Research On Binocular Stereo Matching Technology In Complex Scene

Machine vision is a subject that studies how to make machines use vision to obtain information in the scene like human beings.Binocular measurement is an important branch of machine vision.It imitates the principle of human vision system and can obtain the depth information of scene on the basis of image acquisition.Now it has been widely used in industrial field,traffic assistance,security service,medical and military fields.Compared with the measurement methods relying on other sensors,binocular measurement has the advantages of wider field of view,simple system structure and non-interference between devices.It is an ideal non-contact measurement method for visible objects.Stereo matching task is the core of binocular vision system,which determines the measurement accuracy of the whole system.In this paper,the binocular stereo matching technology is studied.In order to solve the difficult problems in the current stereo matching work and improve the measurement accuracy of binocular vision system,two new stereo matching technologies are proposed.The specific work of this paper is as followsFirstly,a variable window matching algorithm based on object contour information is proposed to solve the problem that the image is difficult to match in the regions with discontinuous depth and uneven illumination.The main contents of the algorithm include: optimizing the edge detection algorithm to make the contour of the object closed better,so as to divide the homogeneous region with high credibility;designing a new similarity measure function,which is robust to illumination mutation,and strengthening the ability of similarity discrimination between windows;proposing a new matching window size self-adjusting strategy to alleviate the image in the depth discontinuity area At the same time,it optimizes the matching effect of the weak texture region in the image;designs a disparity optimization method based on homogeneous region to further optimize the initial disparity map.The algorithm can effectively improve the image matching accuracy in the depth discontinuity region and uneven illumination environment.Secondly,this paper tries to solve the occlusion problem in stereo matching.After introducing the causes of occlusion phenomenon and analyzing all kinds of occlusion situations,a disparity optimization algorithm of adaptive weight image occlusion region is proposed.The content of the new algorithm mainly includes: the occlusion detection method based on the left-right consistency constraint defines the occluded area in the image;proposes an adaptive cost aggregation weight generation strategy to reduce the cumulative cost impact of the unreliable points in the occluded area on the window center point,and improves the matching accuracy of the pixels in the surrounding area;and proposes a new parallax optimization algorithm for the occluded area The pixel set is the most reliable parallax.The algorithm can effectively improve the matching accuracy of the occluded area of the image,and further improve the overall accuracy of the disparity map.Finally,this paper builds the corresponding software and hardware platform of binocular vision ranging.On the basis of camera calibration and image correction,the platform and related algorithms are tested by using real images,which verifies the reliability and advantages of the algorithm.