Research On Depth Information Extraction Technology Based On Binocular Stereo Vision

Binocular stereo vision technology is a major branch of computer vision,and it is also one of the most widely used methods to imitate human eyes to perceive the depth information of three-dimensional objects.It is widely used in three-dimensional reconstruction,unmanned navigation,robot positioning and other practical applications.Binocular stereo vision technology is a process that two cameras placed in different perspectives are used to observe three-dimensional space objects and obtain two images,then binocular calibration,stereo correction,stereo matching and depth information acquisition.The depth information extraction technology based on binocular stereo vision is an important means to perceive the information of three-dimensional environment,which lays the foundation for threedimensional reconstruction,robot navigation,landform detection and other practical applications.This paper focuses on the depth information extraction technology of binocular stereo vision,and the main innovative work is as follows:1.To solve the problem that stereo matching is easily affected by noise and the matching effect is poor in weak texture region and disparity discontinuous region,an improved semi global stereo matching algorithm based on the fusion of census transform and sub-pixel interpolation is proposed.Firstly,the matching value is obtained by the local window census transform after the center point is replaced by the median value and then weighted and summed,then weighted and fused with the sub-pixel interpolation matching value combining the gradient and gray value,then semi global cost aggregation based on adaptive confidence gradient and over counting correction is carried out,finally disparity calculation and disparity optimization are carried out.The experimental results show that the algorithm is not easy to be disturbed by noise,and the matching effect in weak texture area,disparity discontinuity area and occlusion area is improved,and the error matching rate of disparity image is reduced.2.Because the calculation process of disparity is based on a single pixel,in order to avoid the problem of edge discontinuity and noise mutation of disparity,this paper improves the process of disparity optimization,and proposes a weighted median filter algorithm based on phase consistency constraint.The core idea of the algorithm is to introduce the structural information of the maximum moment of phase consistency which embodies the edge features as the regularization term of the weighted median filtering of disparity map.The maximum moment of phase consistency can distinguish different texture complexity and different color regions,and effectively improve the accuracy of matching disparity value.Experimental data show that the algorithm has lower error matching rate than the traditional weighted median filter algorithm and image segmentation algorithm,and has higher precision in the areas of weak texture and disparity discontinuity.3.In order to verify the advantages and disadvantages of the algorithm in the actual scene application and obtain the depth information of three-dimensional objects,a binocular stereo vision development and application platform is built.Firstly,the camera calibration in the binocular stereo vision system is carried out,and the stereo correction is carried out for the left and right image pairs captured in the actual scene.Then,the disparity map is obtained by stereo matching using the algorithm proposed in the paper.Finally,the depth information corresponding to the disparity map is calculated by the triangulation principle.Through the analysis of the experimental results,it is verified that the difference between the obtained depth information and the manual measurement results is very small,the three-dimensional image effect is good,and the platform and the proposed algorithm are feasible in practical application.