Research On Obstacle Detection Method Based On Zoom Stereo Vision

In recent years,automotive intelligence has become a research hotspot in the automotive industry.As one of the most important technologies in intel igent assisted driving,vehicle obstacle detection has been paid more and more attention.Binocular stereo vision can simula te the human binocular perception ability.It can not only provide rich visual information,but also obtain the three-dimensional coordinates of space objects.In this paper,obstacle detection based on zoom stereo vision system is studied.Firstly,the calibration of zoom binocular camera is the basis of zoom stereo vision system.The imaging model and disparity principle of zoom binocular camera are briefly analyzed,and the calibration principle and steps are explained.The single target calibration and stereo calibration are carried out by using the calibration toolbox of MATLAB,and the internal and external parameters and position relations of the binocular camera are obtained.The left and right image pairs are corrected by stereo correction to eliminate the distortion and get the corrected image pairs.Secondly,based on the analysis of the principle of stereo matching and the advantages and disadvantages of the main stereo matching algorithms,a semi-global stereo matching algorit hm is used to match the corrected image pairs.The optimized disparity map is obtained by further edge filtering of the matched disparity map.The U-V disparity theory is analyzed in detail,including the formation of U-V disparity and the manifestation of U-V disparity in differe nt planes.According to the disparity map,the V-disparity map is further calculated.The tilted line representing the ground in the V-disparity map is detected by Hough detection principle,and then the ground is re-projected into the disparity map to filter out.The remaining part is the obstacle.The length of the vertical line segment in the V-disparity map represents the height of the obstacle,so the vertical coordinates of the obstacle can be determined by detecting the length of the vertical line in the V-disparity map.Combined with U-disparity map,the width of the obstacle is calculated,and the spatial position of the obstacle is determined,which is expressed in the form of rod-like pixels.Finally,a zoom stereo vision system is constructed by a micro-electric vehicle equipped with a zoom binocular camera,and the algorithm is transplanted to the embedded developme nt board to complete the vehicle test,which verifies the effectiveness of the proposed algorithm.