CN112465991B - A method for denoising tunnel point cloud and generating visualization model - Google Patents

Abstract

The invention relates to a method for denoising tunnel point cloud and generating a visual model, which belongs to the technical field of graphics and visualization and comprises the steps of collecting tunnel three-dimensional point cloud and preprocessing tunnel point cloud data; denoising: dividing the noise points into noise points in the tunnel and noise points outside the tunnel according to the positions of the noise, dividing the noise points in the tunnel into outlier points caused by a moving target and satellite noise points caused by facilities in the tunnel according to the space distribution condition, carrying out denoising point processing according to the geometric characteristics of the existing point arrangement and the actual condition of the tunnel, and respectively processing the noise in the tunnel and the noise outside the tunnel; seam filling, face constructing and ground adding. The method and the device only rely on the point cloud data to generate the visual model, can divide the point cloud into a plurality of parts for processing, can meet the processing of a large amount of point cloud data, and have higher efficiency.

Description

A method for denoising tunnel point cloud and generating visualization model

technical field

The invention relates to a method for denoising a tunnel point cloud and generating a visualization model, and belongs to the technical field of graphics and visualization.

Background technique

As a traffic project that runs through mountains and rivers, tunnels are an important part of urban traffic systems. However, factors such as the narrow space inside the tunnel and poor lighting conditions have caused certain difficulties for tunnel measurement. As a new type of surveying and mapping technology, 3D point cloud laser scanning technology has the advantages of no need for illumination, non-contact measurement, high degree of automation, comprehensive information collection and short operation cycle, and is widely used in tunnel surveying.

At present, most of the point cloud information is directly visualized based on laser scanning, and the effect is not ideal; the existing tunnel triangular mesh model visualization algorithm can only support the grid generation of a small amount of data.

In the process of visualization, the most important step is to remove noise points such as vehicles, pedestrians, and light tubes in the collected data. Existing methods mostly rely on the design of the tunnel; for old tunnels with missing drawings, the methods used are complex and inefficient. How to provide a simple and efficient visualization model based only on point cloud data has become an urgent technical problem to be solved.

Contents of the invention

Aiming at the deficiencies of the prior art, the present invention provides a method for denoising tunnel point clouds and generating a visualization model, which only relies on point cloud data to generate a visualization model, and can divide the point cloud into multiple parts for processing, which can satisfy a large number of points Cloud data processing, high efficiency.

Explanation of terms:

1. Bounding box: in the present invention, it refers to a virtual cuboid that can completely surround a tunnel.

The present invention adopts following technical scheme:

A method for denoising a tunnel point cloud and generating a visualization model, comprising:

Step 1. Collect the 3D point cloud of the tunnel, and preprocess the tunnel point cloud data;

First, use a 3D laser scanner to collect 3D point clouds of the tunnel. During the collection process, the y-axis of the coordinate system is defined as the direction of the tunnel, the z-axis is perpendicular to the ground, and the x-axis obeys the right-hand rule;

Then preprocess the exported point cloud data, and export the coordinate information of the point cloud when exporting the point cloud data;

The preprocessing process includes:

1.1. Generate point cloud slices: split the point cloud data according to the direction of the tunnel, and split them into point cloud slices with a length of 3 to 5 meters;

1.2. Perform thinning processing for each point cloud slice, and multiple point cloud slices are processed separately in parallel:

Use the existing grid center of gravity thinning method to thin each point cloud slice. The selection of the grid side length depends on the accuracy requirements of the final grid. The smaller the grid side length, the more points are retained. , the smoother the generated mesh model is;

1.3. Determine the slice bounding box: take the maximum value of the slice point cloud coordinates to determine a rectangular bounding box;

Step 2. Denoising points:

Due to the complexity of the tunnel, the interference objects are included when scanning with a 3D laser scanner. According to the position of the noise, it is divided into noise points inside the tunnel and noise points outside the tunnel. Group points and appendage noise points caused by facilities in the tunnel are denoised according to the geometric characteristics of the existing point arrangement and the actual situation of the tunnel, and the noise inside the tunnel and the noise outside the tunnel are processed separately.

Preferably, the process of removing noise points outside the tunnel is:

For each point cloud slice, the range of its bounding box in the x-axis direction is known, and in this direction, it is divided into point cloud columns by decimeter level (the point cloud column is preferably 0.5 meters), and the number of midpoints in each column is counted, which will be compared Exclude and delete less columns (columns with less than 50 points can be considered as fewer columns);

Do the same processing in the z-axis direction, that is, divide the z-direction into point cloud columns (the point cloud column width is preferably 0.5m), count the number of midpoints in each column, and delete less columns (fewer points) in the z-axis direction. Columns greater than 50 can be considered fewer columns);

Preferably, the process of removing noise points in the tunnel includes the following steps:

a. Split each slice again, and split it into several rows and columns on the xoz plane to form a tunnel block, that is, a point cloud block;

Preferably, the width of the row and column is about 0.5m, and it is split into blocks of about 0.5m*0.5m, which should not be too large, so as to avoid insufficient fitting degree of the fitting plane obtained subsequently;

b. Use Principal Component Analysis (PCA) to transform the coordinates of each tunnel block to find the fitting plane:

The original three-dimensional data in the point cloud block obtained by step a is processed into a matrix X of 3 rows of m columns, wherein m represents the number of points in the point cloud block;

Each row of the matrix X (representing the data on the same axis) is zero-meanized, that is, the mean value of this row is subtracted;

其中，C为求得的协方差矩阵，m为点云块中的点数；find the covariance matrix

Among them, C is the obtained covariance matrix, and m is the number of points in the point cloud block;

Find the eigenvalues and corresponding eigenvectors of the covariance matrix;

Arrange the eigenvalue vectors into a matrix from top to bottom according to the corresponding eigenvalue size, and take it as a matrix

Y=PX is the data after coordinate conversion;

Calculate the average value of each row of data in Y to obtain a coordinate M;

The plane determined by the eigenvectors x ₁ , x ₂ and the coordinate M is the fitting plane of the block, and the direction of x ₃ is collinear with the normal vector direction of the fitting plane;

Define the threshold in the _x3 direction, the selection of the threshold is based on the selection of the row and column width in step a, and can be selected to the centimeter level (preferably 0.05m); this step can effectively remove the noise points of the appendages that are relatively close to the tunnel wall.

c. The geometric shape of the tunnel is close to a semi-elliptical shape. G and F represent the points in two adjacent tunnel blocks containing tunnel walls, and the angles α and β are respectively the fitting planes of the tunnel blocks where G and F are located in the xoz plane The angle between the projection above and the x-axis, when the distance between G and F is very small, the difference between the angles α and β is very small, so the approximate value of α can be determined by β, so as to obtain the approximate fitting plane of the tunnel block where G is located. Fitting the normal vector direction of the plane to define the threshold can effectively delete outliers such as light sticks and vehicles with a certain distance from the tunnel wall;

Do the following for each tunnel block:

Get a tunnel block and be O, then find the tunnel block that contains the tunnel wall before the tunnel block O (due to the tunnel block that contains the tunnel wall, so the "previous block" may be the tunnel block on the left side of O, It may also be the tunnel block located on the lower side of O), that is, O ₁ . According to the matrix P ₀ formed by the eigenvectors of O ₁ , calculate the matrix Y ₀ of the tunnel block O ₁ after coordinate conversion, which is similar to the process of step b , to delineate the threshold of the direction of the line where x ₃ is located.

d. Because the ground of many tunnels contains sidewalks, there is a height difference between the ground and the tunnel. The algorithm in step c is only suitable for the data of more than three layers from the ground; for the part close to the ground, the data formed based on the data of the third layer Fit the plane and determine the slope range of the data for this part, so that the tunnel wall is completely separated from the ground.

Preferably, reshape the data after denoising:

Divide the tunnel into left and right sides on average according to the size of the bounding box in the x-axis direction, and then divide the two sides into layers in the z-axis direction (preferably: keep the z-axis direction unchanged, or divide according to 0.5m wide layers), divided into layer data.

Step 3. Fill seams and surfaces:

3.1. Seam filling: Point clouds of adjacent slices or adjacent layers, 5% of the points drawn at the boundary are repeated point sets;

Due to the subsequent need to triangulate the points by layer, there are gaps between segments and layers, and the processing method is to draw 5% of the point cloud of adjacent slices or adjacent layers at the boundary The point is a repeated point set, and the repeated point set exists in two adjacent parts, so as to ensure that the faces are connected when faceting;

3.2. Complement surface: make statistics on the upper and lower 10% points of the point cloud of each layer (this value can be set according to the specific situation, and it should not be too small to cause errors). When the number of points in this part is less than 5% of the layer, determine the upper and lower parts of the layer. Whether there is a missing part in the lower part, search the upper and lower layers of the layer until the non-missing part is found, and add repeated point sets in the non-missing layer, so that the entire tunnel surface remains complete and continuous after the next step of faceting;

Due to the lack of point clouds caused by large-area appendage noise such as light sticks, large-area deletions will appear after tunnel faceting. Therefore, the present invention complements these missing places.

Step 4. Facet, including:

4.1. Use principal component analysis (PCA) to reduce the dimensionality of the three-dimensional point coordinates in each layer of data into two-dimensional data;

Further: Since the 3D faceting algorithm is more complex than the 2D facet algorithm, based on the consideration of modeling efficiency, the 3D points of each layer are reduced to 2D points according to the fitting plane, similar to step b:

First, the three-dimensional points of each layer are formed into a matrix X ₁ with 3 rows and n columns by columns, and each row of the matrix X ₁ is zero-meanized, that is, the mean value of this row is subtracted;

其中C₁表示求得的协方差矩阵，n为该层中点的数量；find the covariance matrix

Where C ₁ represents the obtained covariance matrix, and n is the number of points in the layer;

Find the eigenvalues and corresponding eigenvectors of the covariance matrix;

Arrange the eigenvalue vectors into a matrix from top to bottom according to the corresponding eigenvalue size, and take it as a matrix

Y ₁ = P ₁ X ₁ is the data after coordinate transformation;

Calculate the matrix Y ₁ after the coordinate conversion of each layer of data, and the first two rows of data in Y ₁ are the coordinate set Y ₁ ' after dimensionality reduction;

4.2. Perform Delaunay triangulation on the two-dimensional data Y ₁ ' obtained from each layer of data, correspond the two-dimensional points to the three-dimensional coordinates before dimensionality reduction, generate a triangular mesh model, and then integrate the mesh models of different layers and slices to combine.

In this step, the Delaunay triangulation process can adopt the existing technology, which will not be repeated here.

Step 5. Add the ground:

Since the ground information is not important, all the original information about the ground is discarded, and quadrilaterals are used to construct the ground of each slice.

The present invention is not detailed, all can adopt prior art to carry out.

The beneficial effects of the present invention are:

1. The invention is only based on the point cloud data production model, and does not need to know other information except the point cloud information as auxiliary reference, such as the original design of the tunnel, etc., which is simple, convenient and has strong applicability.

2. The present invention performs fitting by blocks, which has a high degree of fitting and a good noise removal effect, and also has a good effect on complex tunnels.

3. The denoising points and facets of the present invention are performed on a small independent point set, which has high time efficiency and a large degree of parallelism.

4. Multiple point cloud slices of the present invention are processed separately in a parallel manner, which can take into account details and efficiency, and has a better processing effect on point cloud data with a large amount of data.

Description of drawings

Figure 1 is a schematic diagram of the cross-sectional structure of the tunnel;

Figure 2 is a slice of the original tunnel point cloud;

Figure 3 is a point cloud slice after removing noise;

Fig. 4 is the effect diagram when the seam is not repaired;

Fig. 5 is the rendering after completion;

Detailed ways:

In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments, but not limited thereto, and those not described in detail in the present invention shall be conventional techniques in this field.

A method for denoising a tunnel point cloud and generating a visualization model, comprising:

Step 1. Acquire the 3D point cloud of the tunnel, as shown in Figure 2, which is a slice of the original tunnel point cloud, and preprocess the tunnel point cloud data;

The three-dimensional laser scanner used in the present invention is a TrimbleSX10 image scanner, and the illustrated tunnel is the Kaiyuan Tunnel in Jinan. First, the three-dimensional laser scanner is used to collect the three-dimensional point cloud of the tunnel. During the collection process, the y-axis of the coordinate system is defined as the direction of the tunnel. , the z-axis is perpendicular to the ground, and the x-axis obeys the right-hand rule, as shown in Figure 1;

Then preprocess the exported point cloud data, and export the coordinate information of the point cloud when exporting the point cloud data;

The preprocessing process includes:

1.1. Generate point cloud slices: split the point cloud data according to the direction of the tunnel, and split them into 3-meter-long point cloud slices. During the splitting process, record the maximum value of the coordinate axis of each slice as the boundary of the bounding box;

1.2. Perform thinning processing for each point cloud slice, and multiple point cloud slices are processed separately in parallel:

Use the existing grid center of gravity thinning method to thin each point cloud slice, and the grid side length is 0.02m;

1.3. Determine the slice bounding box: take the maximum value of the slice point cloud coordinates to determine a rectangular bounding box;

Step 2. Denoising points:

Due to the complexity of the tunnel, the interference objects are included when scanning with a 3D laser scanner. According to the position of the noise, it is divided into noise points inside the tunnel and noise points outside the tunnel. Group points and appendage noise points caused by facilities in the tunnel are denoised according to the geometric characteristics of the existing point arrangement and the actual situation of the tunnel, and the noise inside the tunnel and the noise outside the tunnel are processed separately.

Preferably, the process of removing noise points outside the tunnel is:

For each point cloud slice, the range of its bounding box in the x-axis direction is known. In this direction, that is, the x-axis direction is divided into columns with a width of 0.5m. The number of midpoints in each column is counted, and the columns with less than 50 points are counted. Delete, thus reducing the range of data in the x-axis direction;

Do the same in the z-axis direction to remove the noise outside the tunnel, that is, divide it into point cloud columns by 0.5m in the z-direction, count the number of points in each column, and delete the columns with less than 50 points;

Preferably, the process of removing noise points in the tunnel includes the following steps:

a. Split each slice again, split it into several rows and columns on the xoz plane, and split it into 0.5m*0.5m blocks, thus forming tunnel blocks, that is, point cloud blocks;

b. Use Principal Component Analysis (PCA) to transform the coordinates of each tunnel block to find the fitting plane:

The original three-dimensional data in the point cloud block obtained by step a is processed into a matrix X of 3 rows of m columns, wherein m represents the number of points in the point cloud block;

Each row of the matrix X (representing the data on the same axis) is zero-meanized, that is, the mean value of this row is subtracted;

其中，C为求得的协方差矩阵，m为点云块中的点数；find the covariance matrix

Among them, C is the obtained covariance matrix, and m is the number of points in the point cloud block;

Find the eigenvalues and corresponding eigenvectors of the covariance matrix;

Arrange the eigenvalue vectors into a matrix from top to bottom according to the corresponding eigenvalue size, and take it as a matrix

Y=PX is the data after coordinate conversion;

Calculate the average value of each row of data in Y to obtain a coordinate M;

The plane determined by the eigenvectors x ₁ , x ₂ and the coordinate M is the fitting plane of the block, and the direction of x ₃ is collinear with the normal vector direction of the fitting plane;

Delineate the threshold in the x ₃ direction. The selection of the threshold is based on the selection of the row and column width in step a, which is 0.05m. This step can effectively remove the noise points of the appendages that are relatively close to the tunnel wall.

c. As shown in Figure 1, the geometric shape of the tunnel is close to a semi-elliptical shape. G and F represent points in two adjacent tunnel blocks containing tunnel walls. The two dotted lines in the figure represent the tunnels where G and F are respectively located. The projection of the fitting plane of the block on the xoz plane can be seen from the figure, when the distance between G and F is very small, the difference between the angles α and β is very small, so the approximate value of α can be determined by β, so as to obtain the approximate value of the tunnel block where G is located Fitting plane, by defining a threshold in the direction of the normal vector of the fitting plane, outliers such as light sticks and vehicles that are a certain distance away from the tunnel wall can be effectively deleted;

Do the following for each tunnel block:

Get a tunnel block and be O, then find the tunnel block that contains the tunnel wall before the tunnel block O (due to the tunnel block that contains the tunnel wall, so the "previous block" may be the tunnel block on the left side of O, It may also be the tunnel block located on the lower side of O), that is, O ₁ . According to the matrix P ₀ formed by the eigenvectors of O ₁ , calculate the matrix Y ₀ of the tunnel block O ₁ after coordinate conversion, which is similar to the process of step b , to delineate the threshold of the straight line direction where x ₃ is located, that is, the threshold of the normal vector, which is 0.1m. When the tunnel wall is missing, that is, the block where G is located is missing as shown in Figure 1, then the threshold is accumulated by 0.05 meters , until the block where G is updated to a non-missing place, restore the threshold to 0.1m.

d. Because the ground of many tunnels contains sidewalks, there is a height difference between the ground and the tunnel. The algorithm in step c is only suitable for the data of more than three layers from the ground; for the part close to the ground, the data formed based on the data of the third layer Fit the plane and determine the slope range of the data for this part, so that the tunnel wall is completely separated from the ground.

Preferably, reshape the data after denoising:

According to the size of the bounding box in the x-axis direction, the tunnel is evenly divided into left and right sides, and the z-axis direction remains unchanged. It is still divided into layers with a width of 0.5m, and divided into layer data.

Figure 3. Point cloud slice after noise removal;

Step 3. Fill seams and surfaces:

Figure 4 is the effect diagram when the seams are not patched and the surface is patched, and the seams and surfaces need to be patched, specifically

3.1. Seam filling: Point clouds of adjacent slices or adjacent layers, 5% of the points drawn at the boundary are repeated point sets;

Due to the subsequent need to triangulate the points by layer, there are gaps between segments and layers, and the processing method is to draw 5% of the point cloud of adjacent slices or adjacent layers at the boundary The point is a repeated point set, and the repeated point set exists in two adjacent parts, so as to ensure that the faces are connected when faceting;

3.2. Complement surface: make statistics on the upper and lower 10% points of the point cloud of each layer (this value can be set according to the specific situation, and it should not be too small to cause errors). When the number of points in this part is less than 5% of the layer, determine the upper and lower parts of the layer. Whether there is a missing part in the lower part, search the upper and lower layers of the layer until the non-missing part is found, and add repeated point sets in the non-missing layer, so that the entire tunnel surface remains complete and continuous after the next step of faceting;

Due to the lack of point clouds caused by large-area appendage noise such as light sticks, large-area deletions will appear after tunnel faceting. Therefore, the present invention complements these missing places.

Step 4. Facet, including:

4.1. Use principal component analysis (PCA) to reduce the dimensionality of the three-dimensional point coordinates in each layer of data into two-dimensional data;

Further: Since the 3D faceting algorithm is more complex than the 2D facet algorithm, based on the consideration of modeling efficiency, the 3D points of each layer are reduced to 2D points according to the fitting plane, similar to step b:

First, the three-dimensional points of each layer are formed into a matrix X ₁ with 3 rows and n columns by columns, and each row of the matrix X ₁ is zero-meanized, that is, the mean value of this row is subtracted;

其中C₁表示求得的协方差矩阵，n为该层中点的数量；find the covariance matrix

Where C ₁ represents the obtained covariance matrix, and n is the number of points in the layer;

Find the eigenvalues and corresponding eigenvectors of the covariance matrix;

Arrange the eigenvalue vectors into a matrix from top to bottom according to the corresponding eigenvalue size, and take it as a matrix

Y ₁ = P ₁ X ₁ is the data after coordinate transformation;

Calculate the matrix Y ₁ after the coordinate conversion of each layer of data, and the first two rows of data in Y ₁ are the coordinate set Y ₁ ' after dimensionality reduction;

4.2. Perform Delaunay triangulation on the two-dimensional data Y ₁ ' obtained from each layer of data, correspond the two-dimensional points to the three-dimensional coordinates before dimensionality reduction, generate a triangular mesh model, and then integrate the mesh models of different layers and slices to combine.

Step 5. Add the ground:

Since the ground information is not important, all the original information about the ground is discarded. For the ground, quadrilaterals are used to construct the ground of each slice.

Fig. 5 is an effect diagram after the processing of the present invention is completed.

The above description is a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, and these improvements and modifications can also be made. It should be regarded as the protection scope of the present invention.

Claims (7)

1. A method for denoising and generating a visualization model for a tunnel point cloud, comprising:

step 1, collecting a tunnel three-dimensional point cloud, and preprocessing tunnel point cloud data;

firstly, a three-dimensional laser scanner is used for collecting three-dimensional point clouds of a tunnel, a y axis of a coordinate system is defined as a direction of trend of the tunnel in the collecting process, a z axis is perpendicular to the ground, and an x axis obeys a right rule;

preprocessing the derived point cloud data, and deriving coordinate information of the point cloud when the point cloud data is derived;

step 2, denoising:

because the tunnel condition is complex, the three-dimensional laser scanner is adopted to scan the tunnel and comprises interference objects, the noise points are divided into noise points in the tunnel and noise points outside the tunnel according to the noise positions, the noise points in the tunnel are divided into outlier points caused by moving targets and accessory noise points caused by facilities in the tunnel according to the space distribution condition, the noise point removing treatment is carried out according to the geometric characteristics of the existing point arrangement and the actual condition of the tunnel, and the noise in the tunnel and the noise outside the tunnel are respectively treated;

step 3, seam and surface repairing;

step 4, constructing a surface;

step 5, adding ground:

constructing the ground of each slice by using a quadrilateral;

in step 1, the pretreatment process comprises the following steps:

1.1, generating point cloud slices: splitting the point cloud data according to the trend of the tunnel into point cloud slices with the length of 3-5 meters;

1.2, performing thinning treatment on each point cloud slice, wherein a plurality of point cloud slices are independently treated in a parallel mode during treatment:

the existing rasterization gravity center thinning method is adopted to carry out thinning treatment on each point cloud slice, the selection of the grid side length depends on the precision requirement on the final grid, the smaller the grid side length is, the more points remain, and the smoother the generated grid model is;

1.3, determining a slice bounding box: and taking the maximum value of the slice point cloud coordinates to determine a rectangular bounding box.

2. The method for denoising and generating a visualized model according to claim 1, wherein in step 2, the process of removing the noise points outside the tunnel is:

for each point cloud slice, dividing the range of the bounding box in the x-axis direction into point cloud columns according to the decimeter level in the direction, counting the number of points in each column, and deleting the columns with the number of points less than 50, wherein the number of points in each column is 0.5 meter;

the same process is performed in the z-axis direction, namely, the columns are divided into point cloud columns in the z-axis according to decimeter level, the width of the point cloud columns is 0.5m, the number of points in each column is counted, and the columns with the number of points less than 50 are deleted.

3. The method for denoising tunnel point cloud and generating visual model according to claim 2, wherein the process of removing noise points in the tunnel comprises the steps of:

a. splitting each slice again, and splitting the slices into a plurality of rows and columns again on a xoz plane, so as to form a tunnel block, namely a point cloud block;

b. and carrying out coordinate conversion on each tunnel block by adopting a principal component analysis method, and finding out a fitting plane:

c, forming a matrix X of 3 rows and m columns of the original three-dimensional data in the point cloud block obtained by processing in the step a according to the columns, wherein m represents the points in the point cloud block;

zero-equalizing each row of the matrix X, namely subtracting the average value of the row;

solving covariance matrix

Wherein C is the covariance matrix obtained, m is the number of points in the point cloud block;

obtaining eigenvalues and corresponding eigenvectors of the covariance matrix;

the eigenvalue vectors are arranged into a matrix according to the corresponding eigenvalue size from top to bottom and are taken as the matrix

Y=px is the coordinate-converted data;

averaging the data of each row of Y to obtain a coordinate M;

feature vector x ₁ ,x ₂ The plane determined by the coordinates M is the fitting plane of the block, x ₃ Is collinear with the normal vector direction of the fitting plane;

demarcating x ₃ A threshold value of the direction, wherein the threshold value is selected according to the row and column width selection in the step a, and is selected to be 0.05m in centimeter level;

c. the geometric shape of the tunnel is close to a semi-ellipse shape, G and F represent points in two adjacent tunnel blocks containing tunnel walls, angles alpha and beta are respectively included angles between projection of a fitting plane of the tunnel block where G, F is located on a xoz plane and an x axis, when the distance between G and F is small, the difference between the angles alpha and beta is small, so that an approximate value of alpha is determined through beta, an approximate fitting plane of the tunnel block where G is located is obtained, a threshold value is defined in the normal vector direction of the fitting plane, a lamp rod is effectively deleted, and an outlier with a certain distance from the tunnel wall such as a vehicle is removed;

the following operations are performed for each tunnel block:

taking a tunnel block as O, and finding the tunnel block containing tunnel wall before the tunnel block O, namely O ₁ According to O ₁ Matrix P of eigenvectors of (a) ₀ Calculate the tunnel block O ₁ Matrix Y after coordinate conversion ₀ Demarcating x, similar to the procedure of step b ₃ A threshold value in the straight line direction;

d. because many tunnel floors comprise sidewalks, the floors have height differences with tunnels, and the algorithm in the step c is only suitable for data with more than three layers from the floors; for a portion near the ground, a data slope range of the portion is determined based on a fitting plane formed by data of the third layer, thereby completely separating the tunnel wall from the ground.

4. The method for denoising and generating a visualized model according to claim 3, wherein in step a, the widths of the rows and columns are about 0.5m, and the rows and columns are split into blocks of 0.5m by 0.5 m.

5. The method of denoising and generating a visualization model of a tunnel point cloud of claim 4, wherein the denoising is followed by reforming data:

dividing the tunnel into left and right sides according to the size of the bounding box in the x-axis direction, and dividing the z-axis direction into layers by the two sides: the z-axis direction is unchanged, or the layer data is divided according to the layer with the width of 0.5 m.

6. The method of denoising and generating a visualization model of tunnel point cloud according to claim 5, wherein step 3 is further:

3.1, seam filling: the point clouds of adjacent slices or adjacent layers mark 5% points at the boundaries as a repeated point set;

since the subsequent triangularization is carried out according to the layer-to-layer points, gaps are formed on the surfaces among the sections and the layers, and the processing method is that the point clouds of the adjacent slices or the adjacent layers are marked with 5% points at the boundaries as repeated point sets, and the two adjacent parts have the repeated point sets, so that the surfaces are connected when the surfaces are constructed;

3.2, face supplementing: and counting 10% of points on the upper and lower sides of the point cloud of each layer, determining whether the upper part and the lower part of the layer are missing when the number of the points of the part is less than 5% of the layer, searching the upper and lower layers of the layer until the non-missing part is searched, and adding a repeated point set in the non-missing layer, so that the whole tunnel curved surface is kept complete and continuous after the next step of surface construction.

7. The method of denoising and generating a visualization model of tunnel point cloud according to claim 6, wherein step 4 is further:

4.1, dimension reduction is carried out on three-dimensional point coordinates in each layer of data into two-dimensional data by using a principal component analysis method:

firstly, three-dimensional points of each layer are formed into a matrix X of 3 rows and n columns according to columns ₁ Matrix X ₁ Zero-equalizing, i.e. subtracting the average value of the line;

solving covariance matrix

Wherein C is ₁ Representing the obtained covariance matrix, wherein n is the number of points in the layer;

obtaining eigenvalues and corresponding eigenvectors of the covariance matrix;

the eigenvalue vectors are arranged into a matrix according to the corresponding eigenvalue size from top to bottom and are taken as the matrix

Y ₁ ＝P ₁ X ₁ Namely, the data after coordinate conversion;

solving for each layer of data after coordinate conversionMatrix Y ₁ ，Y ₁ The first two lines of data are coordinate set Y after dimension reduction ₁ ’；

4.2 two-dimensional data Y obtained for each layer of data ₁ Performing Delaunay triangular planing, corresponding the two-dimensional points to the three-dimensional coordinates before dimension reduction, generating a triangular grid model, and combining grid models of different slices of different layers.

Images