Substation construction site three-dimensional point cloud moving target identification and early warning method

Abstract

The invention discloses a transformer substation construction site three-dimensional point cloud moving target identification and early warning method. The method comprises the steps of obtaining transformer substation construction site three-dimensional point cloud data in real time based on a three-dimensional laser radar; denoising the point cloud of each frame, and filtering out outliers; two frames of processed point cloud data are selected, and a projection frame difference method is adopted to obtain the spatial position of a dynamic target; segmenting different dynamic target point clouds into independent targets; matching with a template based on the morphological characteristics of the target so as to identify the type of the dynamic target; extracting a center coordinate or a bounding box of the dynamic target, marking the position of the center coordinate or the bounding box in the three-dimensional space, and outputting the position in real time; calculating the distance between the dynamic target and the high-voltage live equipment; when the distance is smaller than a set threshold value, safety early warning is triggered. According to the method, the defects in the prior art are overcome, the recognition efficiency of the dynamic target on the construction site of the transformer substation can be effectively improved, and safety management and control and intelligent early warning of construction site operating personnel, engineering vehicles and the like are realized based on the recognized target.

Description

Technical Field

[0001] The present invention relates to the technical field of power operation safety control, and particularly relates to a method for identifying and warning moving targets in a three-dimensional point cloud of a substation construction site. Background Art

[0002] In recent years, with the rapid development of the power industry, the construction scale and complexity of substations have been gradually increasing. The construction site usually includes dynamic targets such as hoisting equipment, engineering vehicles, and operating personnel, and is accompanied by a large number of high-voltage live equipment. This environment poses strict requirements on the safety and efficiency of the construction process.

[0003] Traditional monitoring means mainly rely on video monitoring and manual patrols, and have the following deficiencies:

[0004] (1) Difficulty in identifying dynamic targets: Video monitoring is difficult to accurately capture dynamic targets in complex scenarios, especially in the presence of occlusion or light changes.

[0005] (2) Delayed safety warning: It is impossible to detect the distance between dynamic targets and high-voltage equipment in real time, and the warning mechanism often relies on manual judgment, which may lead to delays.

[0006] (3) High consumption of computing resources: Directly processing the frame difference of three-dimensional point clouds requires point-by-point calculation, which is inefficient and has high requirements for hardware resources, making it difficult to meet real-time requirements. Summary of the Invention

[0007] Aiming at the deficiencies of the prior art, the present invention provides a method for identifying and warning moving targets in a three-dimensional point cloud of a substation construction site, which overcomes the deficiencies of the prior art, can effectively improve the identification efficiency of dynamic targets in a substation construction site, and realizes safety control and intelligent warning for operating personnel, engineering vehicles, etc. at the construction site based on the identified targets.

[0008] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0009] A method for identifying and warning moving targets in a three-dimensional point cloud of a substation construction site includes the following steps:

[0010] Step S1: Perform real-time three-dimensional scanning of the substation construction site based on a three-dimensional lidar; obtain the three-dimensional point cloud data of the substation construction site within a continuous time period; denoted as:

[0011] P t ={(x i ,y i ,z i )|i=1,2,L,N}

[0012] where N is the number of points in the point cloud, and Pt The point cloud of the t-th frame, where (x i , y i , z i ) are the coordinates of the points therein;

[0013] Step S2: Denoise each frame of the point cloud to filter out the outlier points; the processed point cloud is denoted as P t ′;

[0014] Step S3: Select two frames of point cloud data processed in Step S2 and use the projection frame difference method to obtain the spatial position of the dynamic target;

[0015] Step S4: Merge the three-plane information, back-project the changing area of the three-plane frame difference into the three-dimensional point cloud coordinate system to determine the possible dynamic target point cloud area P dynamic ;

[0016] Step S5: Dynamic target segmentation and recognition, perform clustering analysis on P dynamic to segment different dynamic target point clouds into independent targets; based on the morphological features of the targets and template matching to identify the types of dynamic targets;

[0017] Step S6: Calculate the coordinates of the dynamic target, extract the center coordinates or bounding box of the dynamic target, mark its position in the three-dimensional space, and output in real time:

[0018]

[0019] where N d is the number of points in the dynamic target point cloud; Position is the center coordinate of the dynamic target; x i , y i , z i are the three-dimensional coordinates of the i-th point in the dynamic target point cloud; ∑ i x i is the sum of the x coordinates of all points of the dynamic target; ∑ i y i is the sum of the y coordinates of all points of the dynamic target; ∑ z z i is the sum of the z coordinates of all points of the dynamic target;

[0020] Step S7: Safety warning, calculate the distance between the dynamic target and the high-voltage live equipment:

[0021]

[0022] In the formula, d is the spatial distance from the dynamic target to the specified equipment (such as high-voltage equipment); x1, y1, z1 are the center coordinates of the dynamic target; x2, y2, z2 are the three-dimensional coordinates of the specified equipment;

[0023] When the distance d is less than the set threshold, a safety warning is triggered.

[0024] Preferably, in the step S2, the methods for denoising the point cloud include statistical filtering and radius filtering.

[0025] Preferably, in the step S3, the specific steps for obtaining the spatial position of the dynamic target by using the projection frame difference method include:

[0026] Step S31: Data acquisition, acquiring two frames of three-dimensional point cloud data after denoising processing in step S2, denoted as P t ′ and P t ′ +1 , where:

[0027] P t ′ = {(x i , y i , z i ) | i = 1, 2, L, N}

[0028] P′ t+1 = {(x i , y i , z i ) | i = 1, 2, L, N}

[0029] Among them, N is the number of points in the point cloud, P t ′ is the point cloud at time t, P t ′ +1 is the point cloud at time t + 1, (x i , y i , z i ) is the coordinate of the point therein;

[0030] Step S32: Calculate the three-dimensional point cloud projection, project the three-dimensional point cloud data onto the xy, yz, and xz planes respectively to generate two-dimensional projection plane point clouds:

[0031]

[0032]

[0033] Perform the same projection on the (t + 1)-th frame P t ′ +1 to respectively obtain

[0034] Step S33: Calculate the projection frame difference, calculate the frame difference on each projection plane respectively:

[0035]

[0036] Extract the dynamic change area of the point cloud through the frame difference to obtain the projection position of the dynamic target.

[0037] Preferably, in the step S5, the specific steps of dynamic target segmentation and recognition include:

[0038] Step S51: Perform target segmentation by clustering analysis. Use the DBSCAN clustering algorithm to segment the point cloud in the dynamic point cloud region P dynamic into multiple independent segmentation clusters;

[0039] Step S52: Extract target geometric features. Extract geometric features from each segmentation cluster to assist in target classification;

[0040] Step S53: Perform target recognition by template matching. According to the extracted geometric features, match the target with known types;

[0041] Step S54: Results of dynamic target classification. Each segmentation cluster Gi is classified into a specific target type; then output the classification results.

[0042] Preferably, in the step S51, the input parameters of the DBSCAN clustering algorithm are:

[0043] ∈: Neighborhood radius, used to define the density range of each point;

[0044] MinPts: Minimum number of neighborhood points, used to determine whether a point is a core point;

[0045] The steps of the DBSCAN clustering algorithm include:

[0046] Step S511: Traverse each point p in P dynamic If the number of neighborhood points of p ≥ MinPts, mark it as a core point; otherwise, mark it as a boundary point or a noise point;

[0047] Step S512: Group the points with connected core point densities into the same cluster;

[0048] Step S513: Assign boundary points to the cluster of the nearest core point;

[0049] Step S154: Obtain multiple segmentation clusters G1, G2,..., Gk of the dynamic point cloud, where each segmentation cluster is an independent target point cloud.

[0050] Preferably, in the step S52, the geometric features extracted from each segmentation cluster include: bounding box features, point cloud density, shape features, and dynamic features.

[0051] Preferably, the step S53 of performing target recognition by template matching specifically includes the following steps:

[0052] Step S531: Establish a target template library;

[0053] Step S532: Feature matching, calculating the similarity between the target feature and the template:

[0054]

[0055] where ω i is the feature weight; is the i-th eigenvalue of the target; is the i-th eigenvalue of the template;

[0056] Step S533: Select the template with the highest similarity and identify the target type.

[0057] Preferably, step S7 specifically includes the following steps:

[0058] Step S71: Identify fixed power equipment based on template matching;

[0059] Step S72: Identify dynamic targets and extract outer contours;

[0060] Step S73: Calculate the safety distance and give early warnings.

[0061] Preferably, step S71 specifically includes the following steps:

[0062] Step S711: Build a template library for fixed power equipment;

[0063] Build a template library for each type of equipment, denoted as T = {T1, T2, …, T m}, where Ti is the template feature of the i-th type of equipment;

[0064] Step S712: Template matching;

[0065] Search for the area in the point cloud that matches the template,

[0066] Point cloud segmentation: Perform static background segmentation on the scene point cloud to extract the point cloud P static ,

[0067] Calculate the similarity: Calculate the similarity between the target area and the template Ti:

[0068]

[0069] In the formula, ω1, ω2, ω3 are weights; Volume is the bounding box volume; Shape is the shape feature; Surface is the surface curvature;

[0070] Match the area P with the highest similarity and exceeding the threshold to the corresponding equipment type of the template Ti, and record the equipment position (x, y, z) and the outer contour.

[0071] Preferably, the step S72 specifically includes the following steps:

[0072] Step S721: Dynamic target extraction

[0073] Extract the dynamic point cloud P through the frame difference method dynamic ; Use the DBSCAN clustering algorithm to segment P dynamic into independent target point clouds G1, G2, …, Gk;

[0074] Step S722: Target outer contour extraction

[0075] Calculate the minimum axis-aligned bounding box (AABB) or oriented bounding box (OBB) of each dynamic target,

[0076] Record the length, width, height, center position, and surface points of the outer contour Oi.

[0077] Preferably, the step S73 specifically includes the following steps:

[0078] Step S731: Distance calculation

[0079] For each dynamic target Oi and each fixed power device Ej, calculate the closest distance d between their outer contours:

[0080] d = min{||p o - p e |||| p o ∈ O i , p e ∈ E j}

[0081] In the formula, p o is a point on the outer contour of the dynamic target; p e is a point on the outer contour of the power device;

[0082] Step S732: Safety distance judgment

[0083] Set a safety threshold D threshold , if d < D threshold , then trigger an alarm and output the number, type of the dynamic target Oi, and its closest distance to the device Ej.

[0084] The present invention provides a method for identifying and warning moving targets in the three-dimensional point cloud of a substation construction site. It has the following beneficial effects:

[0085] (1) Improve the efficiency of dynamic target detection; by using the projection frame difference method to reduce the dimension of the three-dimensional point cloud, the computational complexity is greatly reduced, and the real-time processing ability of large-scale point cloud data is significantly improved, solving the bottleneck in the computational resource requirements of traditional methods.

[0086] (2) Improve the accuracy of target recognition; by segmenting dynamic targets through the DBSCAN clustering algorithm and combining geometric and motion feature extraction with template matching, different target types can be accurately distinguished, adapting to the diverse dynamic targets in the substation construction scenario.

[0087] (3) Enhance the safety warning ability; based on the spatial distance calculation of the target outer contour and the real-time monitoring mechanism of the present invention, the accurate distance measurement between the dynamic target and the high-voltage equipment is realized. When the distance is lower than the set threshold, the safety warning is immediately triggered, effectively reducing the construction risk.

[0088] (4) Adapt to complex scenarios; by enhancing the anti-noise performance and robustness during the denoising, dimensionality reduction, and feature extraction processes, the present invention can operate stably under complex working conditions with occlusion, illumination changes, and environmental noise. Description of the Drawings

[0089] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the prior art.

[0090] Figure 1 The step flow block diagram of the present invention. Detailed Embodiments

[0091] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention with reference to the drawings in the present invention.

[0092] Embodiment

[0093] As Figure 1 shown, a method for identifying and warning moving targets of three-dimensional point clouds at a substation construction site includes the following steps:

[0094] Step S1: Conduct real-time three-dimensional scanning of the substation construction site based on a three-dimensional lidar; obtain the three-dimensional point cloud data of the substation construction site within a continuous time period; denoted as:

[0095] P t ={(x i , y i , z i )|i = 1, 2, L, N}

[0096] where N is the number of points in the point cloud, P t is the t-th frame of the point cloud, and (x i , y i , z i ) are the coordinates of the points therein;

[0097] Step S2: Denoise each frame of the point cloud to filter out outliers; the processed point cloud is denoted as P t ′;

[0098] Specifically, the methods for point cloud denoising include statistical filtering and radius filtering.

[0099] Statistical filtering removes points that exceed a specific threshold by calculating the mean distance of the neighborhood of each point. It specifically includes the following steps:

[0100] (1) For each point p i ∈P t , search for its k-neighborhood points;

[0101] (2) Calculate the mean distance d i of the point to the neighborhood center;

[0102] (3) Set a threshold to remove points that satisfy d i >μ + ασ, where μ is the mean, σ is the standard deviation, and α is a coefficient (such as 1.5 or 2, etc.).

[0103] Radius filtering removes points with insufficient number of points based on the statistics of the number of points in the neighborhood of each point. It specifically includes the following steps:

[0104] (1) For each point p i , count the number of neighborhood points n i within a radius r.

[0105] (2) Remove points where the number of neighborhood points n i is less than the threshold.

[0106] Step S3: Select two frames of point cloud data processed in Step S2, and use the projection frame difference method to obtain the spatial position of the dynamic target;

[0107] Since the substation construction site involves the hoisting of electrical equipment, personnel working at heights, etc., it is necessary to timely judge the distances between the hanging parts, engineering vehicles, personnel, etc. and the high-voltage live equipment, and conduct safety warnings. Performing direct frame difference based on the three-dimensional point cloud requires calculating the distance of each point in the three-dimensional point cloud of the substation construction site or comparing the point cloud features, which requires high computing resources when processing the large-scale point cloud at the construction site, and it is difficult to ensure real-time performance. At the same time, noise points or measurement errors will interfere with the difference calculation, which may lead to false detections. In scenes with small changes or severe occlusion, direct frame difference cannot effectively separate the target. Therefore, in the present invention, by adopting the projection frame difference method, the three-dimensional point cloud is projected onto multiple two-dimensional planes (such as the xy, yz, xz planes), and the frame differences are calculated respectively. The advantages of adopting the projection frame difference are as follows:

[0108] (1) Dimension reduction reduces the computational complexity: The three-dimensional point cloud is decomposed into data on multiple two-dimensional planes, and the computational complexity of point cloud frame difference calculation is reduced. It is particularly suitable for processing large-scale point clouds and improves real-time performance.

[0109] (2) Strong anti-noise performance: The frame difference of the projected point cloud ignores some insignificant three-dimensional details, reducing the impact of noise points.

[0110] (3) Highlight the main features: After projecting onto a two-dimensional plane, it is easier to capture the lateral or longitudinal motion features of dynamic targets. In a complex environment, the projected frame difference is more likely to separate dynamic targets.

[0111] Specifically, the specific steps of obtaining the spatial position of the dynamic target by using the projected frame difference method include the following:

[0112] Step S31: Data acquisition, obtain two frames of three-dimensional point cloud data after denoising processing in step S2, denoted as P t ′ and P′ t+1 , where:

[0113] P t ′ = {(x i , y i , z i )|i = 1, 2, L, N}

[0114] P t ′ +1 = {(x i , y i , z i )|i = 1, 2, L, N}

[0115] Among them, N is the number of points in the point cloud, P t ′ is the point cloud at time t, P′ t+1 is the point cloud at time t + 1, (x i , y i , z i ) is the coordinate of the point;

[0116] Step S32: Calculate the three-dimensional point cloud projection, project the three-dimensional point cloud data onto the xy, yz, and xz planes respectively to generate two-dimensional projected plane point clouds:

[0117]

[0118] Perform the same projection on the (t + 1)-th frame P t ′ +1 to obtain respectively

[0119] Step S33: Calculate the projected frame difference, calculate the frame difference on each projected plane respectively:

[0120]

[0121] Extract the dynamic change area of the point cloud through the frame difference to obtain the projected position of the dynamic target.

[0122] Step S4: Merge the information of the three planes, back-project the changed regions of the three-plane frame difference to restore them to the three-dimensional point cloud coordinate system, and determine the possible dynamic target point cloud region P dynamic ;

[0123] P dynamic = ΔP xy IΔP yz IΔP xz

[0124] In the formula, P dynamic is the three-dimensional point cloud set of the dynamic target extracted from the three-plane frame difference. ∩ is the set intersection operation. The dynamic target needs to have changes on the xy, yz, and xz planes simultaneously. By merging the information of the three planes, the consistent points can be screened out from the dynamic change regions of different projection planes to restore the three-dimensional information of the dynamic target.

[0125] Step S5: Dynamic target segmentation and recognition. Perform clustering analysis on P dynamic to segment the target point clouds of different dynamic regions extracted from the three-dimensional point cloud into independent targets; based on the morphological features of the targets (such as point cloud density, shape) and template matching to identify the types of dynamic targets (such as hanging parts, engineering vehicles, operators);

[0126] Specifically, the dynamic target segmentation and recognition specifically include the following steps:

[0127] Step S51: Perform target segmentation by clustering analysis. Use the DBSCAN clustering algorithm to segment the point cloud in the dynamic point cloud region P dynamic into multiple independent segmentation clusters;

[0128] Among them, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering algorithm does not need to specify the number of clusters, can identify targets with irregular shapes, and naturally process noise points.

[0129] The input parameters are:

[0130] ∈: Neighborhood radius, used to define the density range of each point;

[0131] MinPts: Minimum number of neighborhood points, used to determine whether a point is a core point;

[0132] The steps of the DBSCAN clustering algorithm include:

[0133] (S511): Traverse each point p in P dynamic . If the number of neighborhood points of p ≥ MinPts, then mark it as a core point; otherwise, mark it as a boundary point or a noise point;

[0134] (S512): Group the points connected by the core point density into the same cluster (target);

[0135] (S513): Assign the boundary points to the cluster of the nearest core point;

[0136] (S154): Obtain multiple segmented clusters G1, G2, …, Gk of the dynamic point cloud, where each segmented cluster Gi is an independent target point cloud.

[0137] Step S52: Target geometric feature extraction, extract geometric features from each segmented cluster to assist in target classification;

[0138] Among them, the geometric features mainly include:

[0139] (1) Bounding box feature

[0140] 1. Axis-aligned bounding box (AABB):

[0141] Definition: The minimum and maximum boundaries of the target point cloud in the x, y, and z directions.

[0142] Calculation: Calculate the point coordinates in Gi respectively:

[0143] Min(x), Min(y), Min(z), Max(x), Max(y), Max(z)

[0144] The axis-aligned bounding box feature can be used to estimate the size and volume of the target.

[0145] ② Oriented bounding box (OBB):

[0146] Definition: Consider the main direction of the target point cloud to make the bounding box more compact.

[0147] Calculation: Determine the main direction of the target through principal component analysis (PCA).

[0148] (2) Point cloud density

[0149] Calculation: The ratio of the number of points in the target point cloud to the volume of the bounding box:

[0150]

[0151] Among them, |G i | is the number of target points, and V bbox is the volume of the bounding box.

[0152] The point cloud density can be used to distinguish sparse targets (such as hanging parts) from dense targets (such as operators).

[0153] (3) Shape features

[0154] Aspect ratios: L / W and W / H are used to judge the planar structure of the target (e.g., a vehicle is rectangular).

[0155] Surface area to volume ratio: Reflects the compactness of the target.

[0156] (4) Dynamic features

[0157] Compare the movement trajectories of the target in different frames:

[0158]

[0159] where (x1, y1, z1) and (x2, y2, z2) are the center coordinates of the target in two frames.

[0160] Step S53: Perform target recognition through template matching. According to the extracted geometric features, match the target with known types;

[0161] Specifically, performing target recognition through template matching specifically includes the following steps:

[0162] (S531): Establish a target template library; Define templates: Establish templates for geometric features and dynamic features for each possible target type (such as operators, engineering vehicles, lifting parts, etc.).

[0163] For example: Operator: Small bounding box, obvious height feature.

[0164] Engineering vehicle: Large aspect ratio, smooth movement trajectory.

[0165] Lifting part: Low point cloud density, obvious dynamic features.

[0166] (S532): Feature matching, calculate the similarity between the target features and the template:

[0167]

[0168] where ω i is the feature weight; is the i-th eigenvalue of the target; is the i-th eigenvalue of the template.

[0169] Step S54: Results of dynamic target classification. Each segmented cluster Gi is classified into a specific target type.

[0170] Output the classification results. Specifically include:

[0171] Target number: Idi; Target type: such as "operator", "engineering vehicle"; Position coordinates: center point of the target (x, y, z). Real-time status: dynamic features such as speed and direction.

[0172] Step S6: Calculate the coordinates of the dynamic target, extract the center coordinates or bounding box of the dynamic target, mark its position in the three-dimensional space, and output in real-time:

[0173]

[0174] where N d is the number of points in the point cloud of the dynamic target; Position is the center coordinates of the dynamic target; x i , y i , z i is the three-dimensional coordinates of the i-th point in the point cloud of the dynamic target; ∑ i x i is the sum of the x coordinates of all points of the dynamic target; ∑ i y i is the sum of the y coordinates of all points of the dynamic target; ∑ z z i is the sum of the z coordinates of all points of the dynamic target.

[0175] By calculating the coordinates of the dynamic target and extracting the center coordinates or bounding box of the dynamic target, the geometric center position of the point cloud of the dynamic target can be calculated for target positioning.

[0176] Step S7: Safety warning, calculate the distance between the dynamic target and the high-voltage energized equipment:

[0177]

[0178] In the formula, d is the spatial distance from the dynamic target to the specified equipment (such as high-voltage equipment); x1, y1, z1 are the center coordinates of the dynamic target; x2, y2, z2 are the three-dimensional coordinates of the specified equipment;

[0179] When the distance d is less than the set threshold, a safety warning is triggered.

[0180] Specifically, the present invention for identifying fixed power equipment based on template matching and performing safety distance warning for dynamic targets includes the following steps:

[0181] Step S71: Identify fixed power equipment based on template matching;

[0182] (1) Build a template library for fixed power equipment;

[0183] Power equipment usually has specific shape features. Therefore, the template library is established through the following steps:

[0184] Collect point clouds of power equipment: Use lidar to scan common power equipment such as transformers, switch cabinets, cable racks, etc., and extract their point cloud features.

[0185] Extract geometric features:

[0186] ① Bounding box parameters (length, width, height).

[0187] ② Shape features (aspect ratio, volume).

[0188] ③ Surface features (surface area, curvature distribution).

[0189] Define templates: Use the above features to construct a template library for each type of power equipment, denoted as T = {T1, T2, …, T m}, where Ti is the template feature of the i-th type of equipment.

[0190] (2) Template matching;

[0191] Search for regions in the point cloud that match the template,

[0192] Point cloud segmentation: Perform static background segmentation on the scene point cloud to extract the point cloud P of the fixed equipment area static ,

[0193] Calculate similarity: Calculate the similarity between the target region and the template Ti:

[0194]

[0195] In the formula, ω1, ω2, ω3 are weights; Volume is the bounding box volume; Shape is the shape feature; Surface is the surface curvature;

[0196] Match the region P with the highest similarity and exceeding the threshold to the corresponding equipment type of the template Ti, and record the equipment position (x, y, z) and the outer contour.

[0197] Step S72: Dynamic target recognition and outer contour extraction;

[0198] (1) Dynamic target extraction:

[0199] Extract the dynamic point cloud P through the frame difference method dynamic ; Use the DBSCAN clustering algorithm to segment P dynamic into independent target point clouds G1, G2, …, Gk.

[0200] (2) Target outer contour extraction:

[0201] Calculate the minimum axis-aligned bounding box (AABB) or oriented bounding box (OBB) of each dynamic target,

[0202] Record the length, width, height, center position, and surface points of the outer contour Oi.

[0203] Step S73: Calculate the safety distance and give an early warning;

[0204] (1) Distance calculation

[0205] For each dynamic target Oi and each fixed power equipment Ej, calculate the closest distance d between their outer contours:

[0206] d = min{||p o - p e || | p o ∈ O i , p e ∈ E j}

[0207] In the formula, p o is a point on the outer contour of the dynamic target. p e is a point on the outer contour of the power equipment.

[0208] (2) Safety distance judgment

[0209] Set a safety threshold D threshold , if d < D threshold , then trigger an early warning, and output the number, type of the dynamic target Oi, and its closest distance to the equipment Ej. Generate an alarm message, such as "The crane is approaching the transformer, the current distance is 5 meters".

[0210] Specifically, the example of the early warning output is as follows:

[0211] Power equipment identification result:

[0212] Equipment type: Transformer.

[0213] Position coordinates: (10.2, 5.3, 2.1).

[0214] Outer contour: Length = 3.2m, width = 2.0m, height = 2.5m.

[0215] Dynamic target and early warning result:

[0216] Dynamic target: Crane.

[0217] Distance: 5 meters (below the safety threshold of 6 meters).

[0218] Early warning: The crane is approaching the transformer, please stop the operation immediately!

[0219] The present invention constructs a full-process intelligent control system from target detection to classification and recognition to early warning output, realizes the global monitoring and accurate early warning of dynamic targets at the construction site, and provides technical support for the intelligent and safe management of power projects.

[0220] The three-dimensional point cloud is dimensionally reduced by the projection frame difference method, thereby greatly reducing the computational complexity, significantly improving the real-time processing ability of large-scale point cloud data, and solving the bottleneck in the computational resource requirements of traditional methods. The dynamic targets are segmented by the DBSCAN clustering algorithm, combined with geometric and motion feature extraction and template matching, which can accurately distinguish different target types and adapt to diverse dynamic targets in the substation construction scenario. In addition, the present invention realizes the accurate distance measurement between the dynamic target and the high-voltage equipment based on the spatial distance calculation and real-time monitoring mechanism of the outer contour of the target. When the distance is lower than the set threshold, a safety warning is immediately triggered, effectively reducing the construction risk. The present invention enhances the noise resistance and robustness during the denoising, dimensional reduction, and feature extraction processes, so that the present invention can also operate stably under complex working conditions with occlusion, illumination changes, and environmental noise.

[0221] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A mobile target recognition and early warning method for a three-dimensional point cloud at a substation construction site, characterized in that: The following steps are involved: Step S1: Perform real-time three-dimensional scanning of the substation construction site based on a three-dimensional laser radar to obtain three-dimensional point cloud data of the substation construction site within a continuous time period; Denoted as: P t ={(x i ,y i ,z i )|i=1,2,L,N} Where N is the number of points in the point cloud, P t The point cloud of the tth frame, (x i ,y i ,z i ) are the coordinates of its midpoint; Step S2: De-noise each frame of point cloud and filter out outliers; the processed point cloud is recorded as P t ′; Step S3: Select the two frames of point cloud data processed in step S2, and use the projection frame difference method to obtain the spatial position of the dynamic target; Step S4: Merge the three-plane information, back-project the change area of ​​the three-plane frame difference, restore it to the three-dimensional point cloud coordinate system, and determine the possible dynamic target point cloud area P dynamic ; Step S5: Dynamic target segmentation and recognition, for P dynamic Perform cluster analysis to segment different dynamic target point clouds into independent targets; match the target’s morphological features with the template to identify the type of dynamic target; Step S6: Calculate the dynamic target coordinates, extract the center coordinates or bounding box of the dynamic target, mark its position in the three-dimensional space, and output in real time: Where N d is the number of points in the dynamic target point cloud; Position is the center coordinate of the dynamic target; x i ,y i ,z i is the three-dimensional coordinate of the i-th point in the dynamic target point cloud; ∑ i x i is the sum of the x coordinates of all points of the dynamic target; ∑ i y i is the sum of the y coordinates of all points of the dynamic target; ∑ z z i is the sum of the z coordinates of all points of the dynamic target; Step S7: Safety warning, calculate the distance between the dynamic target and the high-voltage live equipment: In the formula, d is the spatial distance from the dynamic target to the specified device; x1, y1, z1 are the center coordinates of the dynamic target; x2, y2, z2 are the three-dimensional coordinates of the specified device; When the distance d is less than the set threshold, a safety warning is triggered.

2. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 1 is characterized by: In step S2, the method of denoising the point cloud includes statistical filtering and radius filtering.

3. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 1 is characterized in that: In step S3, the specific steps of using the projection frame difference method to obtain the spatial position of the dynamic target include: Step S31: Data acquisition, obtaining two frames of 3D point cloud data after denoising in step S2, denoted as P and t ′ and P′ t+1 ,in: P t ′={(x i ,y i ,z i )|i=1,2,L,N} P′ t+1 ={(x i ,y i ,z i )|i=1,2,L,N} Where N is the number of points in the point cloud, P t ′ is the point cloud at time t, P t ' +1 is the point cloud at time t+1, (x i ,y i ,z i ) are the coordinates of its midpoint; Step S32: Calculate the three-dimensional point cloud projection, project the three-dimensional point cloud data onto the xy, yz, and xz planes respectively, and generate a two-dimensional projection plane point cloud: P t xy ={(x i ,y i )|(x i ,y i ,z i )∈P t ′} P t yz ={(y i ,z i )|(x i ,y i ,z i )∈P t ′} P t xz ={(x i ,z i )|(x i ,y i ,z i )∈P t ′} For frame t+1, P t ' +1 Perform the same projection to obtain Step S33: Calculate the projection frame difference, and calculate the frame difference on each projection plane: The dynamically changing area of ​​the point cloud is extracted through frame difference to obtain the projection position of the dynamic target.

4. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 1 is characterized by: In step S5, the specific steps of dynamic target segmentation and recognition include: Step S51: Cluster analysis is used to segment the target. The DBSCAN clustering algorithm is used to cluster the dynamic point cloud area P dynamic The point cloud in is segmented into multiple independent segmentation clusters; Step S52: extracting target geometric features, extracting geometric features from each segmentation cluster to assist target classification; Step S53: template matching is used to identify the target, and the target is matched with a known type according to the extracted geometric features; Step S54: Dynamic target classification result, each segmentation cluster Gi is classified into a specific target type; and the classification result is output.

5. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 4 is characterized by: In step S51, the input parameters of the DBSCAN clustering algorithm are: ∈: neighborhood radius, used to define the density range of each point; MinPts: minimum number of neighborhood points, used to determine whether a point is a core point; The steps of the DBSCAN clustering algorithm include: Step S511: traverse P dynamic For each point p in , if the number of neighboring points of p is ≥ MinPts, it is marked as a core point; otherwise, it is marked as a boundary point or a noise point; Step S512: grouping the points that are densely connected to the core points into the same cluster; Step S513: classify the boundary points into the core point cluster closest to them; Step S154: obtaining multiple segmentation clusters G1, G2, ..., Gk of the dynamic point cloud, where each segmentation cluster is an independent target point cloud.

6. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 4, characterized in that: In step S52, the geometric features extracted from each segmentation cluster include: bounding box features, point cloud density, shape features and dynamic features.

7. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 4, characterized in that: In step S53, template matching is used to perform target recognition. The following steps are involved: Step S531: Establish a target template library; Step S532: feature matching, calculating the similarity between the target feature and the template: Among them, ω i is the feature weight; f i target is the i-th eigenvalue of the target; f i template is the i-th eigenvalue of the template; Step S533: Select the template with the highest similarity and identify the target type.

8. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 1, characterized in that: The step S7 specifically includes the following steps: Step S71: Identify fixed power equipment based on template matching; Step S72: dynamic target recognition and outer contour extraction; Step S73: Calculate the safety distance and give warning.

9. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 8, characterized in that: The step S71 specifically includes the following steps: Step S711: construct a fixed power equipment template library; Construct a template library for each type of equipment, denoted as T = {T1, T2, L, T m }, where Ti is the template feature of the i-th type of equipment; Step S712: template matching; Search the point cloud for regions that match the template, Point cloud segmentation: Perform static background segmentation on the scene point cloud and extract the fixed device area point cloud P static , Calculate similarity: Calculate the similarity between the target area and the template Ti: Where ω1, ω2, ω3 are weights; Volume is the bounding box volume; Shape is the shape feature; Surface is the surface curvature; The area P with the highest similarity and exceeding the threshold is matched to the corresponding device type of the template Ti, and the device position (x, y, z) and outer contour are recorded.

10. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 8, characterized in that: The step S72 specifically includes the following steps: Step S721: Dynamic target extraction Extract dynamic point cloud P by frame difference method dynamic ; Use DBSCAN clustering algorithm to cluster P dynamic Segment into independent target point clouds G1, G2, …, Gk; Step S722: Target outer contour extraction Calculate the minimum axis-aligned bounding box (AABB) or oriented bounding box (OBB) for each dynamic object. Record the length, width, height, center position and surface points of the outer contour Oi.

11. The method for mobile target recognition and early warning of three-dimensional point cloud at a substation construction site according to claim 10, characterized in that: The step S73 specifically includes the following steps: Step S731: Distance calculation For each dynamic target Oi and each fixed power equipment Ej, the shortest distance d between their outer contours is calculated: d=min{||p o -p e |||p o ∈O i ,p e ∈E j } In the formula, p o is the outer contour point of the dynamic target; p e It is the outer contour point of the power equipment; Step S732: Safety distance determination Set safety threshold D threshold , if d<D threshold , an early warning is triggered, and the number, type and the closest distance between the dynamic target Oi and the device Ej are output.

Images