3D Reconstruction Algorithm Of Point Cloud Data From Structured Light Scanning

The thesis works comes from the project of “Safety Status Monitoring of Heavyload Train on Board and from Ground;Demonstration and Verification”.The reconstruction algorithms of the 3D point cloud of train bogie are studied here.The content of this thesis work is given as follows:First,using structured light scanning,the data acquisition scheme and algorithms of 3D point cloud of train bogie are studied.A data acquisition platform is developed,and camera and light plane parameters are calibrated.The algorithm to extract the laser stripe centerline is improved,through introducing a function to weight the laser stripe centerlines respectively obtained by the geometric center method and the thinning method.Second,the 3D point cloud data is preprocessed.The spatial position distribution characteristics of noise points is analyzed,and.based on classification concept those redundant points and outliers generated in the non-target scanning field are eliminated separately.The threshold method is used to remove the redundant points to obtain the subject point cloud,and in terms of the point cloud density,the outliers are removed by an adaptive statistical filtering and a multi-condition threshold constraint function.Then,the voxelized down-sampling and the resampling of high-order polynomial interpolation are used to smooth the filtered point cloud while maintaining the geometric form of point cloud,and an average curvature and Gaussian curvature are used to evaluate the smoothing.Third,two surface reconstruction algorithms of 3D point cloud data are studied.The NURBS interpolation surface is used to fit the surface of a train bogie model,and a closed,smooth surface with no holes is obtained.With the regional growth surface reconstruction of spatial distribution,the Delaunay triangulation criterion is used to establish the triangle mesh topology relationship to reconstruct the surface of the train bogie.Compared with the NURBS algorithm,the later has better robustness and achieves better visualization.An RGB pseudo-color index map is constructed based on depth information,and the object texture is mapped to the surface of the reconstructed model through the forward mapping method to improve visualization effect.