Research On Video Image Stabilization Method In Tunnel Environment

Accurate extraction of traffic targets in tunnels is the key to the detection of abnormal tunnel events.In the tunnel scene,there is jitter in the monitoring equipment,which makes the extracted vehicle target deformity,pedestrian target and background connectivity,and seriously interferes with the effective extraction of traffic targets.At the same time,the image quality of the tunnel image is blurred,and the interference of vehicle lights also increases the difficulty of the video stabilization of the tunnel,resulting in the conventional image stabilization method with general effects and weak pertinence.Therefore,researching video stabilization methods based on the characteristics of the tunnel environment has important theoretical and practical significance for improving the detection accuracy of traffic targets such as vehicles and pedestrians in tunnels.This paper analyzes the advantages and disadvantages of common jitter vector estimation methods and comprehensively considers the real-time performance and image stabilization accuracy of various methods,then selects the gray scale projection method with high real-time performance and the high precision feature point matching method as the basic stabilization method of this paper.On this basis,this article separately focuses on the image quality blurring and vehicle light interference suppression in the tunnel image.At the same time,in order to further improve the accuracy and real-time performance of the improved image stabilization method,improvements have been made in grayscale projection windows,feature point distances,matching pair selection strategies,etc.Finally,two types of video stabilization suitable for the characteristics of the tunnel environment have been formed.In the aspect of image stabilization based on gray projection,an improved gray-scale projection image stabilization combining local phase quantization(LPQ)weighting and multi-scale Gaussian estimation is proposed for image blurring and vehicle light interference.Firstly,the weighted fuzzy LPQ quantization coding value is used to increase the projection difference between fuzzy image sequences.Secondly,multi-scale Gaussian estimation method is used to approximate and eliminate the brightness component of the tunnel image,so as to reduce the local difference between the reference frame and the background frame projection curve under light interference.Finally,the method based on multi-sub-region window and sample statistics reduces the influence of local moving objects on the accuracy of grayscale projection image stabilization.The experimental results show that this method can effectively overcome the image quality blur,vehicle lighting and other interference,while ensuring a higher real-time performance while obtaining a smooth,stable video stream.In the aspect of image stabilization based on feature point matching,considering the poor real-time performance of traditional robust features,this paper chooses Fast Retina Keypoint(FREAK)which has strong real-time performance as the basic feature.Aiming at the problem that the feature has high false matching rate under the influence of blurred image quality and vehicle lights,this paper proposes an improved binary feature point matching stabilization method based on recombinant fuzzy robust and illumination robust Hu moment invariants.Firstly,the method calculates the distance of the reconstructed Hu moment invariants in the neighborhood on the basis of the four-level matching of the FREAK feature,eliminates the mis-matching pairs with lower similarity,and secondly,taking into account the high clustering and low resolution of feature points in the neighborhood,this paper adopts a distance constraint to limit the number of feature points to reduce the computational cost.Finally,the accuracy of the jitter vector estimation is further improved based on the Progressive Sample Consensus(PROSAC)and the Hamming distance ratio.The experimental results show that this method can effectively overcome the image quality blur,vehicle lighting and other interference,and can obtain a higher image stabilization accuracy while ensuring a certain time efficiency.Finally,based on the above research results,two video stabilization methods suitable for the characteristics of the tunnel environment are formed,and the tunneling video data collected in the field is used for comparative experiment verification.Compared with the traditional method,in the tunnel environment,the methods presented in this paper have higher inter-frame transform fidelity and better image stabilization effect.At the same time,this paper also compares two improved image stabilization methods.The results show that the improved method based on gray projection is faster in image stabilization real-time and the latter is better in image stabilization accuracy.In practical applications,both image stabilization methods can effectively improve the detection rate of target detection in the tunnel environment and reduce the false detection rate.