Research On Key Technology Of Video Preprocessing Based On Bridge Anti-collision Early Warning System

In the design of large-scale bridge anti-collision warning system,by the imaging system,vehicle vibration and other factors of interference,resulting in shooting video jitter,making our collection of images is not ideal.In the practical application of the project,we are demanding the quality of the image.Digital imaging system in the shooting process of jitter will cause the acquisition of video sequence instability,which will cause the video quality and visual effects worse,seriously affect the follow-up image information processing.At the same time digital imaging in the formation of the process of motion blur,point diffusion blur,under sampling and noise and other factors resulting in degradation.How to accurately detect and track the target,and put forward the challenge to the video stabilization technique and the super-resolution reconstruction technology.In this paper,the video resolution technology and the super-resolution reconstruction of the image are studied,including:(1)In the key technology research project of large-scale bridge anti-collision warning system,the jitter caused by external disturbance in the video acquisition process will affect the calibration and tracking of the subsequent ships.This paper studies the video preprocessing algorithm(video image stabilization).Which is different from the traditional method of stabilization is that the classic image stabilization method is to collect the target object is still.In this paper,combined with the practical application of large-scale bridge anti-collision system,the target object in the video capture is to move large ships,but the moving ships cause interference to the motion estimation part of the electronic image stabilization algorithm.So this paper combined with the practical application of the traditional image stabilization technology has made some improvements.In the stage of motion estimation,considering that the riverbank scene is stationary,the motion estimation can be carried out by using the still part in the collected image.In this paper,the traditional method of feature point matching is used to combine the edge of the bank and the edge of the river and adopt the Canny edge detection method Automatically detect the edge of the river bank and the river,and then erase the edge of the region(ie,the ship's moving area)for feature matching.In the feature point matching part,when the traditional method matches the general frame rate is low,this paper combines the FLANN algorithm and the SURF algorithm,and uses the SURF to extract the key points and descriptors,and matches with the FLANN.The RANSAC algorithm is used to remove the error,and the global motion vector is estimated by the position change of the feature points other than the edge line.The image transformation model(including rotation,scaling and translation)is calculated by combining the image transformation model mentioned above.In the motion filtering stage,the jitter component is used as noise,the global motion vector is processed by Kalman filter,and the noise is filtered by recursively estimating the optimal state of the system,and the effective separation of the subjective scanning component and the jitter component is obtained.In the motion compensation phase,the backward bilinear interpolation method is used to reconstruct the image pixels,and a new compensation method is used to fill the black undefined area in the image,which fully combines the reference frame and the image information of the current frame.Stable image no longer exists.The experimental results show that the electronic image stabilization algorithm proposed in this paper can meet the needs of large bridge anti-collision system.(2)The image is affected by various factors in the collection process resulting in degradation,so that the follow-up target detection and tracking impact,as much as possible to reduce the degradation caused by the detection rate of low-level tracking process of leakage and false alarm Problem,this paper first studies the traditional image super-resolution reconstruction method.Aiming at the practical application,a super-resolution reconstruction technique based on Gaussian pyramid hierarchical Vandewalle registration algorithm is proposed.In the sub-pixel registration phase,taking into account the Gaussian pyramid method is a coarse to fine strategy,in view of Gaussian pyramid hierarchical processing can reduce the search time,making the higher accuracy of the advantages of matching;and Vandewalle registration algorithm is to use no aliasing And the robustness of this method is strong.In this paper,a Vandewalle registration algorithm based on Gaussian pyramid stratification is proposed.Experimental results show that it can preserve image detail and improve the image reconstruction quality.