| As the core of computer vision,3D reconstruction based on binocular vision is of great practical value and research significance in the fields of workpiece detection,medical modeling and high-precision map construction.This thesis focuses on the key technologies involved in the stereo matching and 3D point cloud reconstruction of binocular vision.The main contents of this thesis are as follows:(1)Firstly,the binocular camera is calibrated to obtain the corrected image.On this basis,the image feature extraction algorithm is studied.By graying the image and reducing the amount of calculation,the ORB algorithm is used for feature extraction,and the feature points are roughly matched according to the Hamming distance.Aiming at the problem of false matching of ORB algorithm,the progressive consistent sampling method is used to filter the false matching and retain the matching point pairs with high accuracy.(2)Aiming at the problem that the feature-based stereo matching algorithm can only obtain sparse disparity map,the stereo matching method combining feature extraction and region growth is adopted,and the reserved feature points are used as seed points for region growth to obtain a relatively dense disparity map.Aiming at the problem of holes in the disparity map,the mean filtering method based on integral map is used to fill the empty region of the disparity map.Experimental results show that compared with SGBM algorithm,the consumption time of the proposed stereo matching algorithm is reduced by56.7% on average,and the efficiency of the algorithm is significantly improved,on the basis of the difference between the matching rate and the SGBM algorithm is plus or minus 3%.(3)Research on 3D point cloud reconstruction.According to the calculation principle of the three-dimensional coordinates of the binocular vision system,the three-dimensional coordinates of each point in the two-dimensional image are calculated through the parallax map to obtain the point cloud map.Aiming at the problem that there are some outliers in the point cloud,the outliers are filtered through the statistical filtering method.Aiming at the problem that the single view point cloud cannot completely restore the spatial information in the scene,it is proposed to use the principal component analysis method for rough registration of the point cloud,Then the nearest iterative point algorithm is used to realize the fine registration of point cloud.The experimental results show that the scheme is simple to operate,can better complete the three-dimensional reconstruction of spatial information in the scene,and obtain a more perfect point cloud. |
