Research And Improvement Of Key Technology In Binocular Stereo Vision

Stereo vision is a task to obtain the depth information of the environment by simulating human binocular vision.Its principle is to predict the depth of the object in the threedimensional space through computational imaging and mathematical modeling,so as to achieve the purpose of restoring the three-dimensional information of the scene.As one of the most popular research directions of computer vision,stereo vision is widely used in automatic driving,virtual reality,augmented reality,three-dimensional measurement and three-dimensional reconstruction and other hot fields.In computer stereo vision,the most commonly used method is binocular stereo vision.The difficulty of binocular stereo vision is how to correctly establish the one-to-one mapping relationship of all pixels between binocular images.Therefore,binocular stereo matching has always been the key research direction of binocular stereo vision.Firstly,binocular stereo matching error has been introduced in the process of image distortion correction.How to reduce the influence of distortion correction on binocular stereo matching algorithm is always a difficult problem in binocular stereo vision.In addition,the image often has uneven illumination and contains a lot of noise in the real scene.How to reduce the influence of illumination and noise on binocular stereo matching algorithm is also a problem that has plagued the field of binocular stereo vision for a long time;because there are few features in the weak texture region,it is extremely easy to cause the matching ambiguity.How to improve the accuracy of the binocular stereo matching algorithm in low texture scene is the key problem in the field of binocular stereo vision.Affected by these problems,it is still a very challenging research topic to quickly establish the one-to-one mapping between pixels from binocular images and recover the depth information of the scene completely and correctly.In this paper,aiming at the above difficult problems in binocular stereo vision,a comprehensive and in-depth research is actively carried out,and improved and optimized algorithms are proposed.The main work and innovation of this paper are as follows:1.Local mapping adaptive Gaussian interpolation algorithm in distortion correction.Aiming at the problem of aliasing and edge image quality degradation in binocular distortion correction,this paper proposes a local mapping adaptive Gaussian interpolation algorithm in distortion correction.According to the image distortion mapping,this method uses different Gaussian convolution to interpolate the distorted image in different positions of the distorted image,which can not only do anti-aliasing for the high-frequency components in the image during distortion correction but also solve the image quality degradation of Gaussian interpolation algorithm at the edge of interpolation image,and the accuracy of binocular stereo matching is improved.2.Adaptive weight anti-noise census transform.In order to reduce the radiation and noise in binocular stereo matching,this paper proposes a cross mean bilateral filter,which improves the noise reduction ability of bilateral filter for all kinds of noise,and retains the edge preserving characteristics of bilateral filter.In addition,on the basis of this filter,this paper proposes an adaptive weight anti-noise census transform,which enhances the anti-noise ability of census transform to improve the robustness of binocular stereo matching to noise and radiation.3.Segment tree binocular stereo matching algorithm based on Collaborative Optimization of segmented regions.In order to solve the problem of binocular stereo matching algorithm for weak texture images,this paper proposes a segment tree binocular stereo matching algorithm based on segmentation region collaborative optimization.In this paper,through the collaborative optimization of image segmentation region,the error matching rate of the matching algorithm for the weak texture region is reduced,and the binocular stereo matching results of the algorithm in the low texture scene are greatly improved.In addition,this algorithm also uses an innovative matching cost calculation method based on Gaussian distribution,which combines a variety of pixel features and achieves good matching results.