligent transportation systems have played an important role in improving driving efficiency and alleviating traffic congestion.As a key part of the intelligent transportation systems,the intelligent monitoring system is of great significance for supervising the regulated driving of vehicles.Due to weather changes and vehicle movement,the images captured by the intelligent monitoring system are blurred.Traffic blurred image restoration has become an urgent problem to be solved in the forensics of traffic violations.Traditional methods cannot meet the problem of complex traffic image restoration.This paper conducts research on traffic blur image restoration algorithms based on deep learning.The main contents are as follows:According to the current status of the field of blurred image restoration,this paper analyzes the significance and purpose of the research,the imaging principles of blurred images and related theoretical knowledge are introduced.The classic algorithm models of deep learning and the development process are discussed in detail,deep learning is used as the method of traffic blur image restoration in this paper.In order to improve the practicability and stability of existing blurred image restoration algorithm,a traffic blurred image restoration algorithm based on generative adversarial network is proposed in this paper.The structure is based on the method of image translation,sets up two generative adversarial networks models to realize the conversion from blur domain to clear domain respectively.Using Res Ne Xt structure to improve the training accuracy of the model and choose adversarial loss and perceptual loss to ensure the consistency of image content.The results on the Go Pro dataset and self-made traffic blur dataset are verified that the proposed algorithm model has good migration and generalization ability.Aiming at the defects of traditional methods in the restoration of blurred image edge information and combine the idea of coarse to fine in multi-scale network.Traffic blurred restoration algorithm based on multi-scale network fused with HS-Res Net is proposed.Take encoder-decoder as generator structure and apply multi-scale strategies.Considering the edge information of the image,L2 regularization is used to constrain the image gradient.Take the HS-Res Net block into each scale to integrate receptive field better.In order to reduce training complexity and improve training stability,this model shares weights at different scales of the image.The algorithm is evaluated on self-made traffic blurred dataset,the average PSNR index is 30.060 and the average SSIM index is 0.940.In the subjective vision,the result also proves its effectiveness and practicality.Considering the practical application and analyzing the actual needs of the traffic management system,a traffic blurred image restoration system is designed.The trained multi-scale network traffic blurred image restoration algorithm model integrated with HS-Res Net is embedded in the system.The user interface is designed.The system is easy to operate and use.