Research On Indoor Point Cloud Data Generation Based On Binocular Vision

Data acquisition and 3D modeling of indoor scenes are important contents of smart city construction.How to operate three-dimensional reconstruction of indoor scenes quickly,cheaply and easily has become a hot issue in current research.The three-dimensional reconstruction method based on binocular stereo vision uses ordinary digital cameras as the image acquisition device.By imitating the visual mechanism of the human eye,the computer can realize the perception of the three-dimensional scene through multiple images,and has the advantages of low cost,high efficiency and low labor intensity.It can generate dense point clouds,and is increasingly used in the work of 3D reconstruction of indoor scenes.The binocular stereo vision system flow is divided into three steps: camera calibration,image matching and 3D reconstruction.This paper takes binocular vision to generate indoor point cloud data as the research goal.The main research contents are as follows:(1)Developed a horizontal binocular stereo vision system.The hardware is a simple binocular camera that can output images from left and right viewing angles.The software includes modules such as camera calibration,image stereo matching,point cloud generation and splicing,and each module incorporates multiple algorithms.The system enables a complete flow of point cloud based on binocular vision.(2)Research on the camera calibration algorithm,focusing on the direct linear transformation algorithm,Tsai two-step calibration algorithm and Zhang's calibration algorithm.The characteristics of the three algorithms are analyzed experimentally.The results show that the Zhang's calibration algorithm is accurate and cost-effective,better than Tsai calibration algorithm and direct linear transformation algorithm.(3)The stereo matching algorithm is studied and the advantages and disadvantages of the three types of matching algorithms are compared.The results show that the region-based stereo matching algorithm is more suitable for binocular vision systems.On this basis,this paper proposes a new adaptive weighting SAD-Census matching algorithm,which concludes by comparing the matching results: the algorithm performs well in both effect and efficiency,and it has better robustness in the real scene.(4)Perform a complete binocular stereo vision generation point cloud experiment,use Zhang's calibration algorithm to calibrate,use adaptive weight SAD-Census matching algorithm to match,restore 3D scene,generate 3D point cloud,and finally splicing multiple sets of point cloud data.The splicing of the generated point cloud data shows that the developed binocular stereo vision system and the adaptive weighting SAD-Census matching algorithm proposed in this paper can achieve satisfactory results.