A data and model hybrid driven lidar point cloud 3D modeling method for transmission towers

Abstract

The present invention discloses a data and model hybrid-driven lidar point cloud 3D modeling method for power transmission towers, belonging to the technical field of 3D modeling of power transmission lines. The method comprises the following steps: establishing an independent 3D coordinate system for the tower for input point cloud data, calculating the rotation angle of the point cloud to the coordinate system, and rotating the point cloud to obtain redirected tower point cloud data; performing component-level segmentation on the resulting tower point cloud using a PointNet++ network to obtain point cloud data of the tower head, insulator string, and tower body; the method converts airborne lidar point cloud data into a 3D model, realizing fully automated modeling with high fitting accuracy and strong robustness. The processing process of the method of the present invention is free of approximation, realizing the conversion of 3D point cloud data into a 3D model. The results can be applied to the fields of 3D design, visualization, and management of power grid models, and used for model exchange and information sharing.

Description

Technical Field

[0001] The present invention belongs to the technical field of three-dimensional modeling of transmission lines, and in particular relates to a data and model hybrid driven lidar point cloud three-dimensional modeling method for transmission towers. Background Art

[0002] Transmission towers, as essential infrastructure for power transmission lines, play a crucial role in national economic development and daily production. As key components of power transmission systems, transmission towers are meticulously designed with considerations of power transmission safety and structural stress analysis in mind. Three-dimensional tower model reconstruction methods are gaining increasing attention. Reconstructed 3D tower models can accurately determine tower locations and parameters, enabling applications in disaster monitoring and line monitoring.

[0003] Chinese patent application No. 201811634889.3 discloses an automated tower modeling method based on airborne LIDAR point cloud data. The method involves automatically extracting towers and obtaining point cloud data for individual towers. The method then segments the towers and uses features such as the number of points ni, the eigenvalue ratio ri, and ni / ri in the subdivided subset to determine the tower segmentation locations. The point cloud above the mid-arm crossbeam is projected onto the XY plane, and the principal directions are determined using a PCA algorithm. The tower point cloud is then rotated based on the determined principal directions. A RANSAC algorithm is then used to perform linear fitting based on the previously determined segmentation locations to determine structural parameters and implement the automated modeling method. The selected features are robust to certain noise interference and partial data loss, demonstrating good adaptability. Automatic modeling is possible for towers with a certain amount of noise and partial data loss, and the modeling stability is excellent.

[0004] The above-mentioned disclosed patent cannot effectively solve the problem of automatic modeling of transmission towers when used. At the same time, it cannot perform three-dimensional modeling of lidar point cloud transmission towers through mixed driving of data and models, and cannot improve the efficiency of transmission tower modeling. Summary of the Invention

[0005] To solve the problems raised in the above background technology, the present invention provides a data and model hybrid-driven lidar point cloud transmission tower 3D modeling method, which can realize the conversion of 3D point cloud data into 3D models. The results can be applied to the fields of 3D design, visualization and management of power grid models, and used for model exchange and information sharing.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a data and model hybrid-driven lidar point cloud transmission tower 3D modeling method, comprising the following steps:

[0007] S1: Establish a tower-independent 3D coordinate system for the input point cloud data, calculate the rotation angle of the point cloud to the coordinate system, and rotate the point cloud to obtain the re-oriented tower point cloud data;

[0008] S2: The tower point cloud obtained in step S1 is segmented at the component level using the PointNet++ network to obtain point cloud data of the tower head, insulator string, and tower body;

[0009] S3: matching the tower head point cloud data obtained in step S2 with the tower head in the tower head model library, and then registering with the matched tower head model to obtain a three-dimensional tower head model;

[0010] S4: Extract the coordinates of the hanging points of the insulator strings from the tower head point cloud data obtained in step S2, combine the type and posture information of the insulator strings in the power engine, and hang the corresponding insulator 3D model in the insulator model library on the tower head 3D model obtained in step S3;

[0011] S5: fitting the tower body point cloud data obtained in step S2 to the tower body outer contour model, extracting the tower body steel convergence node coordinates, and generating a three-dimensional model of the tower body internal diagonal material in combination with the outer contour model;

[0012] S6: For the three-dimensional models of the tower head and tower body constructed in the results obtained in steps S4 and S5, the models are reversely restored to the original point cloud coordinate system of the tower according to the rotation angle information obtained in step S1 to obtain the three-dimensional model of the tower;

[0013] In step S2, the PointNet++ network method is used to perform component-level segmentation. The specific method is described as follows: a bilateral filtering method is used to remove ground and vegetation noise outside the tower from the input tower point cloud; a manual labeling method is used to create a point cloud component segmentation dataset of three categories: tower head, insulator string, and tower body; a PointNet++ point cloud component segmentation network model is established and trained to obtain parameters of the tower point cloud component segmentation network model; and the trained model is used to perform component-level segmentation on the tower point cloud to be processed to obtain point clouds after segmentation of the three components: tower head, insulator string, and tower body.

[0014] Furthermore, in step S1, establishing a tower-independent three-dimensional coordinate system for the input point cloud data, calculating the rotation angle of the point cloud to the coordinate system, and rotating the point cloud to obtain the reoriented tower point cloud data includes the following methods and steps:

[0015] (1) Establishing the three-dimensional coordinate system of the tower: Obtain the point cloud at the bottom of the tower, define the average center point of the bottom point cloud as the origin O of the three-dimensional coordinate system of the tower, and define the X, Y, and Z directions of the point cloud as the X-axis, Y-axis, and Z-axis of the coordinate system respectively;

[0016] (2) Reorientation of the tower point cloud: Project the tower bottom point cloud onto the XOY plane and convert it into a binary image. Obtain the minimum circumscribed rectangle of the binary image. Calculate the direction of the rectangle edge and the rotation angle of the independent coordinate system XY axis and rotate the point cloud to obtain the reoriented tower point cloud data.

[0017] Furthermore, in step S3, the tower head point cloud data obtained in step S2 is matched with the tower head in the tower head model library, and then registered with the matched tower head model to obtain a three-dimensional tower head model, which includes the following methods and steps:

[0018] (1) Point cloud to binary image conversion: Project the tower head point cloud segmented in step S2 onto the XOZ plane, set the pixel points at the point cloud position to 255, and the blank area to 0 to generate a binary image of the tower head point cloud;

[0019] (2) Converting the tower head 3D model to a binary image: Project the 3D model in the tower head model library onto the XOZ plane, extract the vertices of the 3D model to obtain the point cloud data of the model, set the pixel points at the point cloud position to 255, and the blank area to 0 to generate a binary image of the tower head 3D model;

[0020] (3) Tower head matching: The binary image of the tower head 3D model is resampled to the same resolution as the point cloud binary image. The edge of the model is projected onto the point cloud image using the TPS thin plate spline transformation method. The matching cost between the point cloud image points and the model image points is calculated. The tower head model image with the minimum cost is selected as the final matching result.

[0021] (4) Tower head model registration: SOR filtering is used to denoise the tower head point cloud, and the ICP registration method is used to align the tower head point cloud obtained in step S2 and the point cloud data corresponding to the matched tower head model to obtain the spatial transformation matrix T from the tower head three-dimensional model to the tower head point cloud coordinate system. The three-dimensional model is multiplied by the transformation matrix T to obtain the tower head three-dimensional model in the point cloud coordinate system.

[0022] Furthermore, in step S5, the outer contour model of the tower body is fitted to the tower body point cloud data obtained in step S2, and the coordinates of the tower body steel convergence nodes are extracted, and a three-dimensional model of the inner oblique material of the tower body is generated in combination with the outer contour model, including the following methods and steps:

[0023] (1) Fitting the outer contour model of the tower body: Divide the tower body point cloud into n layers with equal height Δh along the Z axis direction, and calculate the minimum circumscribed rectangle of the convex hull of each layer of point cloud; extract the corner points in the four directions of the minimum circumscribed rectangle of each layer; all the corner points in a certain direction constitute the point set of a main leg of the tower body, and construct the straight line equation by selecting any two points from the set, and calculate the distance from the remaining points to the straight line. The points less than the threshold Td are defined as inliers, and the above process is repeated to select the straight line with the most inliers as the three-dimensional fitting line in that direction; the corner points in the other three directions are also processed in the same way to obtain the three-dimensional model of the outer contour of the four main legs of the tower body;

[0024] (2) Extracting the transverse diaphragm of the tower: the density is defined as the number of midpoints of each layer, and the length is defined as the maximum value of the distance between the midpoint of each layer and the Z axis in the X direction. The Kth layer that meets both the maximum density and minimum length values ​​within the range L is defined as the transverse diaphragm position;

[0025] (3) Extracting the coordinates of the steel convergence node: converting the tower body component point cloud into a binary image, using the morphological method to remove noise from the obtained image and roughly extract the skeleton information, and using the LSD straight line segment detection method to finely extract the skeleton information, and calculating the intersection coordinates of the extracted straight lines as the coordinates of the steel convergence node;

[0026] (4) Generate a three-dimensional model of the tower body: connect the steel convergence nodes in the two transverse diaphragms with the intersection of the main legs of the transverse diaphragms to obtain a three-dimensional model of the diagonal materials inside the tower body.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention converts airborne lidar point cloud data into a three-dimensional model, achieving fully automated modeling with high fitting accuracy and strong robustness. The processing process of the method of the present invention does not involve approximation, and the conversion of three-dimensional point cloud data into a three-dimensional model is achieved. The results can be applied to the fields of three-dimensional design, visualization and management of power grid models, and used for model exchange and information sharing. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0031] See also Figure 1The present invention provides the following technical solution: a data and model hybrid driven lidar point cloud transmission tower three-dimensional modeling method, comprising the following steps:

[0032] S1: Establish a tower-independent 3D coordinate system for the input point cloud data, calculate the rotation angle of the point cloud to the coordinate system, and rotate the point cloud to obtain the re-oriented tower point cloud data;

[0033] S2: The tower point cloud obtained in step S1 is segmented at the component level using the PointNet++ network to obtain point cloud data of the tower head, insulator string, and tower body;

[0034] S3: matching the tower head point cloud data obtained in step S2 with the tower head in the tower head model library, and then registering with the matched tower head model to obtain a three-dimensional tower head model;

[0035] S4: Extract the coordinates of the hanging points of the insulator strings from the tower head point cloud data obtained in step S2, combine the type and posture information of the insulator strings in the power engine, and hang the corresponding insulator 3D model in the insulator model library on the tower head 3D model obtained in step S3;

[0036] S5: fitting the tower body point cloud data obtained in step S2 to the tower body outer contour model, extracting the tower body steel convergence node coordinates, and generating a three-dimensional model of the tower body internal diagonal material in combination with the outer contour model;

[0037] S6: For the three-dimensional models of the tower head and tower body constructed in the results obtained in steps S4 and S5, the models are reversely restored to the original point cloud coordinate system of the tower according to the rotation angle information obtained in step S1 to obtain the three-dimensional model of the tower;

[0038] In step S2, the PointNet++ network method is used to perform component-level segmentation. The specific method is described as follows: the input tower point cloud is subjected to bilateral filtering to remove ground and vegetation noise outside the tower; a manual labeling method is used to create a point cloud component segmentation dataset of three categories: tower head, insulator string, and tower body; a PointNet++ point cloud component segmentation network model is established and trained to obtain the parameters of the tower point cloud component segmentation network model; the trained model is used to perform component-level segmentation on the tower point cloud to be processed, and the point cloud after the tower head, insulator string, and tower body are segmented is obtained.

[0039] Furthermore, in the present invention, in step S1, a tower-independent three-dimensional coordinate system is established for the input point cloud data, a rotation angle of the point cloud to the coordinate system is calculated, and the point cloud is rotated to obtain the reoriented tower point cloud data, including the following methods and steps:

[0040] (1) Establishing the three-dimensional coordinate system of the tower: Obtain the point cloud at the bottom of the tower, define the average center point of the bottom point cloud as the origin O of the three-dimensional coordinate system of the tower, and define the X, Y, and Z directions of the point cloud as the X-axis, Y-axis, and Z-axis of the coordinate system respectively;

[0041] (2) Reorientation of the tower point cloud: Project the tower bottom point cloud onto the XOY plane and convert it into a binary image. Obtain the minimum circumscribed rectangle of the binary image. Calculate the direction of the rectangle edge and the rotation angle of the independent coordinate system XY axis and rotate the point cloud to obtain the reoriented tower point cloud data.

[0042] By adopting the above technical solution, an independent three-dimensional coordinate system of the tower is established for the input point cloud data, the rotation angle of the point cloud to the coordinate system is calculated, and the point cloud is rotated to obtain the redirected tower point cloud data.

[0043] Furthermore, in the present invention, in step S3, the tower head point cloud data obtained in step S2 is matched with the tower head in the tower head model library, and then aligned with the matched tower head model to obtain a three-dimensional tower head model, which includes the following methods and steps:

[0044] (1) Point cloud to binary image conversion: Project the tower head point cloud segmented in step S2 onto the XOZ plane, set the pixel points at the point cloud position to 255, and the blank area to 0 to generate a binary image of the tower head point cloud;

[0045] (2) Converting the tower head 3D model to a binary image: Project the 3D model in the tower head model library onto the XOZ plane, extract the vertices of the 3D model to obtain the point cloud data of the model, set the pixel points at the point cloud position to 255, and the blank area to 0 to generate a binary image of the tower head 3D model;

[0046] (3) Tower head matching: The binary image of the tower head 3D model is resampled to the same resolution as the point cloud binary image. The edge of the model is projected onto the point cloud image using the TPS thin plate spline transformation method. The matching cost between the point cloud image points and the model image points is calculated. The tower head model image with the minimum cost is selected as the final matching result.

[0047] (4) Tower head model registration: SOR filtering is used to denoise the tower head point cloud, and the ICP registration method is used to align the tower head point cloud obtained in step S2 and the point cloud data corresponding to the matched tower head model to obtain the spatial transformation matrix T from the tower head three-dimensional model to the tower head point cloud coordinate system. The three-dimensional model is multiplied by the transformation matrix T to obtain the tower head three-dimensional model in the point cloud coordinate system.

[0048] By adopting the above technical solution, the tower head point cloud data is matched with the tower head in the tower head model library, and then aligned with the matched tower head model to obtain a three-dimensional model of the tower head.

[0049] Furthermore, in step S5 of the present invention, the tower body point cloud data obtained in step S2 is used to fit the outer contour model of the tower body, and the coordinates of the tower body steel convergence nodes are extracted, and a three-dimensional model of the inner oblique material of the tower body is generated in combination with the outer contour model, which includes the following methods and steps:

[0050] (1) Fitting the outer contour model of the tower body: Divide the tower body point cloud into n layers with equal height Δh along the Z axis direction, and calculate the minimum circumscribed rectangle of the convex hull of each layer of point cloud; extract the corner points in the four directions of the minimum circumscribed rectangle of each layer; all the corner points in a certain direction constitute the point set of a main leg of the tower body, and construct the straight line equation by selecting any two points from the set, and calculate the distance from the remaining points to the straight line. The points less than the threshold Td are defined as inliers, and the above process is repeated to select the straight line with the most inliers as the three-dimensional fitting line in that direction; the corner points in the other three directions are also processed in the same way to obtain the three-dimensional model of the outer contour of the four main legs of the tower body;

[0051] (2) Extracting the transverse diaphragm of the tower: the density is defined as the number of midpoints of each layer, and the length is defined as the maximum value of the distance between the midpoint of each layer and the Z axis in the X direction. The Kth layer that meets both the maximum density and minimum length values ​​within the range L is defined as the transverse diaphragm position;

[0052] (3) Extracting the coordinates of the steel convergence node: converting the tower body component point cloud into a binary image, using the morphological method to remove noise from the obtained image and roughly extract the skeleton information, and using the LSD straight line segment detection method to finely extract the skeleton information, and calculating the intersection coordinates of the extracted straight lines as the coordinates of the steel convergence node;

[0053] (4) Generate a three-dimensional model of the tower body: connect the steel convergence nodes in the two transverse diaphragms with the intersection of the main legs of the transverse diaphragms to obtain a three-dimensional model of the diagonal materials inside the tower body.

[0054] By adopting the above technical solution, the outer contour model of the tower body is fitted according to the tower body point cloud data, and the coordinates of the tower body steel convergence nodes are extracted. Combined with the outer contour model, a three-dimensional model of the tower body's internal oblique materials is generated.

[0055] The working principle and use process of the present invention: When the present invention is used: first obtain the point cloud of the bottom of the tower, define the average center point of the bottom point cloud as the origin O of the three-dimensional coordinate system of the tower, and define the X, Y, and Z directions of the point cloud as the X axis, Y axis, and Z axis of the coordinate system respectively; project the point cloud of the bottom of the tower to the XOY plane and convert it into a binary image, obtain the minimum circumscribed rectangle of the binary image, calculate the direction of the rectangle edge and the rotation angle of the XY axis of the independent coordinate system and rotate the point cloud to obtain the redirected tower point cloud data; project the tower head point cloud segmented in step S2 to the XOZ plane, set the pixel points of the point cloud position to 255, and the blank area to 0 to generate a binary image of the tower head point cloud; The 3D model in the tower head model library is projected onto the XOZ plane, and the vertices of the 3D model are extracted to obtain the point cloud data of the model. The pixel points at the point cloud position are set to 255, and the blank area is set to 0 to generate a binary image of the tower head 3D model; the binary image of the tower head 3D model is resampled to the same resolution as the point cloud binary image, and the TPS thin plate spline transformation method is used to project the edge of the model onto the point cloud image. The matching cost between the point cloud image points and the model image points is calculated, and the tower head model image with the smallest cost is selected as the final matching result; the tower head point cloud is denoised using SOR filtering, and the ICP registration method is used to match the tower head point cloud obtained in step S2 with the point cloud corresponding to the matched tower head model. The data is registered and processed to obtain the spatial transformation matrix T from the three-dimensional model of the tower head to the tower head point cloud coordinate system. The three-dimensional model is multiplied by the transformation matrix T to obtain the three-dimensional model of the tower head in the point cloud coordinate system; the point cloud of the tower body is divided into n layers of equal height Δh along the Z-axis direction, and the minimum circumscribed rectangle of the convex hull of each layer of point cloud is calculated; the corner points in the four directions of the minimum circumscribed rectangle of each layer are extracted; all the corner points in a certain direction constitute the point set of a main leg of the tower body, and any two points in the set are selected to construct the straight line equation, and the distance from the remaining points to the straight line is calculated. The points less than the threshold Td are defined as inliers, and the above process is repeated to select the straight line with the most inliers as the three-dimensional fitting straight line in that direction; The same processing method is used for its corner points in three directions to obtain the three-dimensional model of the outer contour of the four main legs of the tower body; the density is defined as the number of midpoints in each layer, and the length is defined as the value with the maximum distance from the midpoint of each layer to the Z axis in the X direction. The Kth layer that satisfies both the maximum density and minimum length within the range L is defined as the cross-diaphragm position; the point cloud of the tower body components is converted into a binary image, and the resulting image is subjected to morphological methods to remove noise and roughly extract skeleton information. The skeleton information is refined by the LSD straight line segment detection method, and the coordinates of the intersection of the extracted lines are calculated as the coordinates of the steel convergence node; the steel convergence nodes in the two cross-diaphragms are connected with the intersection of the main legs of the cross-diaphragms to obtain a three-dimensional model of the diagonal material inside the tower body.

[0056] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A data and model hybrid-driven lidar point cloud transmission tower 3D modeling method, characterized by: The steps include: S1: Establish an independent three-dimensional coordinate system for the tower based on the input point cloud data, calculate the rotation angle of the point cloud to the coordinate system, and rotate the point cloud to obtain the re-oriented tower point cloud data; S2: The tower point cloud obtained in step S1 is segmented at the component level using the PointNet++ network to obtain point cloud data of the tower head, insulator string, and tower body; S3: matching the tower head point cloud data obtained in step S2 with the tower head in the tower head model library, and then registering with the matched tower head model to obtain a three-dimensional tower head model; S4: Extract the coordinates of the hanging points of the insulator strings from the tower head point cloud data obtained in step S2, combine the type and posture information of the insulator strings in the power engine, and hang the corresponding insulator 3D model in the insulator model library on the tower head 3D model obtained in step S3; S5: Fitting the tower body point cloud data obtained in step S2 to the tower body outer contour model, extracting the tower body steel convergence node coordinates, and generating a three-dimensional model of the tower body internal diagonal material in combination with the outer contour model, including the following steps: (1) Fitting the outer contour model of the tower body: Divide the tower body point cloud into n layers with equal height Δh along the Z axis direction, and calculate the minimum circumscribed rectangle of the convex hull of each layer of point cloud; extract the corner points in the four directions of the minimum circumscribed rectangle of each layer; all the corner points in a certain direction constitute the point set of a main leg of the tower body, and construct the straight line equation by selecting any two points from the set, and calculate the distance from the remaining points to the straight line. The points less than the threshold Td are defined as inliers, and the above process is repeated to select the straight line with the most inliers as the three-dimensional fitting line in that direction; the corner points in the other three directions are also processed in the same way to obtain the three-dimensional model of the outer contour of the four main legs of the tower body; (2) Extracting the transverse diaphragm of the tower: the density is defined as the number of midpoints of each layer, and the length is defined as the maximum value of the distance between the midpoint of each layer and the Z axis in the X direction. The Kth layer that meets both the maximum density and minimum length values ​​within the range L is defined as the transverse diaphragm position; (3) Extracting the coordinates of the steel convergence node: converting the tower body component point cloud into a binary image, using the morphological method to remove noise from the obtained image and roughly extract the skeleton information, and using the LSD straight line segment detection method to finely extract the skeleton information, and calculating the intersection coordinates of the extracted straight lines as the coordinates of the steel convergence node; (4) Generate a three-dimensional model of the tower body: connect the steel convergence nodes in the two transverse diaphragms with the intersection of the main legs of the transverse diaphragms to obtain a three-dimensional model of the diagonal material inside the tower body S6: For the three-dimensional models of the tower head and tower body constructed in the results obtained in steps S4 and S5, the models are reversely restored to the original point cloud coordinate system of the tower according to the rotation angle information obtained in step S1 to obtain the three-dimensional model of the tower; In step S2, the PointNet++ network method is used to perform component-level segmentation. The specific method is described as follows: a bilateral filtering method is used to remove ground and vegetation noise outside the tower from the input tower point cloud; a manual labeling method is used to create a point cloud component segmentation dataset of three categories: tower head, insulator string, and tower body; a PointNet++ point cloud component segmentation network model is established and trained to obtain parameters of the tower point cloud component segmentation network model; and the trained model is used to perform component-level segmentation on the tower point cloud to be processed to obtain point clouds after segmentation of the three components: tower head, insulator string, and tower body.

2. The data and model hybrid-driven lidar point cloud 3D modeling method for transmission towers according to claim 1, characterized in that: The step S1 establishes a tower-independent three-dimensional coordinate system for the input point cloud data, calculates the rotation angle of the point cloud to the coordinate system, and rotates the point cloud to obtain the re-oriented tower point cloud data, including the following methods and steps: (1) Establishing the three-dimensional coordinate system of the tower: Obtain the point cloud at the bottom of the tower, define the average center point of the bottom point cloud as the origin O of the three-dimensional coordinate system of the tower, and define the X, Y, and Z directions of the point cloud as the X-axis, Y-axis, and Z-axis of the coordinate system respectively; (2) Reorientation of the tower point cloud: Project the tower bottom point cloud onto the XOY plane and convert it into a binary image. Obtain the minimum circumscribed rectangle of the binary image. Calculate the direction of the rectangle edge and the rotation angle of the independent coordinate system XY axis and rotate the point cloud to obtain the reoriented tower point cloud data.

3. The data and model hybrid-driven lidar point cloud 3D modeling method for transmission towers according to claim 1, characterized in that: In step S3, the tower head point cloud data obtained in step S2 is matched with the tower head in the tower head model library, and then registered with the matched tower head model to obtain a three-dimensional tower head model, which includes the following methods and steps: (1) Point cloud to binary image conversion: Project the tower head point cloud segmented in step S2 onto the XOZ plane, set the pixel points at the point cloud position to 255, and the blank area to 0 to generate a binary image of the tower head point cloud; (2) Converting the tower head 3D model to a binary image: Project the 3D model in the tower head model library onto the XOZ plane, extract the vertices of the 3D model to obtain the point cloud data of the model, set the pixel points at the point cloud position to 255, and the blank area to 0 to generate a binary image of the tower head 3D model; (3) Tower head matching: The binary image of the tower head 3D model is resampled to the same resolution as the point cloud binary image. The edge of the model is projected onto the point cloud image using the TPS thin plate spline transformation method. The matching cost between the point cloud image points and the model image points is calculated. The tower head model image with the minimum cost is selected as the final matching result. (4) Tower head model registration: SOR filtering is used to denoise the tower head point cloud, and the ICP registration method is used to align the tower head point cloud obtained in step S2 and the point cloud data corresponding to the matched tower head model to obtain the spatial transformation matrix T from the tower head three-dimensional model to the tower head point cloud coordinate system. The three-dimensional model is multiplied by the transformation matrix T to obtain the tower head three-dimensional model in the point cloud coordinate system.

Images