Study Of Localization Of Intelligent Vehicle Based On Monocular Machine Vision

Intelligent vehicle local positioning technology based on monocular machine vision is an important part of the vehicle environment perception technology.It can not only realize the reconstruction of the two-dimensional local environment of the vehicle,but also obtain the relative position and posture of the vehicle in real time to achieve the function of assisted driving.In this paper,the research work is the irregular and complete distortion correction of the monocular wide-angle camera.The main content of this paper is shown as follows:1.This thesis summarizes the background and development of intelligent vehicle environment perception technology and local positioning technology,analyzes the problems and difficulties about them.2.In order to solve the large distortion,uneven distribution of radial and tangential distortion,bad image mosaicing effect problem of the wide-angle monocular camera distortion.In this paper,two interpolation algorithms,TPS(Thin Plate Splines)and Biharmonic,are used to correct wide-angle camera distortion,the experiment verifies its validity.On this basis,aiming at the poor effect of the correction result of corner region,and according to the special distortion condition of wide Angle camera,raised a distortion correction algorithm based on improved TPS interpolation algorithm to realize the global correction of irregular distortion.The algorithm used the improved TPS interpolation algorithm to fit two camera offset surfaces according to the mapping relationship between the detected checkerboard and the expected checkerboard,the distortion correction image is obtained by inverse bilinear interpolation.The experiment verifies the global correction technology of irregular distortion based on improved TPS interpolation,improves the accuracy of camera distortion correction,and established foundation for image Mosaic and local positioning of vehicles.3.In order to solve the slow speed problem of usual image mosaicing method based on feature point matching,apply the improved pyramid Lucas-Kanade optical flow tracking algorithm instead of the feature point matching method,avoid the feature point detection and the complex calculation process effectively,to meet the accuracy of image matching of the vehicle environment,increased the sequential image matching speed largely.4.A position and posture experiment system of the intelligent vehicle based on monocular machine vision was designed,and the effectiveness of the algorithm in this paper was verified through the road scene experiment of the real vehicle.