Point Cloud Quality Improvement And Surface Reconstruction Of Train Key Components

The detection of train key components is really important for the safety of train in operation,and the recognition,registration,segmentation,and reconstruction based on 3D point cloud have been applied to the train components detection systems successfully.However,In the process of acquiring point cloud data by a 3D laser scanner,some problems,such as outliers,mixed points and holes,may be caused in the target point cloud due to the external environment,the discreteness of the laser beam,and the occlusion of objects,which were regarded as a low quality point cloud.The point cloud with a low quality would affect the accuracy of detection and registration,leading to a bad performance in 3D surface reconstruction,too.In order to improve the quality of point cloud,these following researches are conducted in the thesis:1.An filtering and denoising algorithm is proposed in this thesis,A self-adaptive octree is established to generate many grids in point cloud and calculate the density in each grid,combing with the statistical filtering to remove outliers from the point cloud data.Then a plane projection method is used for removing the confounding points from the point cloud data.The denoised experiments on standard point cloud dataset and the real train components pantograph show that while removing outliers and confounding points,the detailed features of the point cloud can be maintained,and the quality of the point cloud is effectively improved.2.There are many related parameters need to be set manually in traditional filtering and denoising algorithms,these parameters would affect the performance of algorithms towards different point cloud.Therefore,this thesis proposed an denoising algorithm via deep learning based on pointcleannet,which uses an edge extraction algorithm to keep the sharp features and details in the point cloud.This method is evaluated based on the standard point cloud dataset and some real key components point cloud that collected from train.The experimental results show the method in this thesis performances better with removing outliers and confounding points effectively in low quality point cloud.3.For solving the problems of uneven insertion points and uneven size of the triangular patches that be added in the process of hole repairing,this thesis proposed a repair algorithm combine with priority value.A large priority value indicates that the area of the triangle patch for repairing is the largest with the hole in point cloud,and then points are inserted in the part with the largest priority value,until the maximum priority value is higher than the threshold value.Finally,the patching is constructed with insertion points and original points.This algorithm is evaluated on the standard point cloud data and the real train pantograph with holes.The experimental results show that the algorithm in this thesis inserting points uniformly,and the size of added triangular patches are uniformly too.4.For 3D surface reconstruction algorithms,this thesis compares several surface reconstruction algorithms that usually used.Then use the deep learning network to predict the normal vector required by the surface reconstruction algorithm,And it is evaluated on the real train components such as wheel sets,screws,the base of bottom and pantograph point cloud data.The experimental results show that the combination of deep learning and traditional algorithms can be well applied to the surface reconstruction of key train components.