Research On Key Algorithms Of Point Cloud Data Preprocessing

Point cloud data processing technology is a key technique in the field of digital detection.Due to the improvement of computational efficiency and high sampling rate,point cloud data processing technology is widely used in the field of reverse engineering and target identification.However,due to the inherent measurement errors and digitization processes of the equipment,the raw data inevitably contains noise;and there are the limitations of data acquisition,such as the limited field of view.Therefore,the denoising and registration of point cloud has become a research hotspot.Reducing the time complexity of point cloud denoising algorithm while maintaining the characteristics of point cloud model is discussed in this work.A point cloud denoising algorithm coupled with multiple feature point parameters is proposed.Firstly,the k-d tree is used to establish the three-dimensional topological relationship,and then the relationship between the parameters of each point(the normal vector of the point,the distance from the point to the center of gravity,and the average distance from the point to the neighbor node)is analyzed.To further explore the influence of the parameters of each feature point on the denoising effect,the weighted parameters of each feature point are discussed,and the differential evolution algorithm is used to optimize the weight with the peak-to-noise ratio as the constraint condition.Experimental results show that the point cloud characteristic model is effectively maintained,the weight of each parameter is established,and the denoising of point cloud is effectively realized.The algorithm provides a rich theoretical support for point cloud denoising.The scale parameters of the point cloud cannot be obtained effectively when facing a large number of interference points,loss points,and singular points.To effectively solve the scale parameters,the solution of the scale parameters of point cloud is studied.A method for solving point cloud scale parameters based on probability estimation is proposed.The solution of the scale parameters of point cloud is transformed into the solution of the statistical parameters of point cloud using the principle of probability estimation.The mean value,variance,and covariance matrix of point cloud is analyzed emphatically,and the relationship between the matrix traces of two points cloud is discussed.Experimental results show that the proposed algorithm can effectively solve the scale parameters,and has high robustness even when the point cloud is accompanied with interference and data loss.Improving the efficiency and the accuracy of point cloud registration has always been the focus of point cloud registration.To improve the registration efficiency and the registration accuracy of point cloud registration,an algorithm for solving point cloud rotation parameters based on the convolution neural network is proposed.Using point cloud data as the input for the convolution neural network is studied and the structure of the convolution neural network is discussed based on the principle of convolution neural network in this work.Firstly,grid statistics and histogram statistics are proposed as the statistical methods of point cloud data to realize the point cloud dimension reduction from three-dimensional data to two-dimensional data as the input of convolution neural network.Secondly,the structure of the convolution neural network is discussed.The size of the convolution kernel,the design of the pooling layer,and the design of the activation function of the convolution neural network are mainly analyzed.Finally,the experimental simulation is conducted and the results show that the algorithm not only has high speed and good registration accuracy but also has high robustness.