Precise elimination method for laser point cloud noise and redundant data

Abstract

According to the precise elimination method for the laser point cloud noise and the redundant data provided by the invention, for an error existing in a laser scanning process, the laser point cloud data contains the noise, a point cloud denoising method based on geometrical characteristic estimation is adopted to carry out noise elimination on the laser point cloud data obtained by scanning; forthe characteristic that the laser point cloud data has a large data volume, redundant data in the laser point cloud data is eliminated by adopting a boundary-reserved adaptive point cloud simplification method; according to the invention, a k neighborhood denoising and bilateral filtering method is improved and fused, and small noise information in the model is smoothed while large outliers in thepoint cloud model are efficiently removed; boundary information of a point cloud model is comprehensively considered while point cloud simplification is carried out, boundary information loss is reduced while feature information of a physical model is reserved, and the precision and reliability of laser point cloud data after preprocessing are guaranteed.

Description

technical field [0001] The invention relates to a method for eliminating point cloud noise and redundant data, in particular to a method for accurately eliminating noise and redundant data of laser point cloud, and belongs to the technical field of laser point cloud data processing. Background technique [0002] 3D laser scanning is a 3D laser point cloud data acquisition and model reconstruction technology, which can obtain a large amount of 3D laser point cloud data on the physical surface in a short time. Compared with traditional single-point measurement, the main advantages of 3D laser scanning technology are: fast data acquisition speed, large data volume, millions of points scanned per second, strong initiative, all-weather operation, simple and convenient operation, The measurement scan controls the operation process through software; the acquired data has full digital characteristics, which is convenient for information transmission, processing and expression. The ...

AI Technical Summary

Problems solved by technology

However, the noise and feature information of the real object have a certain similarity. It is very difficult to distinguish the noise data and the point cloud feature information when the point cloud is denoised in the existing technology. The k-nearest neighbor denoising algorithm cannot deal with the small noise in the laser point cloud data; Bilateral filtering is only suitable for laser point cloud data with low noise. Bilateral filtering for data with high noise tends to make the result too smooth. This algorithm performs smoothing on all points in the point cloud. The smoothing and denoising of the cloud will distort and deform the non-noise data; the denoising accuracy of laser point cloud data is low, and the degree of automation is low; a large number of human-computer interaction methods in the existing technology consume a lot of working time and high work cost;

[0010] The third is that the mainstream point cloud simplification method in the existing technology mainly retains the geometric feature details of the object as much as possible and simplifies the efficiency, and does not consider too much the boundary point detection of the point cloud, which reduces the boundary of the simplified result and affects the subsequent surface reconstruction. precision

In order to preserve the boundary information of the object while satisfying the simplification efficiency and ensuring the geometric information of the point cloud, the adjacency relationship of the data points in the neighborhood is used to detect the boundary feature points. The existing technology is based on the k-nearest neighbors. The angle distribution of the projected points on the tangent plane is uniform It is also possible to determine whether the point is a boundary point, or to divide the point cloud into a three-dimensional grid, find the point closest to each grid center as the initial cluster center, and then use the k-means clustering method to cluster the laser point cloud data. Class, through the boundary clustering center to detect the corresponding boundary clustering point cloud, thus avoiding the calculation of each point in the point cloud, but this method selects the initial center and then clusters the point cloud, and at the same time, each The clustering set also needs to calculate the maximum deviation of the normal vector in each subset to decide whether to subdivide the cluster. This algorithm is very sensitive to noise data, the efficiency is not high, and there will be a certain loss at the boundary;

[0011] The fourth is that the boundary shrinkage caused by the point cloud simplification process of the prior art, all points are judged as boundary points, the efficiency is very low, and the existence of redundant data reduces the efficiency of data storage and calculation. The point cloud simplification method of the prior art only looks at The simplification rate or simplification speed, simplification rate, simplification speed and point cloud simplification accuracy cannot reach a balance. It is impossible to use as few points as possible to represent more information about the object while ensuring the point cloud simplification accuracy. The operating efficiency of the algorithm Low, the elimination precision and accuracy of point cloud redundant data are low, and the quality and accuracy of point cloud data are obviously low

Images