A method for obstacle framing based on laser point cloud in a converter station scene

Abstract

The application relates to the technical field of obstacle detection, in particular to an obstacle framing method based on laser point clouds in a converter station scene, which comprises the following steps: based on the converter station environment, three-dimensional scanning is carried out through a laser radar, point cloud data is collected, noise filtering is carried out on the data, spatial alignment of the data is carried out, and processed point cloud data is obtained. In the application, three-dimensional scanning is carried out through the laser radar, and the scanning angle and resolution are adjusted, the coverage rate and precision of the point cloud data are optimized, the point cloud data is mapped to the adjacent grid center, the data structure is simplified, the point cloud data is horizontally cut, and the linear feature of the electric wire is identified, the obstacle recognition process is optimized, the accuracy and delicacy of the obstacle recognition are improved through the combination of spatial statistical analysis and clustering analysis, and the obstacles are framed through the construction of the bounding box of each obstacle, so that three-dimensional space positioning and shape recognition of the obstacles are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of obstacle detection, and in particular to an obstacle framing method based on laser point cloud in a converter station scenario. BACKGROUND

[0002] The technical field of obstacle detection mainly focuses on identifying, locating and classifying objects in the environment through various sensor systems. Obstacle detection technology is widely used in autonomous vehicles, drones, robot navigation and security monitoring systems. Key technologies include image and signal processing, machine learning algorithms and sensor fusion technology. Obstacle detection not only involves feature extraction from sensor data, but also includes real-time analysis using deep learning and other methods to improve detection accuracy and reliability.

[0003] Among them, the obstacle framing method based on laser point cloud in the converter station scenario involves using laser point cloud technology to detect and locate obstacles in the converter station environment. Laser point cloud is a large number of precise spatial point data collected by laser scanners, which can construct a three-dimensional digital model of the environment. In the converter station, this technology is used to identify obstacles that affect drone inspection, including equipment, pipelines and power lines within the converter station. By identifying and framing obstacles, it ensures that drones can perform inspections in the complex environment of the converter station, improving the operational efficiency of the converter station.

[0004] Traditional methods rely on pre-set parameters and fixed algorithms, which cannot cope with the challenges brought by the diversity of equipment, pipelines and power line layout within the converter station. Traditional methods are limited in data processing efficiency by the complexity of the algorithm and computing resources, and have problems such as low data processing efficiency and difficulty in accurately locating small-range obstacles. Common problems include data overload, where a large amount of unoptimized point cloud data collected by sensors can easily cause processing bottlenecks. Traditional methods are not fine enough in classifying and analyzing the specific attributes of obstacles, and cannot effectively distinguish between different types of obstacles, which can easily lead to misjudgment or omission, increasing the risk in the automated inspection process. SUMMARY

[0005] The purpose of the present application is to solve the shortcomings in the prior art and to propose an obstacle framing method based on laser point cloud in a converter station scenario.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical solution: an obstacle framing method based on laser point cloud in a converter station scenario, comprising the following steps:

[0007] S1: Based on the converter station environment, perform three-dimensional scanning by laser radar, collect point cloud data, and filter out noise from the data, perform spatial alignment of the data, and obtain processed point cloud data;

[0008] S2: mapping each point in the data to a neighboring grid center based on the processed point cloud data, calculating the coordinate average of points in each grid, setting the result point as the grid representative, and obtaining simplified point cloud data;

[0009] S3: based on the simplified point cloud data, filtering the point cloud data in contact with the ground, creating a simulated cloth model, inverting the point cloud data, adjusting the position of the simulated cloth to fit the ground represented by the point cloud, segmenting the ground and non-ground point cloud, and obtaining ground segmentation data;

[0010] S4: based on the ground segmentation data, performing horizontal cutting on the point cloud data, projecting each layer of point cloud onto a two-dimensional plane, identifying the linear features of the wires on each layer, and obtaining wire identification point cloud data;

[0011] S5: based on the wire identification point cloud data, measuring the density between the point clouds, marking the densely aggregated areas as potential obstacles, classifying the obstacles, and obtaining obstacle classification labels;

[0012] S6: based on the obstacle classification labels, calculating the geometric center point and the circumscribed rectangular bounding box of each marked obstacle, constructing the bounding box of each obstacle, and obtaining the obstacle bounding result.

[0013] As a further scheme of the present application, the processed point cloud data includes position coordinates, reflection intensity and color information of the point cloud, the simplified point cloud data includes voxelized point cloud, average position point inside each voxel and reconstructed point cloud model with reduced spatial resolution, the ground segmentation data includes ground point cloud identified after simulated cloth fitting, non-ground point cloud part and ground and non-ground threshold, the wire identification point cloud data includes identified wire feature points and wire point cloud segments, the obstacle classification labels include divided obstacle categories and corresponding obstacle clustering labels, and the obstacle bounding result includes three-dimensional coordinates, length, width and height dimensions of each obstacle bounding box and obstacle volume size.

[0014] As a further scheme of the present application, based on the converter station environment, three-dimensional scanning is performed by laser radar, point cloud data is collected, noise filtering is performed on the data, spatial alignment of the data is performed, and the step of obtaining the processed point cloud data is specifically as follows:

[0015] S101: based on the converter station environment, adjusting the scanning angle and resolution of the laser radar, performing panoramic three-dimensional scanning on the converter station, and obtaining a basic point cloud data set;

[0016] S102: Based on the basic point cloud data set, by analyzing the density and reflection characteristics of the point cloud, and filtering out the noise of the data, identifying and removing the data points with abnormal reflection intensity and deviating from the normal position, obtaining the point cloud data filtered out of noise;

[0017] S103: Based on the point cloud data filtered out of noise, adjusting the coordinates of the data points, eliminating the offset generated by scanning, and verifying the continuity of the point cloud data through continuity detection, obtaining the processed point cloud data.

[0018] As a further scheme of the application, based on the processed point cloud data, each point in the data is mapped to the adjacent grid center, the coordinate average of the points in each grid is calculated, and the result point is set as the grid representative, to obtain the simplified point cloud data. The steps are specifically:

[0019] S201: Based on the processed point cloud data, spatial compression is performed, and the point cloud data is mapped into a pre-set spatial grid according to the spatial position of each point, to obtain grid-mapped point cloud data;

[0020] S202: Based on the grid-mapped point cloud data, the coordinate average of the point cloud set in each grid is calculated, to obtain the grid coordinate average calculation result;

[0021] S203: Based on the grid coordinate average calculation result, the coordinate average is set as the representative point of each grid, to obtain the simplified point cloud data.

[0022] As a further scheme of the application, based on the simplified point cloud data, the point cloud data in contact with the ground is screened, a simulated cloth model is created, the point cloud data is inverted, the position of the simulated cloth is adjusted to fit the ground represented by the point cloud, and the ground and non-ground point clouds are segmented, to obtain the ground segmentation data. The steps are specifically:

[0023] S301: Based on the simplified point cloud data, by analyzing the Z coordinate of each point and the predetermined ground height difference, the data points in contact with the ground are screened, and the point set close to and equal to the ground height is identified, to obtain the ground point cloud information;

[0024] S302: Based on the ground point cloud information, the physical properties of the cloth are simulated, the elastic coefficient and grid size of the cloth model are set, the covering behavior of the cloth in the virtual environment is simulated, and the cloth is adjusted to adapt to the terrain, to obtain the simulated cloth configuration;

[0025] S303: Based on the simulated cloth configuration, the cloth is matched with the inverted point cloud, the position of the cloth model is adjusted, the cloth model is fitted to the inverted terrain features, the ground and non-ground areas are segmented through the contact points of the cloth and the terrain, and the ground segmentation data is obtained.

[0026] As a further scheme of the present application, based on the ground segmentation data, the point cloud data is horizontally cut, each layer of point cloud is projected onto a two-dimensional plane, the linear feature of the electric wire on each layer is identified, and the step of obtaining the electric wire recognition point cloud data is specifically:

[0027] S401: Based on the ground segmentation data, the entire point cloud data set is processed by height interval layering, each layer height is fixed, the point cloud is cut into multiple horizontal layers according to a preset interval, and layer segmentation point cloud is obtained.

[0028] S402: Based on the layer segmentation point cloud, the point cloud of each horizontal layer is projected onto a two-dimensional plane, the XY coordinates of the point cloud are plotted on the two-dimensional plane, and two-dimensional projection data is obtained.

[0029] S403: Based on the two-dimensional projection data, a linear feature detection technology is used to analyze the linear arrangement feature of the projection graph, the projection point cloud with linear feature is screened, the point set with linear arrangement consistent with the characteristics of the electric wire is marked, and the electric wire recognition point cloud data is obtained.

[0030] As a further scheme of the present application, the linear feature detection technology is according to the formula:

[0031] ;

[0032] The linear feature score of the point cloud data is calculated, wherein, represents the length of the longest linear feature in the projection data, represents the average distance between adjacent features, represents the standard deviation of the linear feature direction, 、 and is an adjustment coefficient, represents the linear feature score of the point cloud data.

[0033] As a further scheme of the present application, based on the electric wire recognition point cloud data, the density between the point clouds is measured, the region with dense aggregation is marked as a potential obstacle, the obstacle is classified, and the step of obtaining the obstacle classification label is specifically:

[0034] S501: Based on the electric wire recognition point cloud data, a kernel density estimation method is used to perform spatial statistical analysis on each point cloud block, the number and distribution of points inside the point cloud block are calculated, the density of each point cloud region is identified, and point cloud density information is generated.

[0035] S502: Based on the point cloud density information, by comparing with a preset density reference, the point cloud region exceeding the preset density reference is identified as a potential obstacle, and marked obstacle region data is obtained.

[0036] S503: Based on the marked obstacle region data, analyze the color information of the point cloud region, and classify the obstacles into corresponding types according to the color characteristics of the point cloud, including metal, wood and plastic, to obtain obstacle classification labels.

[0037] As a further scheme of the present application, the kernel density estimation method is according to the formula:

[0038] ;

[0039] Calculate the point cloud density , wherein, represents the point cloud density, represents the number of points in the point cloud block, represents the position of the th point, is the average position of the point cloud block, is the standard deviation of the point position, is the base of the natural logarithm, is the ratio of the circumference to the diameter.

[0040] As a further scheme of the present application, based on the obstacle classification label, the geometric center point and the circumscribed rectangular bounding box of each marked obstacle are calculated, and the bounding box of each obstacle is constructed to obtain the obstacle framing result. The steps are specifically:

[0041] S601: Based on the obstacle classification label, the average value of each clustered point cloud coordinate is calculated to identify the geometric center of the cluster, and the center point is taken as the representative of the obstacle position to obtain the obstacle center point coordinate;

[0042] S602: Based on the obstacle center point coordinate, the distance from each cluster farthest point to the center point is measured to identify the size and position of the circumscribed rectangular bounding box of each obstacle, and the obstacle bounding box parameter is obtained;

[0043] S603: Based on the obstacle bounding box parameter, a three-dimensional bounding box of each obstacle is constructed by three-dimensional modeling, each obstacle is covered by the bounding box, and the position and shape of the bounding box in space are recorded to obtain the obstacle framing result.

[0044] Compared with the prior art, the present application has the following advantages and positive effects:

[0045] In the application, three-dimensional scanning is performed by laser radar, and the scanning angle and resolution are adjusted to optimize the coverage and accuracy of point cloud data. Through optimization of point cloud data processing such as noise filtering and spatial alignment, the data usability is enhanced. The point cloud data is mapped to the adjacent grid center, and the coordinate average value is calculated to simplify the data structure and make subsequent data processing more efficient. The point cloud data in contact with the ground is screened, and a simulated cloth model is used to segment the ground and non-ground point cloud. The point cloud data is horizontally cut and the linear features of the wire are identified to optimize the obstacle recognition process. Through the combination of spatial statistical analysis and cluster analysis, the accuracy and detail of obstacle recognition are improved. By constructing the bounding box of each obstacle, the obstacle is framed to realize the three-dimensional spatial positioning and shape recognition of the obstacle. It ensures that the unmanned aerial vehicle can safely and effectively avoid obstacles when performing tasks, and enhances the safety and efficiency of the converter station inspection. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The main steps of the application are shown in the figure;

[0047] Figure 2 The S1 refinement of the application is shown in the figure;

[0048] Figure 3 The S2 refinement of the application is shown in the figure;

[0049] Figure 4 The S3 refinement of the application is shown in the figure;

[0050] Figure 5 The S4 refinement of the application is shown in the figure;

[0051] Figure 6 The S5 refinement of the application is shown in the figure;

[0052] Figure 7 The S6 refinement of the application is shown in the figure. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and do not limit the application.

[0054] In the description of the present application, it should be understood that the terms "length", "width", "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application. In addition, in the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0055] Please refer to Figure 1 The present application provides a technical solution: a method for obstacle framing based on laser point cloud in a converter station scene, comprising the following steps:

[0056] S1: Based on the environment of the converter station, three-dimensional scanning is performed by laser radar, and the scanning angle and resolution are adjusted to optimize the coverage and accuracy of the point cloud data. The point cloud data is collected, and the data is filtered to remove abnormal points caused by scanning errors and correct the coordinate offset of the scanning data. The data is spatially aligned to optimize the continuity between the point clouds, and the processed point cloud data is obtained;

[0057] S2: Based on the processed point cloud data, the point cloud data is spatially compressed, each point in the data is mapped to the center of the adjacent grid, the average value of the coordinates of each point in the grid is calculated, and the result point is set as the grid representative. Simplified point cloud data is obtained;

[0058] S3: Based on the simplified point cloud data, the point cloud data in contact with the ground is screened, a simulated cloth model is created, the point cloud data is inverted, the position of the simulated cloth is adjusted to fit the ground represented by the point cloud, the ground and non-ground point clouds are segmented, and ground segmentation data is obtained;

[0059] S4: Based on the ground segmentation data, the point cloud data is horizontally cut according to the preset height interval, the point cloud is divided into multiple horizontal layers, each layer of point cloud is projected onto a two-dimensional plane, the linear features of the wires on each layer are identified, the point cloud of the wire part is marked, and wire identification point cloud data is obtained;

[0060] S5: Based on the wire identification point cloud data, the density between the point clouds is measured through spatial statistical analysis, clustering analysis is performed using the density benchmark, the densely aggregated area is marked as a potential obstacle, and the obstacle is classified according to the color characteristics of the point cloud, and an obstacle classification label is obtained;

[0061] S6: Based on the obstacle classification label, calculate the geometric center point and the circumscribed rectangular bounding box of each marked obstacle, construct the bounding box of each obstacle through three-dimensional space modeling, locate and characterize the spatial position and morphology of the obstacle, and obtain the obstacle bounding result.

[0062] The processed point cloud data includes position coordinates, reflection intensity and color information of the point cloud, the simplified point cloud data includes voxelized point cloud, average position point inside each voxel and reconstructed point cloud model with reduced spatial resolution, the ground segmentation data includes ground point cloud identified after simulating cloth fitting, non-ground point cloud part and ground and non-ground threshold, the wire identification point cloud data includes identified wire feature points and wire point cloud segments, the obstacle classification label includes divided obstacle categories and corresponding obstacle cluster labels, and the obstacle bounding result includes three-dimensional coordinates, length, width and height dimensions and obstacle volume size of each obstacle bounding box.

[0063] Please refer to Figure 2 , based on the converter station environment, three-dimensional scanning is performed by laser radar, and the scanning angle and resolution are adjusted to optimize the coverage and accuracy of the point cloud data, the point cloud data is collected, and the data is filtered to remove abnormal points generated by scanning errors, and the coordinate offset of the scanning data is corrected, the data is spatially aligned, the continuity between the point clouds is optimized, and the steps of obtaining the processed point cloud data are as follows:

[0064] S101: Based on the converter station environment, adjust the scanning angle and resolution of the laser radar, and perform panoramic three-dimensional scanning of the converter station to obtain the basic point cloud data set, and the process is as follows:

[0065] Based on the converter station environment, set the scanning parameters of the laser radar, including adjusting the scanning angle of the laser radar to a suitable range to cover all directions of the converter station, and at the same time, set the scanning resolution to ensure high-precision data capture. High-precision laser radar is used for three-dimensional scanning to cover every corner of the converter station to obtain the basic point cloud data set. In this process, the laser radar repeatedly scans the preset path to ensure the integrity and accuracy of the data, and records the environmental conditions such as temperature and humidity at each scanning to facilitate correction in subsequent data analysis.

[0066] S102: Based on the basic point cloud data set, analyze the density and reflection characteristics of the point cloud, and filter out the noise to identify and remove data points with abnormal reflection intensity and position deviation from the normal, and the process of obtaining the noise-filtered point cloud data is as follows:

[0067] Based on the obtained basic point cloud data set, the point cloud density distribution diagram is used to analyze the data point density of each region, the low density region is identified and excluded by setting the density threshold, and the low density region is usually incomplete due to scanning angle or shielding. The reflection characteristics of the point cloud are analyzed, and the points are marked as abnormal by setting the reflection intensity threshold, such as greater than 800 units. The entire point cloud is processed using a Gaussian blur filter to remove environmental noise and device errors. For data points deviating from the normal position, their spatial relationship with surrounding points is compared, and data points exceeding the preset spatial distance threshold (for example, 5 cm) are removed to obtain clean point cloud data after noise filtering.

[0068] S103: Based on the point cloud data filtered by noise, the coordinates of the data points are adjusted to eliminate the scanning generated offset, and the continuity of the point cloud data is verified by continuity detection to obtain the processed point cloud data. The process is as follows:

[0069] Based on the point cloud data filtered by noise, the coordinates of the data points are adjusted. According to the position and scanning angle information of the laser radar, the accurate position of each data point in the global coordinate system is calculated, and the coordinates of the data points are adjusted using the conversion matrix to eliminate the coordinate deviation caused by the movement or rotation of the scanning device. The continuity detection is performed by establishing a distance matrix between points to analyze the distance between each point and its nearest neighbor point, for example, if the distance is less than 3 cm, the data is considered continuous, and the continuity of the point cloud data is verified. Ensure that there is no obvious jump or break between data points to ensure accurate reconstruction of the converter station model and form the processed point cloud data.

[0070] Please refer to Figure 3 , based on the processed point cloud data, the point cloud data is spatially compressed, each point in the data is mapped to the center of the adjacent grid, the average value of the coordinates of the points in each grid is calculated, and the result point is set as the grid representative. The steps of obtaining the simplified point cloud data are as follows:

[0071] S201: Based on the processed point cloud data, spatial compression is performed, and the point cloud data is mapped to the preset spatial grid according to the spatial position of each point to obtain the grid mapping point cloud data. The process is as follows:

[0072] Based on the processed point cloud data, a three-dimensional spatial grid is designed according to the actual structure and spatial characteristics of the converter station. The size and shape of the grid are pre-set to adapt to the spatial distribution of each part of the converter station. For example, the volume of each grid can be set to 3 cubic centimeters to ensure sufficient resolution and processing efficiency. The point cloud data is mapped into these pre-set spatial grids, and each data point is assigned to the corresponding grid according to its coordinates. Through this mapping, the original point cloud data is effectively compressed, and the point cloud collection within each grid reflects the geometric and spatial characteristics of the region, resulting in grid-mapped point cloud data, while recording the original reflection characteristics and color information of each point to ensure the integrity of the compressed data.

[0073] S202: Based on the grid-mapped point cloud data, the coordinate average value of each grid in the point cloud collection is calculated, and the grid coordinate average value calculation result is obtained.

[0074] Based on the grid-mapped point cloud data, the coordinate average value of each grid in the point cloud collection is calculated according to the formula:

[0075]

[0076] and

[0077]

[0078] and

[0079] ;

[0080] The grid coordinate average value is calculated. In the formula, , , represent the average values of x, y, z coordinates, represent the total number of points in the point cloud collection, , , represent the x, y, z coordinates of the th point.

[0081] Formula details and formula calculation derivation process:

[0082] In point cloud data processing, calculating the average value of coordinates is a basic operation to determine the center position or mean position of all points in each grid. For each grid, it is set to contain points, each point has three-dimensional coordinates , , . By summing up the coordinates of the same dimension of all points and dividing by the total number of points , the average coordinate of that dimension can be calculated. The process is performed separately for x, y, and z coordinate dimensions.

[0083] For demonstration calculation, set the coordinates of the following points in a grid:

[0084] Point 1: (2, 3, 5).

[0085] Point 2: (4, 1, 6).

[0086] Point 3: (3, 4, 4).

[0087] Calculate the average of the x coordinates:

[0088] ;

[0089] Similarly, the average of the y coordinates:

[0090] ;

[0091] The average of the z coordinates:

[0092] ;

[0093] The process demonstrates how to determine the center or mean position of point cloud data in each grid through mathematical operations.

[0094] S203: Based on the grid coordinate average calculation result, set the coordinate average as the representative point of each grid, and obtain the simplified point cloud data process is as follows:

[0095] Based on the coordinate average calculation result of each grid, the point cloud data is simplified. Set the coordinate average of each grid as the representative point of the grid, so that only one point data of each grid is retained and other points are removed, reducing the size of the data set while retaining the key information of the structure. Simplified point cloud data facilitates more advanced analysis and processing, such as modeling and visualization. In the simplified point cloud data, the selection of each grid representative point is based on geometric centrality and statistical importance of data points, ensuring the representativeness and accuracy of the data.

[0096] Referring to Figure 4 , based on the simplified point cloud data, the point cloud data in contact with the ground is screened, the simulated cloth model is created, the point cloud data is inverted, the position of the simulated cloth is adjusted to fit the ground represented by the point cloud, the ground and non-ground point clouds are segmented, and the ground segmentation data is obtained. The steps are as follows:

[0097] S301: Based on the simplified point cloud data, by analyzing the Z coordinate of each point and the predetermined ground height difference, the data points in contact with the ground are screened, the point set close to and equal to the ground height is identified, and the ground point cloud information is obtained. The process is as follows:

[0098] Based on the simplified point cloud data, ground point cloud recognition and screening are performed. For example, a predetermined ground height of 0 cm is defined, and based on this, the Z coordinate of each data point is checked to calculate the height difference from the ground. A height difference threshold, such as ±4 cm, is set to identify all points close to the ground height. All data points in contact or almost in contact with the ground are screened out to form a new point cloud dataset, i.e., ground point cloud information. The dataset includes all ground or near-ground points, and the selection of points is based on the comparison of Z coordinates with the predetermined ground height to ensure that only points close to the ground are selected.

[0099] S302: Based on the ground point cloud information, simulate the physical properties of the cloth, set the elastic coefficient and grid size of the cloth model, simulate the covering behavior of the cloth in the virtual environment, adjust the cloth to adapt to the terrain, and obtain the simulated cloth configuration process as follows:

[0100] Based on the obtained ground point cloud information, the physical properties of the cloth are simulated to adapt to the terrain. The elastic coefficient of the cloth model is set, for example, to 0.05 Newton per square meter, to simulate the elastic behavior of real cloth. The size of the cloth grid is set to 5 cm x 5 cm to ensure that the model can accurately cover complex terrain. In the virtual environment, the cloth is adjusted to naturally cover the terrain. By gradually adjusting the position and shape of the cloth grid, the cloth model is completely fitted to the terrain, simulating the behavior of the cloth reflecting the real physical properties such as stretching and compression, ensuring the practicality and accuracy of the simulation results.

[0101] S303: Based on the simulated cloth configuration, match the cloth with the inverted point cloud, adjust the position of the cloth model, and fit the cloth model to the inverted terrain features. Through the contact points between the cloth and the terrain, the ground and non-ground areas are segmented to obtain the ground segmentation data. The process of S302: simulation and configuration of the cloth model.

[0102] Based on the simulated cloth configuration, the cloth model is matched with the inverted terrain features. The terrain point cloud data is inverted, i.e., the Z coordinate value of each point is reversed, to simulate the natural behavior of the cloth sagging. The position of the cloth model is adjusted to ensure that the cloth can accurately cover the inverted terrain features. Through this matching, the contact points between the cloth and the terrain are used to segment the ground and non-ground areas. At each contact point, the contact between the cloth and the terrain is accurately recorded, and these contact points are used as boundaries to clearly separate the ground and non-ground areas. Through geometric and physical matching, the accuracy and practicality of the ground segmentation data are ensured.

[0103] Please refer to Figure 5, based on the ground segmentation data, the point cloud data is horizontally cut, and the point cloud is divided into multiple horizontal layers according to the preset height interval, and each layer of point cloud is projected onto a two-dimensional plane, the linear features of the wires on each layer are identified, the point cloud of the wire part is marked, and the step of obtaining the wire recognition point cloud data is specifically:

[0104] S401: Based on the ground segmentation data, the entire point cloud dataset is processed by height interval layering, each layer has a fixed height, and the point cloud is cut into multiple horizontal layers according to the preset interval, and the layer segmentation point cloud is obtained. The process is specifically:

[0105] Based on the ground segmentation data, the point cloud dataset is processed by height interval layering. For example, set the fixed height of each layer to 10 cm, and layer the entire point cloud dataset according to the standard. Use a preset height interval, for example, start from the ground height of 0 cm, and set each 10 cm as a layer until the highest point is reached. In this way, the entire point cloud dataset is cut into multiple horizontal layers, each layer containing all point cloud data within that height range. The point cloud data is organized into multiple independent layers, each uniformly covering a specific height range, resulting in layer segmentation point cloud. Such layering helps subsequent detailed analysis and data processing.

[0106] S402: Based on the layer segmentation point cloud, the point cloud of each horizontal layer is projected onto a two-dimensional plane, and the XY coordinates of the point cloud are plotted on a two-dimensional plane to obtain two-dimensional projection data. The process is specifically:

[0107] Based on the obtained layer segmentation point cloud, the point cloud of each horizontal layer is projected onto a two-dimensional plane. Select the XY coordinates of each point and plot these coordinates on a two-dimensional plane. During the process, the Z coordinate of each point is ignored and only the horizontal distribution is considered. In this way, each layer of point cloud is converted into a two-dimensional graph, forming two-dimensional projection data. The two-dimensional graph reflects the distribution characteristics of each layer of point cloud in the horizontal plane, facilitating further graphical analysis and feature extraction.

[0108] S403: Based on the two-dimensional projection data, a linear feature detection technique is used to analyze the linear arrangement features of the projection graph, and the projection point cloud with linear features is selected, and the point set with linear arrangement consistent with the characteristics of the wire is marked. Obtain the wire recognition point cloud data.

[0109] The linear feature detection technique is according to the formula:

[0110] ;

[0111] Calculate the linear feature score of the point cloud data, where, represents the length of the longest linear feature in the projection data, represents the average distance between adjacent features, Standard deviation of linear feature direction, , and are adjustment coefficients to balance the influence of length, spacing, and direction deviation on the overall recognition result. The comprehensive consideration of parameters helps to accurately define which point cloud data belongs to the power line, thereby improving selectivity and accuracy, adjusts the weight of the longest linear feature length in recognition, adjusts the influence of adjacent feature spacing on the result, and balances the influence of direction deviation on recognition accuracy. represents the linear feature score of point cloud data, used to determine which point cloud set has a significant linear arrangement consistent with the power line feature. By calculating this score, power lines can be accurately distinguished from other non-linear structures.

[0112] Formula: ;

[0113] Formula details and formula calculation derivation process:

[0114] In the formula, represents the linear feature score of point cloud data, used to determine which point cloud set has a significant linear arrangement consistent with the power line feature. By calculating this score, power lines can be accurately distinguished from other non-linear structures.

[0115] represents the length of the longest linear feature observed in the projected data. For example, through point cloud data analysis, it is determined that the longest straight line length of a potential power line feature is 10 meters.

[0116] is the average spacing of adjacent features, which represents the spatial distribution between linear features in power line detection. For example, from point cloud data, it is calculated that the average spacing of power lines is 0.5 meters.

[0117] represents the standard deviation of the linear feature direction, measuring the consistency of the linear feature in direction. For example, through directional analysis, it is determined that the standard deviation of the power line direction is 5 degrees.

[0118] , and are adjustment coefficients, for example, set to 0.7, 0.2, and 0.1 respectively. The selection of coefficients is based on the relative importance of linear features in power line detection. is higher because the longest feature length is the most critical in recognition; and are lower, indicating that spacing and direction deviation are important but have less direct impact on the result.

[0119] Formula calculation derivation process:

[0120] , , , m, m, degrees.

[0121] Substitute the formula calculation:

[0122] ;

[0123] The linear feature score calculated is 9.07, indicating that the identified linear feature is strong and has the typical characteristics of a wire. Based on this score, this point cloud data can be labeled as a wire. Please refer to

[0124] , based on the wire identification point cloud data, through spatial statistical analysis, measure the density between point clouds, use density benchmark for cluster analysis, mark the densely aggregated area as potential obstacles, and according to the color characteristics of point cloud, classify the obstacles, get the steps of obstacle classification label: Figure 6 S501: Based on the wire identification point cloud data, use kernel density estimation method to perform spatial statistical analysis on each point cloud block, calculate the number and distribution of points inside the point cloud block, identify the density of each point cloud area, and generate point cloud density information.

[0125] Kernel density estimation method, according to the formula:

[0126]

[0127] ;

[0128] Calculate the point cloud density , where represents the point cloud density, represents the number of points in the point cloud block, represents the position of the th point, is the average position of the point cloud block, is the standard deviation of the point position, is the base of the natural logarithm, is the ratio of the circumference to the diameter.

[0129] Formula:

[0130] ;

[0131] Formula explanation and formula calculation derivation process:

[0132] ​: Number of points within a point cloud block, directly counted from the dataset. For example, there are 200 points counted in a particular point cloud block.

[0133] : Position of each point in the point cloud, usually defined by three-dimensional spatial coordinates, e.g., which can be three-dimensional coordinate values directly obtained from laser scanning equipment.

[0134] : Average position of the point cloud block, calculated by taking the arithmetic mean of all point coordinates. For example, if all point coordinates are summed and divided by the number of points , the average coordinate value is obtained.

[0135] : Standard deviation of point positions, obtained by calculating the square root of the average of the squares of the deviations of each point's coordinates from the average position. This indicates the dispersion of points in the point cloud relative to the average position.

[0136] Specific example:

[0137] , average position 12 meters, standard deviation 6 meters, point positions range from 6 meters to 18 meters.

[0138] Calculation process:

[0139] ;

[0140] To simplify the calculation process, use and to calculate the of two extreme cases:

[0141] ;

[0142] ;

[0143] Approximate calculation:

[0144] ;

[0145] The results show that the point cloud density is , which is a relatively low density value, indicating that the points inside this point cloud area are relatively dispersed and may not belong to power lines or other densely arranged linear structures.

[0146] S502: Based on the point cloud density information, by comparing with the preset density benchmark, the point cloud area exceeding the preset density benchmark is identified as a potential obstacle, and the process of marking the obstacle area data is as follows:

[0147] Based on the generated point cloud density information, by comparing with the preset density benchmark, the point cloud area exceeding the preset density benchmark is identified as a potential obstacle. The process includes comparing the density of each point cloud block with the benchmark value, identifying all the exceeding areas, and marking the areas as potential obstacle areas. This marking helps to quickly identify and respond to possible obstacles in subsequent maintenance and safety operations, and obtains the marked obstacle area data.

[0148] S503: Based on the marked obstacle area data, analyze the color information of the point cloud area, and classify the obstacles according to the color characteristics of the point cloud, including metal, wood and plastic, and the process of obtaining the obstacle classification label is as follows:

[0149] Based on the marked obstacle area data, analyze the color information of the point cloud area. According to the RGB color value of each point in the point cloud data, classify the obstacles. Analyze the color characteristics of the point cloud in each marked area, and classify the obstacles according to the typical color range, including metal (usually gray or silver white), wood (brown or yellow), and plastic (may show a variety of bright colors). The color characteristics of each material type are set in advance to ensure the accuracy of classification, and a specific material type label is assigned to each obstacle area to generate the obstacle classification label.

[0150] Please refer to Figure 7 , based on the obstacle classification label, calculate the geometric center point and the outer rectangular bounding box of each marked obstacle, build the bounding box of each obstacle through three-dimensional space modeling, and locate and represent the spatial position and morphology of the obstacle, and the steps of obtaining the obstacle bounding box result are as follows:

[0151] S601: Based on the obstacle classification label, identify the geometric center of the cluster by calculating the average value of each cluster point cloud coordinate, and take the center point as the representative of the obstacle position, and obtain the obstacle center point coordinate.

[0152] Based on the obstacle classification label, calculate the average value of each cluster point cloud coordinate according to the formula:

[0153]

[0154] and

[0155]

[0156] and

[0157]

[0158] The geometric center of the cluster is calculated, where , , represent the average values of x, y, z coordinates, represent the total number of points in the cluster, , , represent the x, y, z coordinates of the th point.

[0159] Formula details and formula calculation derivation process:

[0160] The key step in identifying the location of the obstacle is to determine the geometric center of the cluster point cloud, usually by calculating the average of the coordinates. For example, a cluster contains points, each with three-dimensional coordinates , , . The average value of each coordinate dimension can be obtained by summing the coordinate values of each point and then dividing by the total number of points .

[0161] Suppose a cluster contains the following points:

[0162] Point 1: (3, 2, 7).

[0163] Point 2: (5, 3, 8).

[0164] Point 3: (4, 5, 6).

[0165] Calculate the average value of x coordinate:

[0166] ;

[0167] Similarly, the average value of y coordinate:

[0168] ;

[0169] The average value of z coordinate:

[0170] ;

[0171] The calculation method accurately determines the geometric center of the obstacle, which is the basis for positioning and processing obstacles. The process relies on the actual point cloud data.

[0172] S602: Based on the coordinates of the obstacle center point, measure the distance from each cluster's farthest point to the center point, identify the size and position of each obstacle's bounding rectangle, and obtain the obstacle bounding box parameter flow:

[0173] Based on the obstacle center point coordinates, the distance from the farthest point in each cluster to the center point is measured to identify the size and position of the bounding box of each obstacle. By calculating the Euclidean distance of each point to the center point, the farthest point is found, and the distance is defined as the radius of the bounding box. Considering the shape of the obstacle, the length, width and height of the bounding box are determined by calculating the maximum and minimum coordinates of each dimension. The bounding box parameters of each obstacle are obtained, including its position and size in three-dimensional space.

[0174] S603: Based on the obstacle bounding box parameters, a three-dimensional bounding box of each obstacle is constructed by three-dimensional modeling, each obstacle is covered by the bounding box, and the position and shape of the bounding box in space are recorded to obtain the bounding result of the obstacle. The process is as follows:

[0175] Based on the obtained obstacle bounding box parameters, three-dimensional modeling is performed to construct a three-dimensional bounding box for each obstacle. A standard cube or cuboid model is used to adjust the size and position of the bounding box according to the bounding box parameters, ensuring that the bounding box completely covers the corresponding obstacle cluster. In this way, each obstacle is surrounded by an accurate three-dimensional bounding box, which records the specific position and shape of the obstacle in space, providing necessary data for visualization and further processing of the obstacle. Such bounding results are crucial information for obstacle management and space planning.

[0176] The above is only the preferred embodiment of the present application, and does not limit the form of the present application in other ways. Any skilled person in the art can use the above disclosed technical content to make changes or modifications as equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made according to the technical essence of the present application without departing from the technical solution content of the present application shall be within the protection scope of the present application.

Claims

1. A method for obstacle bounding based on laser point cloud in a converter station, characterized in that, The method comprises the following steps: Based on the converter station environment, three-dimensional scanning is performed by a laser radar, point cloud data is collected, noise filtering is performed on the data, spatial alignment of the data is performed, and processed point cloud data is obtained; Based on the processed point cloud data, each point in the data is mapped to the center of the adjacent grid, the average value of the coordinates of the points in each grid is calculated, the result point is set as the grid representative, and simplified point cloud data is obtained; Based on the simplified point cloud data, the point cloud data in contact with the ground is screened, a simulated cloth model is created, the point cloud data is inverted, the position of the simulated cloth is adjusted to fit the ground represented by the point cloud, the ground and non-ground point clouds are segmented, and ground segmentation data is obtained; Based on the ground segmentation data, the point cloud data is horizontally cut, each layer of point cloud is projected onto a two-dimensional plane, the linear features of the wires on each layer are identified, and wire identification point cloud data is obtained; Specifically: Based on the ground segmentation data, the entire point cloud data set is processed by height interval layering, each layer has a fixed height, and the point cloud is divided into multiple horizontal layers according to a preset interval, and layer segmentation point cloud is obtained; Based on the layer segmentation point cloud, the point cloud of each horizontal layer is projected onto a two-dimensional plane, the XY coordinates of the point cloud are drawn on the two-dimensional plane, and two-dimensional projection data is obtained; Based on the two-dimensional projection data, a linear feature detection technique is used to analyze the linear arrangement features of the projection graph, the projection point cloud with linear features is screened, the point set with linear arrangement consistent with the characteristics of the wire is marked, and wire identification point cloud data is obtained; The linear feature detection technique is according to the formula: ； Computing linear feature scores for point cloud data , wherein the linear feature scores are used to accurately distinguish power lines from other non-linear structures, represents the length of the longest linear feature in the projected data, represents the average spacing between adjacent features, represents the standard deviation of the linear feature direction, 、 and are adjustment coefficients used to balance the influence of length, spacing, and direction deviation on the recognition result; Based on the wire identification point cloud data, the density between the point clouds is measured, the region with dense aggregation is marked as a potential obstacle, the obstacle is classified, and an obstacle classification label is obtained; Based on the obstacle classification label, the geometric center point and the circumscribed rectangular bounding box of each marked obstacle are calculated, a bounding box of each obstacle is constructed, and an obstacle bounding result is obtained.

2. The method of claim 1, wherein the method further comprises: The processed point cloud data includes the position coordinates, reflection intensity and color information of the point cloud, the simplified point cloud data includes the voxelized point cloud, the average position point inside each voxel and the point cloud model with reduced spatial resolution after reconstruction, the ground segmentation data includes the ground point cloud identified after the simulated cloth is fitted, the non-ground point cloud part and the threshold value between the ground and the non-ground, the wire identification point cloud data includes the identified wire feature points and the point cloud line segment of the wire, and the obstacle classification label includes the divided obstacle categories and the corresponding obstacle clustering label. The obstacle bounding result includes the three-dimensional coordinates, length-width-height size and obstacle volume size of each obstacle bounding box.

3. The method of claim 1, wherein the method further comprises: Based on the converter station environment, three-dimensional scanning is performed by a laser radar, point cloud data is collected, noise filtering is performed on the data, spatial alignment of the data is performed, and processed point cloud data is obtained, which comprises the following steps: Based on the converter station environment, the scanning angle and resolution of the laser radar are adjusted, panoramic three-dimensional scanning of the converter station is performed, and a basic point cloud data set is obtained; Based on the basic point cloud data set, by analyzing the density and reflection characteristics of the point cloud, and filtering the data noise, the data points with abnormal reflection intensity and deviating from the normal position are identified and removed, and the point cloud data filtered by noise is obtained; Based on the point cloud data filtered by noise, the coordinates of the data points are adjusted to eliminate the offset generated by scanning, and the continuity of the point cloud data is verified through continuity detection, and the processed point cloud data is obtained.

4. The method of claim 1, wherein, Based on the processed point cloud data, each point in the data is mapped to the center of the adjacent grid, the average value of the coordinates of the points in each grid is calculated, and the result point is set as the representative of the grid, and the simplified point cloud data is obtained. Based on the processed point cloud data, spatial compression is performed, and the point cloud data is mapped into a preset spatial grid according to the spatial position of each point, and the grid mapping point cloud data is obtained. Based on the grid mapping point cloud data, the average value of the coordinates in each grid point cloud set is calculated, and the grid coordinate average value calculation result is obtained. Based on the grid coordinate average value calculation result, the average value of the coordinates is set as the representative point of each grid, and the simplified point cloud data is obtained.

5. The method of claim 1, wherein, Based on the simplified point cloud data, the point cloud data in contact with the ground is screened, a simulated cloth model is created, the point cloud data is inverted, the position of the simulated cloth is adjusted to fit the ground represented by the point cloud, and the ground and non-ground point clouds are segmented, and the ground segmentation data is obtained. Based on the simplified point cloud data, the data points in contact with the ground are screened by analyzing the Z coordinate of each point and the predetermined ground height difference, and the point set close to and equal to the ground height is identified, and the ground point cloud information is obtained. Based on the ground point cloud information, the physical properties of the cloth are simulated, the elastic coefficient and grid size of the cloth model are set, the coverage behavior of the cloth in the virtual environment is simulated, and the cloth is adjusted to adapt to the terrain, and the simulated cloth configuration is obtained. Based on the simulated cloth configuration, the cloth is matched with the inverted point cloud, the position of the cloth model is adjusted, the cloth model is fitted to the inverted terrain features, the ground and non-ground areas are segmented through the contact points of the cloth and the terrain, and the ground segmentation data is obtained.

6. The method of claim 1, wherein, Based on the wire recognition point cloud data, the density between the point clouds is measured, the densely gathered area is marked as a potential obstacle, the obstacle is classified, and the obstacle classification label is obtained. Based on the wire recognition point cloud data, the kernel density estimation method is used to perform spatial statistical analysis on each point cloud block, the number and distribution of points inside the point cloud block are calculated, the density of each point cloud area is identified, and the point cloud density information is generated. Based on the point cloud density information, by comparing with the preset density benchmark, the point cloud area exceeding the preset density benchmark is identified as a potential obstacle, and the marked obstacle area data is obtained. Based on the marked obstacle area data, the color information of the point cloud area is analyzed, the obstacles are classified into corresponding types according to the color characteristics of the point cloud, including metal, wood and plastic, and the obstacle classification label is obtained.

7. The method of claim 6, wherein the method further comprises: The kernel density estimation method is according to the formula: ； Computing point cloud density wherein, representing point cloud density, representing the number of points in a point cloud block, representing the position of the th point, is the average position of the point cloud block, is the standard deviation of the point positions, is the base of the natural logarithm, is the mathematical constant pi.

8. The method of claim 1, wherein, Based on the obstacle classification label, the geometric center point and the circumscribed rectangular bounding box of each labeled obstacle are calculated, the bounding box of each obstacle is constructed, and the obstacle bounding result is obtained. The steps are as follows: Based on the obstacle classification label, the average value of each clustered point cloud coordinate is calculated, the geometric center of the cluster is identified, and the center point is taken as the representative of the obstacle position. The obstacle center point coordinate is obtained. Based on the obstacle center point coordinate, the distance from each cluster farthest point to the center point is measured, the size and position of the circumscribed rectangular bounding box of each obstacle are identified, and the obstacle bounding box parameter is obtained. Based on the obstacle bounding box parameter, a three-dimensional model is constructed to build a three-dimensional bounding box for each obstacle. The bounding box covers each obstacle and records the position and shape of the bounding box in space, and the obstacle bounding result is obtained.

Images