Research On Vehicle3D Reconstruction Method Based On Monocular Polarized Light

As we all know, our automobile industry has made a huge progress in recentyears. On one hand, the vehicle holdings has increased significantly in whole country.On the other hand, the traffic issue has been showing apparently. The better method toimprove traffic environment is to strength management of road traffic. It needs toanalysis road traffic flow and behavior of traffic participates, then it is necessary toget relevant data of important traffic participates (automobile). To build automobile3D model can obtain some information about depth of traffic scene and help to makea more thorough research on traffic flow data. The binocular3D reconstruction is thebest method in3D rebuild field. While, because the cost of using high precisioncamera is very high, then decrease the cost is the one of the direction on this study.Based on above two issues, it proposes a method of3D automobile reconstructionaccording to the theory of monocular polarized light. It can cut down the cost of3Dreconstruction efficiently and also make3D reconstruction reality.Firstly, this paper introduces the study of polarized light and the research statusof stereo vision on domestic and overseas. With comprehending all aspects ofapplication research on polarized light and it provides ideas about making full use ofpolarized light for this research. Besides, according to analysis of the basic theory andspecific implementation steps of stereo vision, the advantages and disadvantages ofvarious methods can be given on the stereo vision research.Secondly, this research makes a detailed analysis on theory of the key stepsduring3D reconstruction, introducing camera calibration model, and analyzes thedisadvantages of traditional calibration, active vision calibration and self-calibration.Contrasted to the theory of features point extraction between Harris and SUSAN, itdescribes both of strengths and weaknesses. Two categories are divided into featurematching algorithm and gray window regional algorithm on account of feature pointsmatching algorithm. It introduces two methods as SIFT and covariance window grayrelated operate and also describes both of matching theory and implement process indetails.Then, based on a series of polarized light features, it builds a system which canget some different scene images information by monocular camera in order toimplement binocular camera images capture. It can reduce the system hardware costin3D reconstruction effectively. As followed, it is a part of key steps problems solving and implementation. Itcompletes camera stereo calibration and verifies its result, then to correct capturedimages and propose improved feature points detection algorithm to get sub-pixel levelfeature points and also keep scale invariant. According to take window gray featurepoints matching algorithm and improved matching operator to judge the matchingpoints, it gains the disparity of binocular images.Finally, it makes the trigonometry operations to get the information about depthof traffic scene and draw3D points cloud figure as well. It verifies system’seffectiveness and the error of outdoor road test. Above all, it can conclude that it canmake a better3D reconstruction to the road traffic scene based on images collectionof monocular polarized light and3D reconstruction algorithm.