Research On Image Clarity Correction Method For Intelligent Transportation

In recent years,people pay more attention to blur image clarity correction,and it is also a very challenging topic.The ultimate goal is to provide high-quality clear images for outdoor monitoring,aerospace,medical imaging and other fields.Intelligent transportation promotes the development of the city and provides a strong guarantee for smooth public travel and sustainable economic development.However,in the process of image acquisition and transmission,noise interference is inevitable,such as vehicle speed is too fast,imaging distance is too far and equipment resolution is too low.Therefore,it is particularly important to carry out road traffic blur image sharpening correction.The research on this subject has great significance for road traffic supervision departments.Firstly,the imaging mechanism of the blurred image and the development of deblurring are introduced in this thesis.Then,it analyzes the disadvantages of traditional deblurring methods and the shortcomings of existing deep learning methods,and systematically sorted out the related theoretical foundations of deep learning.Finally,based on the ideas and concepts of generative adversarial,with the help of deep learning methods.Two blurred image clarification methods,multi-scale encoder-decoder network,and multi-scale feature pyramid network are respectively proposed.Blind restoration of blurred images is carried out in an end-to-end manner.The main contents of this paper are as follows:1、A multi-scale encoder-decoder deep convolutional neural network structure is proposed,which adopts the network design from coarse to fine.Firstly,taking the cyclic multi-scale encoder-decoder network as the generator.An(Optimized Multi-scale Residual Block,OMSRB)structure is proposed based on the common Residual block and Multi-scale Residual block in the encoder-decoder,which reduces the number of parameters and improves the nonlinear expression ability and stability of the network.In addition,the skip connections were added between the encoder-decoder network to provide more detailed information for the restoration results.Finally,the L2 loss of each layer network is calculated to ensure that the image after deblurring is closer to reality.2、A multi-scale pyramid blind deblurring network structure is proposed.This method extracts and fuses road traffic images at multiple scales and features based on the principle of feature pyramid in the generator part,ensuring that the predicted results of each layer fully incorporate high-resolution and strong semantic features,which is more conducive to image clarity and restoration.On the basis of local and global discrimination,the discriminator innovatively uses the downsampling multi-scale discrimination,and comprehensively judges the images with different resolutions and random patches,so as to more accurately guide the network to carry out reverse parameter adjustment optimization.Finally,a multi-scale structural similarity loss function is introduced to further constrain the generation of high-quality images.The two methods proposed in this thesis are tested on Go Pro dataset and collected road traffic dataset,and compared with the existing methods.The simulation results show that compared with the classical network model in recent years,the method 1 can remove the blur in the image better,so as to obtain better restoration results.Method 2 can obtain higher-quality restoration results due to sufficient feature extraction and full discrimination.Compared with method 1,the average PSNR,SSIM and MOS are improved by 4.9%,2.6%and 3.3%,respectively.The two methods can not only enhance the visual effect of road traffic image greatly,but also achieve stable and high-quality restoration of monitoring blurred image,and have good generalization performance in different road traffic scenes.