Point cloud automatic extraction and registration method and device for underground tunneling

Abstract

The application relates to a point cloud automatic extraction and registration method and device for underground tunneling, and belongs to the field of three-dimensional point cloud data, and the technical key points are as follows: original tunnel point cloud is subjected to statistical filtering to remove noise, target point cloud is identified and extracted by utilizing the high light intensity characteristics of an L-shaped plane target, and a plurality of L-shaped targets are segmented; least square plane fitting is carried out on each target point cloud, projection is carried out on the fitting plane, and uniform transformation is carried out on the XOY plane; two center axes of each target are extracted, the random sampling consistency algorithm is adopted to fit the center axes and obtain the intersection, inverse transformation is carried out to restore to the original coordinate system, and the coordinates of the target homonymic feature points are obtained; at least four pairs of homonymic feature points are used to carry out coarse registration through the quaternion method, and fine registration is realized in combination with the ICP algorithm, the problem of difficult initial pose estimation in various underground spaces such as weak texture, heavy dust and strong water mist is solved, and the accuracy and efficiency of point cloud registration are improved.

Description

Technical Field

[0001] This application relates to the field of three-dimensional point cloud data technology, and in particular to a method and apparatus for automatic extraction and registration of point clouds in underground tunnels or roadways. Background Technology

[0002] With the development of technology and the improvement of industrial automation, the intelligent construction of coal mine roadways and tunnels has become crucial for improving coal mine safety and efficiency. In the intelligent construction of coal mine roadways and tunnels, building accurate 3D models of the roadways or tunnels is fundamental for applications such as monitoring surrounding rock deformation and classifying roof bolts. To achieve this goal, Simultaneous Localization and Mapping (SLAM) technology has become the core of research. Laser SLAM technology, unaffected by dust obstruction and changes in lighting, can obtain accurate 3D spatial distance information in real time, making it particularly suitable for environments with low light and high dust levels, such as underground coal mines and tunnels.

[0003] In laser SLAM technology, point cloud registration is a crucial process. It involves spatially aligning point cloud data acquired at different times or locations to ensure the continuity and consistency of the mapping. Point cloud registration is typically divided into two stages: coarse registration and fine registration. The purpose of coarse registration is to estimate the approximate transformation relationship between two point clouds, providing an initial estimate for fine registration, which aims to achieve high-precision point cloud alignment. Currently, researchers have proposed various point cloud coarse registration algorithms, including those based on geometric feature descriptions and those based on global search. Although these methods improve the accuracy and efficiency of registration to some extent, they still have certain limitations in the special environment of coal mine roadways or underground tunnels. For example, while the PCA algorithm based on geometric feature description can calculate the covariance and rotation matrices of multiple point clouds, it has low robustness to noise and outliers. The RANSAC algorithm based on global search can handle noisy data, but its single probability-based constraints lead to low algorithm stability and a tendency to produce false matches. Therefore, a generalized four-point consistency coarse registration algorithm assisted by a planar target is adopted.

[0004] However, existing target feature point extraction methods are not very applicable when dealing with single, narrow structures such as coal mines, roadways, and tunnels in underground spaces. In practical applications, due to the influence of factors such as low light, humidity, and noise, the scanned planar target point cloud data is prone to missing or redundant phenomena, which further increases the difficulty of point cloud registration.

[0005] Therefore, developing a new method for extracting and registering feature points from target point clouds in underground mining to improve the accuracy and efficiency of point cloud registration has become an urgent technical problem to be solved. Summary of the Invention

[0006] The embodiments of this application provide a method and apparatus for automatic point cloud extraction and registration in underground tunnels or roadways, which at least have the technical effect of improving the accuracy and efficiency of point cloud registration in low-light and high-dust environments.

[0007] To address the aforementioned technical problems, according to one aspect of this application, an automatic point cloud extraction and registration method for underground tunnels or excavation roadways is provided, comprising:

[0008] A kd-tree index structure is constructed in the original point cloud data and the target point cloud data, and statistical filtering is performed to remove discrete noise points and retain the main structure of the alley or tunnel.

[0009] Based on the high reflectivity of L-shaped planar targets, target point clouds are extracted using dynamic intensity thresholding, and Euclidean clustering is performed on the extracted point clouds to identify and separate multiple L-shaped planar targets.

[0010] For each target point cloud, least squares plane fitting is performed, the target point cloud is projected onto the fitting plane, and then unified to the XOY plane through coordinate transformation to eliminate attitude differences;

[0011] Based on the principle of average coordinate statistics, two central axes are extracted for each L-shaped target. The random sampling consistency algorithm is used to fit the central axes and find the intersection points. The inverse transformation is then used to restore the original coordinate system to obtain the three-dimensional coordinates of the target's corresponding feature points.

[0012] By utilizing at least four pairs of target feature points with the same name in the two point clouds, the initial transformation matrix is ​​calculated using the quaternion method for coarse registration, and then the ICP algorithm is combined for fine registration to achieve high-precision alignment of the point clouds.

[0013] Among them, obtaining the three-dimensional coordinates of the target's corresponding feature points includes:

[0014] A statistical analysis algorithm is used to extract the same feature points. Several random intervals are randomly generated within a given range. The coordinates of all points in the point cloud within each interval are determined. The average coordinates of each interval are calculated, and a new point cloud is generated at the coordinates of the average coordinates.

[0015] The new point cloud is fitted with a straight line. This process is repeated multiple times. The model with the most interior points is selected as the final estimation result. The approximate axes of the two long sides of the L-shaped planar target are then fitted.

[0016] A matrix is ​​constructed and a linear regression model is applied to calculate the slope and intercept of the two lines. The coordinates of the intersection point are calculated using the intersection formula of the line equations. The coordinates of the corresponding feature points of the L-shaped planar target are obtained by the inverse transformation of the fitting plane rotation and translation.

[0017] According to another aspect of this application, an automatic point cloud extraction and registration device for underground tunnels or roadways is also claimed, comprising:

[0018] The acquisition unit is configured to construct a kd-tree index structure in the original point cloud data and the target point cloud data, and perform statistical filtering to remove discrete noise points and retain the main structure of the alley or tunnel.

[0019] The first processing unit is configured to extract target point clouds based on the high reflectivity characteristics of L-shaped planar targets by using dynamic intensity thresholds, and to perform Euclidean clustering segmentation on the extracted point clouds in order to identify and separate multiple L-shaped planar targets.

[0020] The second processing unit is configured to perform least squares plane fitting on each target point cloud, project the target point cloud onto the fitting plane, and unify it to the XOY plane through coordinate transformation to eliminate attitude differences.

[0021] The third processing unit is configured to extract the two orthogonal central axes of each L-shaped target based on the principle of average coordinate statistics, use the random sampling consistency algorithm to fit the central axes and find the intersection points, and restore them to the original coordinate system through inverse transformation to obtain the three-dimensional coordinates of the target's corresponding feature points.

[0022] The unit is determined and configured to perform coarse registration using the quaternion method based on the coordinates of the feature points of the two-phase point cloud targets, and then perform fine registration using the ICP algorithm.

[0023] According to another aspect of this application, a computer-readable storage medium is also claimed, the computer-readable storage medium including a stored program wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the automatic point cloud extraction and registration method for underground tunnels or roadways as described in any of the preceding claims.

[0024] According to another aspect of this application, an electronic device is also claimed, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including a point cloud automatic extraction and registration method for performing any of the above-described underground tunneling roadways or tunnels.

[0025] This application has the following beneficial effects:

[0026] The method for automatic point cloud extraction and registration of underground tunnels or roadways in this application identifies and extracts L-shaped planar targets by utilizing intensity differences. After performing planar fitting and projecting onto the fitted plane, the target planar point cloud is transformed to the XOY plane. The two central axes of the L-shaped target are extracted, and the central axes are fitted with straight lines and the intersection points are calculated. The original fitted plane position is transformed back to obtain the coordinates of the target's corresponding feature points. Coarse registration is performed using four pairs of corresponding target feature points, and then fine registration is performed using ICP, thus completing the point cloud registration. This solves the problem that it is difficult to obtain the initial pose of a single roadway or tunnel feature.

[0027] The point cloud automatic extraction and registration device for underground tunnels or roadways disclosed in this application is applicable to the tunneling process of tunnels or roadways during underground space construction. First, it acquires the original point cloud data. Then, according to the roadway point cloud preprocessing method, it filters out noise from the initial roadway point cloud concentration. Subsequently, it identifies and extracts L-shaped planar targets using intensity differences, performs planar fitting, and projects the target planar point cloud onto the fitted plane. After transforming the target planar point cloud to the XOY plane, it extracts the two central axes of the L-shaped target, performs linear fitting of the central axes, and finds the intersection point. It then transforms back to the original fitted plane position to obtain the coordinates of the target's corresponding feature points. Coarse registration is performed using four pairs of corresponding target feature points, followed by fine registration using ICP, thus completing the point cloud registration. This solves the problem of difficulty in obtaining the initial pose of a single roadway feature. Attached Figure Description

[0028] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0029] Figure 1 A flowchart illustrating an automatic point cloud extraction and registration method for underground tunnels or excavation roadways provided according to an embodiment of this application is shown.

[0030] Figure 2 A schematic diagram illustrating the statistical principle of average coordinates provided according to an embodiment of this application is shown;

[0031] Figure 3 A schematic diagram of a centerline straight-line fitting provided according to an embodiment of this application is shown;

[0032] Figure 4 A structural block diagram of an automatic point cloud extraction and registration device for underground tunnels or roadways provided according to an embodiment of this application is shown. Detailed Implementation

[0033] As described in the background section, existing methods for automatic extraction and registration of target homonymous feature points are not highly applicable when dealing with single, narrow structures in underground coal mine roadways. Furthermore, in practical applications, the scanned planar target point cloud data is prone to missing or redundant data due to factors such as low light, humidity, and noise, further increasing the difficulty of point cloud registration. Therefore, to address the difficulties in existing target homonymous feature point extraction and registration methods, embodiments of this application provide an automatic point cloud extraction and registration method, apparatus, computer, readable storage medium, and electronic device for underground tunnels or roadways.

[0034] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0035] This embodiment provides an automatic point cloud extraction and registration method for underground tunnels or roadways. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than that shown here.

[0036] Figure 1 This is a flowchart illustrating an automatic point cloud extraction and registration method for underground tunnels or excavation roadways, according to an embodiment of this application. The steps include:

[0037] Step S101: Construct a kd-tree index structure in the original point cloud data and the target point cloud data, and perform statistical filtering to remove discrete noise points and retain the main structure of the alley or tunnel.

[0038] Step S102: Based on the high reflectivity of the L-shaped planar target, the target point cloud is extracted by dynamic intensity threshold, and the extracted point cloud is segmented by Euclidean clustering to identify and separate multiple L-shaped planar targets.

[0039] Step S103: Perform least squares plane fitting on each target point cloud, project the target point cloud onto the fitting plane, and unify it to the XOY plane through coordinate transformation to eliminate attitude differences.

[0040] Step S104: Based on the principle of average coordinate statistics, extract the two central axes of each L-shaped target, use the random sampling consistency algorithm to fit the central axes and find the intersection points, and restore them to the original coordinate system through inverse transformation to obtain the three-dimensional coordinates of the target's corresponding feature points.

[0041] Step S105: Using at least four pairs of target feature points with the same name from the two point clouds, the initial transformation matrix is ​​calculated using the quaternion method for coarse registration, and then fine registration is performed by combining the ICP algorithm to achieve high-precision alignment of the point clouds.

[0042] It should be noted that the method and apparatus of this application are not only applicable to coal mine roadways and tunnel excavation, but also applicable to the construction of large-scale underground space excavation, underground garages, underground storage, underground civil defense, hydropower generation, mining, coal mining, road (railway) ventilation shafts and other low-light and high-dust environments, depending on the environment.

[0043] Depending on the underlying environment, this method includes the following basic steps:

[0044] Based on the high reflectivity of L-shaped planar targets, target point clouds are extracted using dynamic intensity thresholding, and Euclidean clustering is performed on the extracted point clouds to identify and separate multiple L-shaped planar targets. Before identifying and extracting L-shaped planar targets, the method further includes: obtaining an initial set of roadway or tunnel point clouds; preprocessing the initial set of roadway or tunnel point clouds to obtain the final set of roadway or tunnel point clouds, thereby filtering out noise in the initial set of roadway or tunnel point clouds.

[0045] Specifically, point cloud processing software is used to manually crop the point cloud of the alley or tunnel to remove the incompletely scanned part of the point cloud of the alley or tunnel and the large-scale noise point cloud; statistical filtering denoising method is used to remove individual discrete point clouds to complete the point cloud preprocessing.

[0046] After extracting the target point cloud based on the reflectance intensity characteristics of L-shaped planar targets in the point cloud of the alleyway or tunnel, Euclidean clustering segmentation is performed to identify and extract the L-shaped planar targets, including:

[0047] The target point cloud area is roughly selected manually for coarse extraction. Then, the average value of the maximum and minimum backlight reflection intensity within the coarsely extracted target point cloud area is calculated. 1.5 times Point clouds exceeding the threshold are extracted as the target point cloud fine extraction result.

[0048] To facilitate individual processing of each L-shaped planar target, the target point cloud is extracted and then clustered for segmentation. First, a seed point is selected as the starting point of the current cluster. Then, all unclassified points are traversed, and the Euclidean distance between them and the seed point is calculated. If the distance is less than a preset threshold, these points are assigned to the same cluster as the seed point. Next, for points in the same cluster, the above process is repeated, adding points whose distance to any point in the current cluster is less than the threshold to that cluster. This process continues until all points are classified into a cluster, thus completing the entire Euclidean clustering segmentation process.

[0049] The least squares plane fitting method is used to fit the target point cloud. After projecting the target point cloud onto the fitting plane, it is rotated and translated to the XOY plane. Specifically, the 3D point cloud is projected onto the best-fit plane, and the coordinate transformation relationship of the point cloud projection and rotation is derived. The corresponding projection relationship is shown in the following formula:

[0050] Let the general equation of a three-dimensional plane be:

[0051]

[0052] Assuming the three-dimensional spatial coordinates of the L-shaped target point cloud that is not on a plane are: The coordinates of its projection point on the plane are Then the formula for the coordinates of the projection point is:

[0053]

[0054] This allows us to obtain the projection coordinates of the L-shaped target point cloud onto the fitting plane. After obtaining the projection plane of the L-shaped target, the transformation matrix is ​​obtained using the Rodriguez formula based on the normal vector of this plane. The rotation matrix R is then multiplied by the target point cloud data coordinates on the fitting plane to obtain the rotated point cloud.

[0055] Based on the principle of average coordinates, the two centerlines of the L-plane target are extracted. Straight line fitting is performed on the centerlines, and the intersection points are calculated. The original positions are then transformed to obtain the coordinates of the target's corresponding feature points. Specifically, taking the calculation of the average abscissa by generating intervals on the Y-axis as an example, several random intervals are randomly generated within a given range on the Y-axis, denoted as... Determine all point clouds contained within each interval, in each interval Find all points in the point cloud whose coordinates lie within this interval, and denote them as... For each interval containing point clouds Calculate the average coordinates of these point clouds on the X-axis. That is, for each interval... Calculate all points The average value of the coordinates on the x-axis The formula is as follows:

[0056]

[0057] in, It is an interval The number of midpoint clouds, x(p) These are the coordinates of point p on the X-axis. For interval Minimum Y-coordinate of all points within the area For interval The maximum value of the Y-coordinate of all points within the area. , These are the x and y coordinates of a random interval, calculated for each interval. A new point cloud is generated at the coordinates.

[0058] Linear fitting is performed on the newly generated point cloud. A model is fitted to a small subset of samples from the data. The distances from other data points to the model are calculated. Data points with distances less than a certain threshold are classified as inliers, and those with distances greater than the threshold are classified as exliers. This process is repeated multiple times, and the model with the most inliers is selected as the final estimation result, thus approximating the axes of the two longer sides of the L-shaped planar target.

[0059] By constructing a matrix and applying a linear regression model to calculate the slope and intercept of the two lines, the problem is transformed into finding the intersection point of two-dimensional plane lines. The coordinates of the intersection point are calculated using the intersection point formula of the line equation. Then, by using the inverse transformation of the fitting plane rotation and translation mentioned above, the coordinates of the corresponding feature points of the L-shaped plane target can be obtained.

[0060] Figure 2 To illustrate, using a rectangle as an example, here is a diagram showing the points near the midline of its longest side. Figure 3 This is a schematic diagram illustrating the process of fitting a straight line after extracting points near the central axis.

[0061] After coarse registration using SVD decomposition of the coordinates of corresponding feature points from the two point cloud targets, fine registration is performed using the ICP algorithm to complete the point cloud registration. Specifically, the aforementioned method for extracting corresponding feature points from the L target is used to extract four pairs of corresponding coordinate points from the two point clouds. The transformation matrix between the original point cloud data and the target point cloud data is calculated using the SVD matrix decomposition method. The original point cloud data is then transformed as a whole and concatenated with the target point cloud data to complete the coarse registration. Finally, ICP registration is performed on the two point clouds to complete the fine point cloud registration.

[0062] To enable those skilled in the art to better understand the technical solution of this application, the implementation process of the automatic point cloud extraction and registration method for underground tunnels or tunnels of this application will be described in detail below with reference to specific embodiments.

[0063] In one embodiment, a method for automatic point cloud extraction and registration of underground tunnels or roadways is provided, comprising the following steps:

[0064] Step S1: Construct a kd-tree index structure in the original point cloud data and the target point cloud data, and perform statistical filtering to remove discrete noise points and retain the main structure of the alley or tunnel.

[0065] Step S2: Based on the high reflectivity of the L-shaped planar target, the target point cloud is extracted by dynamic intensity thresholding, and the extracted point cloud is segmented by Euclidean clustering to identify and separate multiple L-shaped planar targets.

[0066] Step S3: Perform least squares plane fitting on each target point cloud, project the target point cloud onto the fitting plane, and unify it to the XOY plane through coordinate transformation to eliminate attitude differences;

[0067] Step S4: Extract the two orthogonal central axes of each L-shaped target based on the principle of average coordinate statistics, use the random sampling consensus algorithm to fit the central axes and find the intersection points, and restore them to the original coordinate system through inverse transformation to obtain the three-dimensional coordinates of the target's corresponding feature points;

[0068] Step S5: After coarse registration by performing SVD decomposition using the coordinates of the same feature points of the two phases of point cloud targets, fine registration is then performed using the ICP algorithm to complete the point cloud registration.

[0069] After statistical filtering of the original roadway or tunnel point cloud, L-shaped planar targets are identified and extracted using intensity differences. Planar fitting is performed, and after projection onto the fitted plane, the target planar point cloud is transformed to the XOY plane. The two central axes of the L-shaped target are extracted, and the central axis straight lines are fitted and the intersection points are found. The original fitted plane position is transformed back to obtain the coordinates of the target's corresponding feature points. Coarse registration is performed using four pairs of corresponding target feature points, and then fine registration is performed using ICP, thus completing the point cloud registration. This solves the problem that it is difficult to obtain the initial pose of a single roadway or tunnel feature.

[0070] This application also provides an automatic point cloud extraction and registration device for underground tunnels or roadways, along with an automatic point cloud extraction and registration method and device for underground tunnels or roadways. Both can be applied to coal mining machines, tunneling machines, shield tunneling machines, or tunnel excavation equipment in underground spaces. It should be noted that the automatic point cloud extraction and registration device for underground tunnels or roadways in this application can be used to execute the automatic point cloud extraction and registration method for underground tunnels or roadways provided in this application. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0071] The following describes the automatic point cloud extraction and registration device for underground tunnels or roadways provided in the embodiments of this application.

[0072] Figure 4 This is a structural block diagram of an automatic point cloud extraction and registration device for underground tunnels or excavation roadways, provided according to an embodiment of this application.

[0073] like Figure 4 As shown, the device includes:

[0074] The acquisition unit 51 is configured to construct a kd-tree index structure in the original point cloud data and the target point cloud data, and perform statistical filtering to remove discrete noise points and retain the main structure of the alley or tunnel.

[0075] The first processing unit 52 is configured to extract target point cloud based on the high reflectivity characteristics of L-shaped planar targets by using dynamic intensity thresholds, and to perform Euclidean clustering segmentation on the extracted point cloud to identify and separate multiple L-shaped planar targets.

[0076] The second processing unit 53 is configured to perform least squares plane fitting on each target point cloud, project the target point cloud onto the fitting plane, and unify it to the XOY plane through coordinate transformation to eliminate attitude differences.

[0077] The third processing unit 54 is configured to extract the two orthogonal central axes of each L-shaped target based on the average coordinate statistical principle, use the random sampling consistency algorithm to fit the central axis and find the intersection point, and restore it to the original coordinate system through inverse transformation to obtain the three-dimensional coordinates of the target's corresponding feature points.

[0078] Unit 55 is configured to perform coarse registration using the quaternion method based on the coordinates of the target feature points in the two-phase point cloud, and then perform fine registration using the ICP algorithm.

[0079] The aforementioned method and apparatus for automatic point cloud extraction and registration of underground tunnels or roadways includes a processor and a memory. The acquisition unit, first processing unit, second processing unit, third processing unit, and determination unit are all stored as program units in the memory. The processor executes these program units stored in the memory to achieve the corresponding functions. All of the above modules are located in the same processor; alternatively, the modules may be located in different processors in any combination.

[0080] The processor contains a kernel, which retrieves the corresponding program unit from memory. One or more kernels can be configured, and adjusting kernel parameters can address the low accuracy and efficiency of existing coarse registration methods for single features of roadways or tunnels.

[0081] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0082] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the automatic point cloud extraction and registration method for underground tunnels or roadways.

[0083] This invention provides a processor for running a program, wherein the program executes the method for automatic point cloud extraction and registration of underground tunnels or roadways.

[0084] This invention provides a device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps: Constructing a kd-tree index structure in the original point cloud data and target point cloud data, and performing statistical filtering; based on the high reflectivity characteristics of L-shaped planar targets, extracting target point clouds through dynamic intensity thresholding, and performing Euclidean clustering segmentation on the extracted point clouds to identify and separate multiple L-shaped planar targets; to facilitate subsequent extraction of target homonymous feature points, performing least-squares plane fitting on each target point cloud, projecting the target point cloud onto the fitting plane, and then rotating and translating it to the XOY plane; using the statistical analysis algorithm proposed in this paper to extract homonymous feature points, extracting the two central axes of the L-shaped planar target according to the average coordinate statistical principle, using a random sampling consensus algorithm to perform straight-line fitting of the central axes and find the intersection points, transforming back to the original positions to obtain the coordinates of the target homonymous feature points; then, using the coordinates of the target feature points in the two point clouds for coarse registration using the quaternion method, and then using the ICP algorithm for fine registration, finally completing the point cloud registration. The devices mentioned in this article can be servers, PCs, tablets, mobile phones, etc.

[0085] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps: constructing a kd-tree index structure in the original point cloud data and target point cloud data, performing statistical filtering to remove discrete noise points and retain the main structure of the roadway or tunnel; extracting the target point cloud using the intensity features of the L-shaped planar target and then performing Euclidean clustering segmentation to identify and extract the L-shaped planar target; to facilitate the subsequent extraction of target homonymous feature points, performing least squares plane fitting on each target point cloud, projecting the target point cloud onto the fitting plane and then rotating and translating it to the XOY plane; using the statistical analysis algorithm proposed in this paper to extract homonymous feature points, extracting the two central axes of the L-shaped planar target according to the average coordinate statistical principle, using the random sampling consensus algorithm to perform straight-line fitting of the central axes and find the intersection point, transforming back to the original position to obtain the coordinates of the target homonymous feature points; then, using the coordinates of the target feature points in the two point clouds to perform coarse registration using the quaternion method, and then using the ICP algorithm to perform fine registration, finally completing the point cloud registration.

[0086] This application also provides an electronic device, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include methods for executing any of the above-described automatic point cloud extraction and registration methods for underground tunnels or excavation roadways. After statistical filtering of the original roadway or tunnel point cloud, an L-shaped planar target is identified and extracted using intensity differences. Planar fitting is performed, and after projection onto the fitted plane, the target planar point cloud is transformed to the XOY plane. The two central axes of the L-shaped target are extracted, and straight-line fitting of the central axes is performed to find the intersection point. The position is transformed back to the original fitted plane to obtain the coordinates of the target's corresponding feature points. Coarse registration is performed using four pairs of corresponding target feature points, followed by fine registration using ICP, thus completing the point cloud registration. This solves the problem of difficulty in obtaining the initial pose of a single roadway or tunnel feature.

[0087] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0088] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for automatic point cloud extraction and registration of underground tunnels or roadways, characterized in that, include: A kd-tree index structure is constructed in the original point cloud data and the target point cloud data, and statistical filtering is performed to remove discrete noise points and retain the main structure of the alley or tunnel. Based on the high reflectivity of L-shaped planar targets, target point clouds are extracted using dynamic intensity thresholding, and Euclidean clustering is performed on the extracted point clouds to identify and separate multiple L-shaped planar targets. For each target point cloud, least squares plane fitting is performed, the target point cloud is projected onto the fitting plane, and then unified to the XOY plane through coordinate transformation to eliminate attitude differences; Based on the principle of average coordinate statistics, two orthogonal central axes of each L-shaped target are extracted. The random sampling consensus algorithm is used to fit the central axes and find the intersection points. The inverse transformation is used to restore the original coordinate system and obtain the three-dimensional coordinates of the target's corresponding feature points. By utilizing at least four pairs of target feature points with the same name in the two point clouds, the initial transformation matrix is ​​calculated using the quaternion method for coarse registration, and then the ICP algorithm is combined for fine registration to achieve high-precision alignment of the point clouds. Among them, obtaining the three-dimensional coordinates of the target's corresponding feature points includes: A statistical analysis algorithm is used to extract the same feature points. Several random intervals are randomly generated within a given range. The coordinates of all points in the point cloud within each interval are determined. The average coordinates of each interval are calculated, and a new point cloud is generated at the coordinates of the average coordinates. The new point cloud is fitted with a straight line. This process is repeated multiple times. The model with the most interior points is selected as the final estimation result. The approximate axes of the two long sides of the L-shaped planar target are then fitted. A matrix is ​​constructed and a linear regression model is applied to calculate the slope and intercept of the two lines. The coordinates of the intersection point are calculated using the intersection formula of the line equations. The coordinates of the corresponding feature points of the L-shaped planar target are obtained by the inverse transformation of the fitting plane rotation and translation.

2. The method for automatic point cloud extraction and registration of underground tunnels or roadways according to claim 1, characterized in that, Before identifying and extracting the L-shaped planar target, the process also includes: acquiring raw point cloud data; preprocessing the raw point cloud data to obtain a set of point clouds for the alleyway or tunnel, wherein the preprocessing includes manually cropping the raw point cloud data, removing incompletely scanned parts of the alleyway or tunnel point cloud and large-scale noise point cloud, and removing individual discrete point clouds using statistical filtering denoising method.

3. The method for automatic point cloud extraction and registration of underground tunnels or excavation roadways according to claim 1, characterized in that, The step of constructing a kd-tree index structure in the original point cloud data and the target point cloud data, and performing statistical filtering to remove discrete noise points and retain the main structure of the roadway or tunnel includes: For the acquired raw point cloud data and target point cloud data, each level of the kd tree uses a hyperplane perpendicular to the corresponding axis to divide all the children along a preset dimension. Specifically, dividing all the children along the preset dimension means that at the root of the tree, all the child nodes will be divided according to the first dimension, and each level in the tree will be divided in the next dimension. When other dimensions are exhausted, it returns to the first dimension. By filling point cloud data into a kd-tree data structure, statistical analysis of all points in the point cloud can be accelerated, and the average distance between a point and its k nearest neighbors can be calculated. in, It is a point The distance to a point in its neighborhood. It is a point The average distance to its neighboring points It is a point The three-dimensional coordinates yes The coordinates of any point within the KNN neighborhood, It is the number of nearest neighbors, i.e., the K value in KNN; Calculate the mean distance and standard deviation of global points in the point cloud: Set statistical filter threshold ,in This represents the average distance from all points in the point cloud to their KNN neighborhood. It is a calculation coefficient. It is the standard deviation of the average distance from all points in the point cloud to their neighborhood. It is the total number of points in the point cloud, when > When the point is out of range, it is marked as an outlier and removed; the statistical filtering of the point cloud is completed by traversing all points in the point cloud.

4. The method for automatic point cloud extraction and registration of underground tunnels or excavation roadways according to claim 1, characterized in that, Based on the high reflectivity of L-shaped planar targets, the following steps are taken to extract target point clouds using dynamic intensity thresholding and then perform Euclidean clustering on the extracted point clouds to identify and separate multiple L-shaped planar targets: A coarse extraction is performed on the target point cloud region, and then the average of the maximum and minimum values ​​of the backlight reflection intensity within the coarsely extracted target point cloud region is calculated. 1.5 times Point clouds with values ​​greater than the threshold are extracted as the target point cloud fine extraction result; Select a seed point as the starting point of the current cluster, then traverse all unclassified points and calculate their Euclidean distance from the seed point. If the distance is less than a preset threshold, then classify these points into the same cluster as the seed point. For points in the same cluster, repeat the above process, adding points whose distance from any point in the current cluster is less than a threshold to that cluster, until all points are classified into a certain cluster.

5. The method for automatic point cloud extraction and registration of underground tunnels or excavation roadways according to claim 1, characterized in that, The steps of performing least-squares plane fitting on each target point cloud, projecting the target point cloud onto the fitting plane, and unifying it to the XOY plane through coordinate transformation to eliminate attitude differences include: projecting the three-dimensional point cloud onto the best fitting plane; after obtaining the projection plane of the L-shaped target by coordinate transformation of the point cloud projection and rotation; obtaining the transformation matrix using the Rodriguez formula based on the normal vector of this plane; and multiplying the target point cloud data coordinates on the fitting plane by the rotation matrix R to obtain the rotated point cloud.

6. The method for automatic point cloud extraction and registration of underground tunnels or roadways according to claim 1, characterized in that, The steps for achieving high-precision alignment of point clouds by utilizing at least four pairs of corresponding target feature points from two point clouds, calculating the initial transformation matrix using the quaternion method for coarse registration, and then combining this with the ICP algorithm for fine registration include: extracting four pairs of corresponding coordinate points from the two point clouds using the L-target corresponding feature point extraction method, and calculating the transformation matrix between the original point cloud data and the target point cloud data using the SVD matrix decomposition method. After the original point cloud data is transformed as a whole, it is stitched together with the target point cloud data to complete the coarse registration. Then, ICP registration is performed on the two point clouds to complete the fine registration of the point cloud.

7. An automatic point cloud extraction and registration device for underground tunnels or roadways, characterized in that, include: The acquisition unit is configured to construct a kd-tree index structure in the original point cloud data and the target point cloud data, and perform statistical filtering to remove discrete noise points and retain the main structure of the alley or tunnel. The first processing unit is configured to extract target point clouds based on the high reflectivity characteristics of L-shaped planar targets by using dynamic intensity thresholds, and to perform Euclidean clustering segmentation on the extracted point clouds to identify and separate multiple L-shaped planar targets. The second processing unit performs least-squares plane fitting on each target point cloud, projects the target point cloud onto the fitting plane, and unifies it to the XOY plane through coordinate transformation to eliminate attitude differences. The third processing unit is configured to extract the two orthogonal central axes of each L-shaped target based on the principle of average coordinate statistics, use the random sampling consistency algorithm to fit the central axes and find the intersection points, and restore them to the original coordinate system through inverse transformation to obtain the three-dimensional coordinates of the target's corresponding feature points; The unit is determined and configured to perform coarse registration using the quaternion method based on the coordinates of the feature points of the two-phase point cloud targets, and then perform fine registration using the ICP algorithm.

8. The automatic point cloud extraction and registration device for underground tunnels or roadways according to claim 7 is applicable to coal mining machines, tunneling machines, shield tunneling machines, or tunnel excavation equipment in underground spaces.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a computing device, the computing device performs the method for automatic point cloud extraction and registration of underground tunnels or roadways as described in any one of claims 1 to 6.

10. An electronic device, characterized in that, The method includes: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include a method for performing automatic point cloud extraction and registration of underground tunnels or tunnels as described in any one of claims 1 to 6.

Images