Real-time Stereo Registration Algorithm Based On Binocular Vision

With the development of Internet technology,computer vision has extended to all fields of life.Binocular vision,as an important part of computer vision,is widely used in driverless,medical imaging,intelligent navigation and other emerging fields.Binocular vision simulates human visual function,shoots the same scene,obtains 3D information by calculating parallax and 3D ranging principle,and then makes the computer have a certain perception of the objective world.In this paper,the real-time stereo registration algorithm based on binocular vision is studied,which improves the real-time and accuracy of 3D reconstruction.The specific contents of this study include:(1)Binocular camera calibration and image correction processing.Firstly,the camera imaging principle and coordinate system conversion rule are introduced to calculate the binocular camera parameters,then the camera is calibrated by Zhang Zhengyou chessboard calibration method,and the internal and external parameters of the binocular camera are obtained.On this basis,the corrected left and right images are located at the same horizontal baseline to speed up the subsequent stereo matching.(2)Feature point extraction and stereo matching.The feature point extraction of the corrected image is carried out.By comparing SIFT?SURF?ORB three feature point extraction algorithms,the best performance ORB is selected as the corrected image extraction feature point,and the feature point is described to make it have rotation invariance and scale invariance.Then stereo matching is carried out.Based on the theory of stereo matching algorithm,by comparing GC?BM?SGBM three stereo matching algorithms,the parallax value of each pixel is obtained by selecting SGBM for stereo matching.(3)In view of the problem of poor real-time performance and low accuracy in the three-dimensional reconstruction,the agent aggregation algorithm is adopted to improve the speed of obtaining parallax map,thus improve the speed of generating point cloud,and conduct noise reduction with the graph optimization;the pose estimation algorithm is used to find out adjacent points and integrate and obtain the complete 3D point cloud.(4)Build a real time 3D reconstruction experiment platform based on binocular vision.VS2017 as the compilation environment using C language and Opencv3.1.0 ? PCL1.8.0.Realize realize modules: image acquisition,camera calibration,feature point extraction,3D matching,3D reconstruction,and loop detection,and use multiple sets of datasets to analyze and compare the reconstruction effect,which verify the real time nature of binocular visual reconstruction.