Method for improving position accuracy of vehicle-mounted laser point cloud assisted by airborne point cloud

Abstract

The present invention discloses a method for improving the position accuracy of vehicle-mounted laser point clouds assisted by airborne point clouds, comprising the following steps: 1) indexing the airborne point clouds and vehicle-mounted point clouds respectively, and dividing the entire space of the airborne point clouds and vehicle-mounted point clouds into regular three-dimensional cubic spaces according to a given spatial resolution; 2) calculating geometric features of all voxels in the airborne point clouds and vehicle-mounted point clouds respectively, and screening out voxels whose distribution features are surface features; 3) taking the airborne point clouds as a reference, taking the surface features of the voxels in the airborne point clouds obtained in step 2) as a reference plane, finding all corresponding plane feature pairs of the airborne point clouds and vehicle-mounted point clouds; 4) taking all corresponding plane feature pairs as registration primitives, calculating the rotation matrix and translation vector from the vehicle-mounted point clouds to the airborne point clouds, and using the rotation matrix and translation vector to optimize the position of the vehicle-mounted point clouds. The present invention corrects the vehicle-mounted laser point clouds to the airborne point cloud coordinate framework, and improves the position accuracy of the vehicle-mounted point clouds.

Description

Technical Field

[0001] The invention relates to laser scanning data processing technology, and in particular to a method for improving the position accuracy of vehicle-mounted laser point clouds assisted by airborne point clouds. Background Art

[0002] The 3D laser scanning system has a fast acquisition speed and high sampling frequency. It can directly obtain a large number of 3D coordinate point clouds with information such as intensity and color in a short time by non-contact measurement. At the same time, 3D laser radar technology is not restricted by weather and time and has penetrating properties. The acquired point cloud is an important 3D geographic information data. The 3D point cloud is very important in earth science research, including global change, smart cities, resource surveys, basic surveying and mapping, etc. It has a wide range of applications. With the development and maturity of 3D point cloud acquisition technology and 3D laser radar scanning systems, the theory and application of point cloud processing have gradually improved, including data quality improvement, multi-source point cloud data fusion, point cloud classification and target extraction, and on-demand multi-level modeling and expression.

[0003] The vehicle-mounted laser scanning system can obtain position and attitude information synchronized with time in real time, so that the acquired laser point cloud is a three-dimensional coordinate with geographic reference information. In urban scenes, the GNSS signal received by the vehicle-mounted laser scanning system may be blocked by buildings, trees and other objects, causing the GNSS signal to be lost or of poor quality, resulting in a deviation between the final solved point cloud and the actual position. In order to ensure the accuracy and robustness of subsequent point cloud processing, the quality of point cloud data with deviations must first be improved. Therefore, point cloud quality improvement is a key step in the point cloud processing process, providing raw input for point cloud classification, target extraction, and vectorized modeling.

[0004] Traditionally, the accuracy of vehicle-mounted point cloud position is improved by repeated observation strip adjustment or by setting control points along the way. However, there are also certain problems. The method of using repeated observation of vehicle-mounted point cloud for strip adjustment is not applicable to areas without round-trip observation; at the same time, the method of manually setting control points is very time-consuming and labor-intensive. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for improving the position accuracy of vehicle-mounted laser point cloud assisted by airborne point cloud in view of the defects in the prior art.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a method for improving the position accuracy of vehicle-mounted laser point cloud assisted by airborne point cloud, comprising the following steps:

[0007] 1) Establish indexes for the airborne point cloud and the vehicle-borne point cloud respectively, and divide the entire space of the airborne point cloud and the vehicle-borne point cloud into regular three-dimensional cube spaces, i.e., voxels, according to a given spatial resolution;

[0008] 2) Calculate the geometric features of all voxels in the airborne point cloud and the vehicle-mounted point cloud respectively, and select the voxels whose distribution features are surface features; calculate the plane parameters for the surface features of the selected voxels;

[0009] 3) Taking the airborne point cloud as a reference, the surface features of the voxels in the airborne point cloud obtained in step 2) are used as reference planes to construct the distance of plane features with the same name, search for the plane feature voxels corresponding to the spatial neighborhood of the vehicle-borne point cloud, and find all corresponding plane feature pairs between the airborne point cloud and the vehicle-borne point cloud;

[0010] The plane feature in each voxel in the airborne point cloud is used as the reference plane, and then for each reference plane feature, all the neighborhoods of its corresponding vehicle-borne point cloud are searched for plane features to be matched that meet the requirements, and the plane feature pairs in the entire point cloud space are obtained;

[0011] 4) All corresponding planar feature pairs of the airborne point cloud and the vehicle-borne point cloud are used as registration primitives, the rotation matrix and translation vector from the vehicle-borne point cloud to the airborne point cloud are calculated, and the position of the vehicle-borne point cloud is optimized using the rotation matrix and translation vector.

[0012] According to the above scheme, in step 1), the airborne point cloud and the vehicle-borne point cloud are indexed based on the octree structure respectively, and the entire space of the airborne point cloud and the vehicle-borne point cloud is divided into a regular three-dimensional cube space.

[0013] According to the above scheme, in step 2), the geometric features of all voxels in the airborne point cloud and the vehicle-mounted point cloud are calculated, and voxels whose distribution features are surface features are screened out, as follows:

[0014] 2.1) Using the principal component analysis method, the geometric features of the point cloud within the voxel are calculated through the eigenvalue decomposition of the covariance matrix to screen out voxels with surface features;

[0015] p i , i = 1, 2, ..., k, is the k points in the voxel, and the covariance matrix M of the point cloud in the voxel is calculated by the following formula:

[0016]

[0017] in,

[0018] 2.2) The covariance matrix M obtains the corresponding eigenvector and three eigenvalues ​​λ1, λ2, and λ3 through the eigenvalue decomposition process, where λ1≥λ2≥λ3>0; the three eigenvalues ​​together describe the distribution of the point cloud in the voxel, and the geometric characteristics of the point cloud are expressed by defining the operation between the three features, which are a 1D 、a 2D 、a 3D , which represent line features, surface features, and volume features respectively. The geometric features are calculated using the following formula:

[0019]

[0020] when a 2D When the value is the largest, that is, when the following formula is satisfied, it indicates that the distribution feature of the point cloud in the current voxel is a surface feature;

[0021]

[0022] 2.3) According to the above steps, the geometric features of all voxels in the airborne point cloud and the vehicle-mounted point cloud are calculated respectively, and the voxels with significant planar features are screened out.

[0023] According to the above scheme, before calculating the plane parameters in step 2), the plane features of the screened voxels are refined as follows:

[0024] The random sampling consistency method is adopted to refine the planar features of the screened voxels;

[0025] By randomly selecting points in the point cloud space, the point cloud is divided into plane interior points and exterior points, the plane interior points within the voxels are retained, and the plane parameters are calculated;

[0026] In 3D Euclidean space, the plane equation is expressed as follows:

[0027] ax+by+cz=d

[0028] For a three-dimensional plane, the plane normal vector n is expressed as n = (a, b, c). If the normal vector n is a normalized vector, then n·n = 1. The d in the plane equation is the distance from the plane to the origin of the coordinate system. For each plane feature voxel after refinement, the plane point set {(x i ,y i ,z i ),1≤i≤k}, calculate its plane parameters.

[0029] According to the above scheme, the step 3) finds all corresponding plane feature pairs of the airborne point cloud and the vehicle-borne point cloud, as follows:

[0030] For any reference plane in the airborne point cloud, the neighborhood voxels of its corresponding vehicle-borne point cloud can be obtained, and the reference plane features P can be calculated respectively. i To all neighboring voxels, the plane features P to be registered j The plane characteristic distance Δd of {0≤j≤26} ij , the plane features with the smallest distance form a plane feature pair;

[0031]

[0032] Among them, n i and n j They are the reference plane features P i and the plane feature P of the neighboring voxels to be registered j The normal vector of i and x j They are the plane feature point sets composed of the refined voxels {(x k ,y k ,z k ),1≤k≤n}, where

[0033] According to the above scheme, the rotation matrix and translation vector are calculated in step 4) by using the angle relationship between the normal vectors of the same-name plane feature pairs and the difference in the origin distance between the same-name plane feature pairs to respectively calculate the rotation matrix and the translation vector.

[0034] The beneficial effects produced by the present invention are:

[0035] 1. Utilize the higher-precision airborne point cloud and overcome the difficulty of density difference between airborne point cloud and vehicle-borne point cloud, correct the vehicle-borne laser point cloud to the airborne point cloud coordinate framework, and improve the position accuracy of the vehicle-borne point cloud.

[0036] 2. In order to solve the problem of deviation in the position accuracy of vehicle-mounted laser point clouds, the plane features in non-homologous point clouds are used as registration primitives to complete the registration of vehicle-mounted point clouds to the reference airborne point clouds, thereby improving the position accuracy of vehicle-mounted point clouds. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0038] Figure 1 is a method flow chart of an embodiment of the present invention;

[0039] Figure 2 Schematic diagram of plane feature registration according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0041] like Figure 1 As shown, a method for improving the position accuracy of vehicle-mounted laser point cloud assisted by airborne point cloud comprises the following steps:

[0042] Step 1: Use the octree structure to index the airborne point cloud and the vehicle-mounted point cloud respectively, and divide the regular voxels according to the given spatial resolution, that is, according to the given spatial resolution r, divide the entire space of the airborne point cloud and the vehicle-mounted point cloud into regular 3D cube space, namely voxels;

[0043] This step is a voxelization process based on the octree. Given a spatial resolution r, which is 5 m in the embodiment, the entire space of the airborne point cloud and the vehicle-borne point cloud is divided into a regular 3D cube space.

[0044] Step 2: For the voxels divided by the vehicle-mounted point cloud and the airborne point cloud, the principal component analysis method is used for the points in the voxels to determine the geometric features of the voxels, obtain the voxels with significant plane features, and then refine the plane features of the voxels with significant plane features, eliminate the non-planar points in the voxels, and calculate the plane feature parameters.

[0045] In the embodiment, step 2 is implemented as follows:

[0046] 2.1) Using the principal component analysis method, the geometric features of the point cloud within the voxel are calculated through the eigenvalue decomposition of the covariance matrix to screen out voxels with surface features;

[0047] p i (i=1,2,…,k) is the k points in the voxel. The covariance matrix M of the point cloud in the voxel can be calculated by formula (1):

[0048]

[0049] in,

[0050] The covariance matrix M can be decomposed into corresponding eigenvectors and three eigenvalues ​​λ1, λ2, and λ3 (λ1≥λ2≥λ3>0) through the eigenvalue decomposition process. The three eigenvalues ​​together describe the distribution of the point cloud in the voxel. The geometric characteristics of the point cloud are expressed by defining the operation between the three features, which are a 1D 、a 2D 、a 3D , which represent line features, surface features, and volume features respectively, which can be calculated using formula (2):

[0051]

[0052] when a 2D When the value is the largest, that is, when formula (3) is satisfied, it indicates that the distribution feature of the point cloud in the current voxel is a surface feature.

[0053]

[0054] According to the above technical steps, the geometric features of all voxels in the airborne point cloud and the vehicle-mounted point cloud are calculated respectively, and the voxels with significant planar features are screened out.

[0055] 2.2) In order to make the extracted plane features more robust and the calculated plane parameters more accurate, this technical solution adopts a random sampling consistency method to refine the roughly extracted plane features and then calculate the plane parameters.

[0056] Random sampling consistency is a method of extracting specific components when there is a lot of noise interference in the data. This embodiment uses a method of randomly selecting points in the point cloud space to divide the point cloud into plane inner points and outer points. The plane parameters are calculated for the plane inner points retained in the voxel.

[0057] In 3D Euclidean space, the plane equation is expressed by formula (4):

[0058] ax+by+cz=d (4)

[0059] Among them, a, b, c, d are plane parameters;

[0060] For a three-dimensional plane, the plane normal vector n can be expressed as n = (a, b, c). If the normal vector n is a normalized vector, then n·n = 1. The d in the plane equation is the distance from the plane to the origin of the coordinate system. For each plane point set {(x i ,y i ,z i ),1≤i≤k}, its plane parameters can be calculated. In the embodiment, for any point p in the point set i =(x i ,y i ,z i ), and its distance to the plane is calculated by formula (5):

[0061] d i =|ax i +by i +cz i -d| (5)

[0062] For the plane that best fits the point set, the sum of the distances from all points to the plane Should be minimum. Using the orthogonal regression method, the normalized normal vector n of the plane can be obtained from the eigenvector corresponding to the minimum eigenvalue of matrix A. Matrix A can be obtained by formula (6):

[0063]

[0064] in,

[0065] After the normal vector n is obtained, that is, the first three parameters a, b, c of the plane equation are obtained, the fourth parameter d of the plane equation can be calculated, which can be obtained by formula (7):

[0066]

[0067] Step 3: Using the airborne point cloud as a reference and the plane features obtained in step 2 as a reference, construct the distance of plane features with the same name, and search for plane feature voxels corresponding to the spatial neighborhood of the vehicle-borne point cloud. Find all corresponding plane feature pairs between the airborne point cloud and the vehicle-borne point cloud.

[0068] In the embodiment, the plane feature voxel in the airborne point cloud is used as a reference, and the search range of the plane feature with the same name corresponding to the vehicle-borne point cloud is composed of its corresponding 26 adjacent neighborhood voxels.

[0069] In order to correct the vehicle-mounted point cloud to the airborne point cloud benchmark, the plane features in each voxel in the airborne point cloud are used as reference planes, and then for each reference plane feature, all other plane features to be matched that meet the requirements in the neighborhood are searched to obtain plane feature pairs in the entire point cloud space.

[0070] For any reference plane in the airborne point cloud, the neighborhood voxels of its corresponding vehicle-borne point cloud can be obtained, and the reference plane features P can be calculated using formula (7). i To all neighboring voxels, the plane features P to be registered j The plane characteristic distance Δd of {0≤j≤26} ij ,like Figure 2 shown.

[0071]

[0072] Among them, n i and n j They are the reference plane features P i and the plane feature P of the neighboring voxels to be registered j The normal vector of i and x j They are the plane feature point sets composed of the refined voxels {(x k ,y k ,z k ),1≤k≤n}. Among them,

[0073] Step 4: All corresponding plane feature pairs of the airborne point cloud and the vehicle-mounted point cloud obtained in step 3 are used as registration primitives, and the normal vector angle relationship between the plane feature pairs with the same name and the difference in distance from the origin between the plane feature pairs with the same name are used to calculate the rotation matrix R and the translation vector T respectively. This technical solution divides the rotation matrix R and the translation vector T into two steps, and uses the least squares method to solve the parameters of the plane feature point pairs with the same name. For solving the rotation matrix, an antisymmetric matrix is ​​introduced to solve the rotation matrix R.

[0074] In the embodiment, step 4 is as follows:

[0075] 1) The rotation matrix is ​​used to describe the rotation transformation relationship between two point cloud point sets. The registration primitives extracted by this method are multiple plane feature pairs. The rotation matrix R can be solved by using the spatial rotation relationship between the normal vectors between the plane feature pairs.

[0076] This embodiment also provides a calculation method: using three independent parameters α, β, γ to replace the above three rotation parameters, and introducing an antisymmetric matrix, combined with the properties of the Rodriguez matrix, to solve the 3×3 rotation matrix R. Introducing the antisymmetric matrix, assuming that the matrix S is composed of three independent parameters, as shown in formula (8)

[0077]

[0078] In this technical solution, the rotation matrix between the vehicle-mounted point cloud and the airborne point cloud can be expressed by the matrix R, as shown in formula (9).

[0079]

[0080] Where Δ=1+α 2 +β 2 +γ 2 . At the same time, the rotation matrix R is the Rodriguez matrix.

[0081] Therefore, by finding the corresponding direction unit vectors in the two point clouds, the rotation transformation matrix between the point clouds can be solved. Plane features are extracted based on voxels, and multiple plane feature pairs are obtained by constructing plane feature distances based on the voxel neighborhood relationship.

[0082] For any plane feature pair, the normal vectors (a1, b1, c1) and (a2, b2, c2) are the corresponding direction vectors that meet the conditions. Therefore, we can make v = (a1, b1, c1) T , w=(a2,b2,c2) T , and satisfies formula (10).

[0083] vw=S(v+w) (10)

[0084] Let l = vw = (l z ,l y ,l z ) T , b=v+w=(b x ,b y ,b z ) T , the three independent parameters α, β, and γ can be expressed in the form of formula (11):

[0085]

[0086] Therefore, solving the R matrix can be directly transformed into solving the three parameters α, β, and γ. According to the n (n ≥ 2) plane feature pairs extracted in step 3, the least squares method can be used to solve it, and the matrix formula (12) can be listed.

[0087]

[0088] Among them, the B matrix and the l matrix are coefficient matrices calculated by multiple sets of direction vector pairs (v, w) according to formula 10. x is the matrix (α, β, γ) composed of three independent parameters α, β, and γ T . x is solved according to formula (13)

[0089] x=(B T WB) -1 B T Wl (13)

[0090] Among them, the matrix W is the diagonal weight matrix, according to the sum of the plane fitting residuals Δd in each plane feature pair i The elements w(i,i) on the diagonal of the matrix W can be obtained by formula (14):

[0091]

[0092] λ is the factor of the Gaussian kernel. Finally, the rotation matrix R is obtained.

[0093] 2) After solving the rotation matrix R, for any plane feature pair, the reference plane feature has parameters (a1, b1, c1, d1), and the plane feature to be registered has parameters (a2, b2, c2, d2). The plane to be registered is rotated according to the rotation matrix, and the fourth parameter d2 of the plane feature to be registered is recalculated to obtain new registered plane parameters (a2′, b2′, c2′, d2′). After the plane to be registered is rotated, it is considered that the normal vector of the corresponding reference plane is consistent, and the translation vector T can be obtained by simply solving the distance between the two planes.

[0094] The solution of the translation vector T requires at least three plane feature pairs, and the difference between the plane feature pairs is the translation. Assume that the translation between the plane feature to be registered and the reference plane feature is x0, y0, z0, and the normal vectors of the plane feature to be registered and the reference plane feature after rotation transformation are both (a, b, c);

[0095] ax0+by0+cz0=d1-d2 (15)

[0096] Similarly, for n (n ≥ 3) plane feature pairs, the least squares method can be used to solve the translation vector, and formula (16) is listed:

[0097] A·T=l (16)

[0098] The A matrix is ​​the coefficient matrix of the plane feature normal vector, and l is the difference in distance between the plane feature and the origin. The translation vector can be obtained by formula 17.

[0099] T=(A T WA) -1 A T Wl (17)

[0100] Similarly, the matrix W is a diagonal weight matrix and is calculated by formula (14).

[0101] Through the above calculations, we obtain the rotation matrix and translation vector from the vehicle-mounted point cloud to the airborne point cloud, and use the rotation matrix and translation vector to optimize the position of the vehicle-mounted point cloud.

[0102] The present invention utilizes airborne point clouds to improve the position accuracy of vehicle-mounted laser point clouds. The technical solution of the present invention aims to solve the problem of deviation in the position accuracy of vehicle-mounted laser point clouds. Planar features in non-homologous point clouds are used as registration primitives to complete the registration of vehicle-mounted point clouds to reference airborne point clouds, thereby improving the position accuracy of vehicle-mounted point clouds.

[0103] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.

Claims

1. A method for improving the position accuracy of vehicle-mounted laser point cloud assisted by airborne point cloud, characterized in that: The following steps are involved: 1) Establish indexes for the airborne point cloud and the vehicle-borne point cloud respectively, and divide the entire space of the airborne point cloud and the vehicle-borne point cloud into regular three-dimensional cube spaces, i.e., voxels, according to a given spatial resolution; 2) Calculate the geometric features of all voxels in the airborne point cloud and the vehicle-mounted point cloud respectively, and select the voxels whose distribution features are surface features; calculate the plane parameters for the surface features of the selected voxels; 3) Taking the airborne point cloud as a reference, the surface features of the voxels in the airborne point cloud obtained in step 2) are used as reference planes to construct the distance of plane features with the same name, search for the plane feature voxels corresponding to the spatial neighborhood of the vehicle-borne point cloud, and find all corresponding plane feature pairs between the airborne point cloud and the vehicle-borne point cloud; Step 3) Find all corresponding plane feature pairs between the airborne point cloud and the vehicle-borne point cloud, as follows: For any reference plane in the airborne point cloud, the neighborhood voxels of its corresponding vehicle-borne point cloud are obtained, and the plane features Pi to be registered for all the neighborhood voxels are calculated. j The plane characteristic distance Δd of {0≤j≤26} ij , the plane features with the smallest distance form a plane feature pair; Among them, n i and n j They are the reference plane features P i and the plane feature Ρ of the neighboring voxels to be registered j The normal vector of i and x j They are the plane feature point sets composed of the refined voxels {(x k ,y k ,z k ),1≤k≤n}, where 4) Using all corresponding planar feature pairs between the airborne point cloud and the vehicle-borne point cloud as registration primitives, calculate the rotation matrix and translation vector from the vehicle-borne point cloud to the airborne point cloud, and use the rotation matrix and translation vector to optimize the position of the vehicle-borne point cloud.

2. The method for improving vehicle-mounted laser point cloud position accuracy assisted by airborne point cloud according to claim 1, characterized in that: In the step 1), an octree structure is used to respectively index the airborne point cloud and the vehicle-borne point cloud, and the entire space of the airborne point cloud and the vehicle-borne point cloud is divided into a regular three-dimensional cube space.

3. The method for improving vehicle-mounted laser point cloud position accuracy assisted by airborne point cloud according to claim 1, characterized in that: In step 2), the geometric features of all voxels in the airborne point cloud and the vehicle-mounted point cloud are calculated, and voxels whose distribution features are surface features are screened out, as follows: 2.1) Using the principal component analysis method, the geometric features of the point cloud within the voxel are calculated through the eigenvalue decomposition of the covariance matrix to screen out voxels with surface features; p i , i = 1, 2, ..., k, is the k points in the voxel, and the covariance matrix M of the point cloud in the voxel is calculated by the following formula: in, 2.2) The covariance matrix M obtains the corresponding eigenvector and three eigenvalues ​​λ1, λ2, and λ3 through the eigenvalue decomposition process, where λ1≥λ2≥λ3>0; the three eigenvalues ​​together describe the distribution of the point cloud in the voxel, and the geometric characteristics of the point cloud are expressed by defining the operation between the three features, which are a 1D 、a 2D 、a 3D , which represent line features, surface features, and volume features respectively. The geometric features are calculated using the following formula: when a 2D When the value is the largest, that is, when the following formula is satisfied, it indicates that the distribution feature of the point cloud in the current voxel is a surface feature; 2.3) According to the above steps, the geometric features of all voxels in the airborne point cloud and the vehicle-mounted point cloud are calculated respectively, and the voxels with significant planar features are screened out.

4. The method for improving vehicle-mounted laser point cloud position accuracy assisted by airborne point cloud according to claim 1, characterized in that: Before calculating the plane parameters in step 2), the plane features of the screened voxels are refined as follows: A random sampling consistency method is adopted to refine the planar features of the screened voxels.

5. The method for improving vehicle-mounted laser point cloud position accuracy assisted by airborne point cloud according to claim 3, characterized in that: The plane parameters calculated in step 2) are specifically as follows: By randomly selecting points in the point cloud space, the point cloud is divided into plane interior points and exterior points, the plane interior points within the voxels are retained, and the plane parameters are calculated; In 3D Euclidean space, the plane equation is expressed as follows: ax+by+cz=d; For a three-dimensional plane in space, the normal vector n of the plane is expressed as n = (a, b, c). If the normal vector n is a normalized vector, then n·n = 1 is satisfied; d in the plane equation is the distance from the plane to the origin of the coordinate system; For each face feature voxel, the plane point set {(x i ,y i ,z i ),1≤i≤k}, calculate its plane parameters; For the plane that best fits the point set, the sum of the distances from all points to the plane Minimum, using the orthogonal regression method, the normalized normal vector n of the plane is obtained by the eigenvector corresponding to the minimum eigenvalue of the matrix A. The matrix A is obtained by in, After the calculation, the normal vector n is obtained, that is, the first three parameters a, b, c of the plane equation are obtained, and the fourth parameter d of the plane equation can be calculated, which is obtained by the following formula:

6. The method for improving vehicle-mounted laser point cloud position accuracy assisted by airborne point cloud according to claim 1, characterized in that: The calculation of the rotation matrix and the translation vector in step 4) is to use the angle relationship between the normal vectors of the same-name plane feature pairs and the difference in the origin distance between the same-name plane feature pairs to respectively calculate the rotation matrix and the translation vector.

Images