Research On Dehazing Of Traffic Image Based On Simultaneous Polarization Imaging

In recent years,target recognition of optical imaging systems under low-contrast environmental conditions has become a key research direction in my country’s development strategy projects.Polarization imaging technology has the advantage of "penetrating through clouds and fog",which is to increase the detection information dimension of the image by collecting images of different polarization angles of the same target scene.Therefore,the polarization imaging technology can clear the haze image affected by the chaotic medium.Under severe weather conditions such as fog and haze,the performance of outdoor optical imaging systems is degraded,resulting in reduced image quality and loss of image details.Aiming at the problem that the traffic image is affected by haze weather,the visibility of road traffic is deteriorated,and the traffic scene cannot be accurately identified.This paper studies the image processing algorithm of the simultaneous polarization imaging system based on Wollaston prism.The main research contents are as follows:First,the overall design of the simultaneous polarization imaging system based on Wollaston prism is carried out.In view of the excessive number of detectors used in the current simultaneous polarization imaging system,and the high cost and complexity of the system,this paper designs a polarization imaging system that shares a front objective lens and an imaging detector.Angled polarized image.The dual-axis sub-aperture convergent imaging is designed,and the two orthogonal polarization state lights are condensed and imaged respectively,which reduces the aberration of the polarized image and facilitates post-image processing.Secondly,the image registration algorithm is studied.The SIFT registration algorithm based on point features is studied.In view of the problem that the algorithm fails to remove some edge points,resulting in mismatching of feature points and slow processing time,an improved SIFT and RANSAC algorithm based on random sampling are proposed.The registration algorithm is optimized to further remove mismatched point pairs and improve the registration rate and processing rate of the image.Using the improved registration algorithm,the registration of four polarized images of the same target scene is completed.Finally,the dehazing algorithm for traffic images is studied.Aiming at improving the accurate estimation of atmospheric light,a polarization dehazing algorithm using target and atmospheric polarization information is studied.The four polarized images are differentially filtered by Gaussian filtering fused with gradient information,and the corresponding atmospheric light images are estimated.A search method of quadtree subspace division is proposed,which estimates the atmospheric light intensity at infinity in the atmospheric light image,constructs an improved polarization dehazing model,and realizes the clear traffic image in fog,haze,rain and snow weather.Using the polarization imaging system designed in this paper,four polarization images of the traffic scene were collected in severe weather such as haze.Through the dehazing results of the images obtained by multiple sets of experiments,it is concluded that the evaluation indicators of the improved polarization dehazing algorithm in this paper are better than Dark channel prior and Stokes vector polarization dehazing algorithm.