3D Reconstruction Based On Binocular Stereo Vision

Three-dimensional reconstruction is to restore the 3D information of all or part of the surface of the target object in the scene.It has a wide range of applications in many fields such as aerospace,unmanned driving,building modeling,and industrial inspection.Binocular stereo vision is a way to achieve three-dimensional reconstruction.It uses two identical cameras to simulate the human visual system,obtains two images of the surface of the same space object from different perspectives,and uses the principle of parallax and similar triangles to restore the depth of the surface of the space object.Information is simple to implement and effective.This paper mainly studies 3D reconstruction methods and technologies based on binocular stereo vision,including camera calibration,stereo matching,and 3D reconstruction.The focus is on improving the two stereo matching algorithms.main tasks as follows:(1)In the camera calibration part,the simple operation and high precision Zhang Zhengyou calibration method were studied,and in the experimental part,Open CV and Mat Lab were used to realize the camera calibration link.On the basis of comparing the accuracy of the two,we obtained The more accurate camera parameters are used to constrain two different images taken by the same two cameras and the subsequent three-dimensional reconstruction visualization.(2)The traditional stereo matching algorithm based on Census transform has a high dependence on the gray value of the center pixel and poor anti-interference ability.Propose an improved Census+Guided Filter(GF)stereo matching algorithm.First,the pixel detection module is introduced,and the pixel gray value is recalculated by weighted median;second,the matching cost is calculated by fusing the absolute difference(AD)cost with the Census cost;finally,on the basis of GF Above,the robustness of the algorithm in different areas is strengthened.The experimental comparison results show that the algorithm proposed in this paper has better noise resistance,and the resulting disparity map is clearer.(3)Aiming at the problem of mismatches caused by sudden changes in pixels of ADCensus under external conditions and blurring of the edge regions of objects in the disparity map,a stereo matching algorithm of APC+DASW(ADPhase Census+Double Adaptive Support Window)is proposed.First,the initial image is denoised;secondly,the phase value is added when calculating the AD cost,and a matching cost calculation method that combines the AD cost,the phase angle cost and the Census transform cost is proposed;finally,in the cost aggregation step Improved the calculation rule of the adaptive window.The experimental comparison results with the original algorithm show that,compared with the original algorithm,the algorithm proposed in this paper produces better results in both the edge area and the smooth area.(4)In the 3D reconstruction part,the depth information of the actual object surface is calculated from the disparity map using the principle of similar triangles,the depth information and pixel color values are saved as point cloud information,and the 3D reconstruction results of the object are displayed through 3D visualization.