Research On Road Surveillance Video Clarification Based On Deep Learning

With the widespread use of electronic imaging equipment,road surface monitoring plays an important role in urban traffic management and social safety.However,due to the relative movements of the monitoring equipment and the vehicle,the road surveillance video is often partially blurred.Moreover,due to various factors,additive noise also appears in the surveillance video during transmission and recording.All these problems have caused the image degradation of surveillance video.In order to obtain high-quality surveillance video that meets the needs of society,the problem of denoising and deblurring the video is studied in this article.The main research content and results are as follows:(1)In view of the salt and pepper noise existing in the video,according to the characteristics of relatively isolated noise points(blocks),a video image denoising algorithm based on connectivity is proposed.The algorithm first uses the area growth algorithm to determine whether the pixels belong to the connected area of the image,and then uses the median filter to filter out the detected noise points.Experimental results show that the denoising algorithm proposed in this paper is better than several comparison algorithms at a noise density of 0.1 to 0.9.(2)In order to find the blurred area in the video image,a new blur detection neural network is designed based on the U-shaped neural network,which has sub-module inputs different from the original network structure and new jump connections.In order to verify the effectiveness of the designed network,this paper compares the blur detection network with several classic blur detection algorithms,and the experimental results show that the network structure can achieve better blur detection results.The experimental results show that the network structure can achieve better blur detection results.(3)Considering the correlation and continuity between video frames,a recurrent neural network with multiple adjacent frame image blocks as input is designed.Then,using the recurrent neural network as a generator,combined with Markov discriminator,a generative adversarial network for deblurring image blocks is constructed.Finally,the video deblurring algorithm is designed by combining the deblurring network and the blur detection algorithm.Comparative experiments show that this video deblurring algorithm can achieve better deblurring effects than several other algorithms.Finally,according to the proposed algorithm,the software part of the road video clearing processing system was designed,and the graphical user interface of the software was designed with Tkinter.This software can realize the de-noising and de-blurring of road surveillance video so it has practical application value.