| Lidar-based 3D laser scanner is widely used in reverse engineering,the point cloud data obtained by it has the characteristics of high density,high precision,and become a new and reliable data source for surface structure information acquisition.Therefore,the algorithm study of three-dimensional point cloud rule surface fitting is very important.At present,researchers at home and abroad have put forward a variety of fitting algorithms,however,the traditional fit algorithm based on the least square algorithm in practical engineering applications often ignore the error in the coefficient matrix,which affects the precision of fitting.In view of this situation,a three-dimensional point cloud rule surface fitting algorithm based on the overall minimum square,with spherical surface,column surface,tapered surface,etc.as the fitting object,and design a set of experimental process for scattered point cloud data,including noise reduction,scanti-severation,registration,fitting,etc.,to verify the feasibility of the algorithm.The main elements and results are as follows:(1)The scattered point cloud is scattered to establish a neighborhood spatialrelationship.Based on the neighborhood relationship,the target and curvature ofeach point are estimated.The curvature of the point cloud allows you to calculate the Euclidean distance between any two points.On this basis,the corresponding threshold is set to determine whether the noise is the surface noise of the point cloud subject or the outlier noise.Depending on the nature of noise,different filtering methods can be used to reduce noise.Experiments show that the noise of outlier can be effectively removed by using statistical filtering,and the noise of the main body of the point cloud can be effectively removed by using radius filtering.(2)In order to improve computing efficiency and save space,the data redundancy of point cloud is reduced.Reducing data redundancy is divided into two aspects:directly reducing point cloud points and eliminating unnecessary data information.Using the random resampling algorithm,the point cloud can be randomly deleted according to the pre-set threshold,and the number of point clouds can be reduced directly.The octagonal tree dilution algorithm can effectively further compress the point cloud data,eliminate unnecessary information,and retain only the location information of the point cloud.(3)Point cloud registration is divided into coarse registration and fine registration,and coarse registration is the premise of fine registration.First,the feature points of the common part of the two point clouds are extracted using the SIFT(Scale Invariant Feature Transform algorithm),and then,according to the Euclidean distance between the characteristic point vectors,the characteristic points of the two point clouds are paired,and the characteristic point pairs are purified by using the angle of the method vector Finally,through the unit quad snumber method,the rotation and translation matrix are solved,and the rough registration is completed.On the basis of this rough registration,a good initial iteration value is provided for the fine registration.The fine registration adopts the ICP algorithm,and iteratively calculates the source cloud to the target point cloud as a whole.(4)Point cloud rule surface fit.In order to overcome the defects of error in the coefficient matrix that cannot be taken into account in the traditional minimum-by-two-fit algorithm,this paper uses the overall minimum square-by-multiplier algorithm to improve.First,the initial value of the iteration is calculated by the minimum square-by-second algorithm,and the iterative calculation is calculated by using the overall minimum square-by-multiply algorithm.Using this method,the spherical surface,column surface and cone surface are tested,and the fitting algorithm is programmed to be implemented in the C-language,so as to verify the feasibility of the algorithm.The experimental results show that the algorithm fits better. |
PDF Full Text Request