Method, system and device for reconstructing three-dimensional scene of power transmission corridor and medium

Abstract

The invention discloses a method, a system and equipment for reconstructing a three-dimensional scene of a power transmission corridor, and a medium. The method comprises the following steps: acquiring a multi-view image and first point cloud data of the power transmission corridor; performing feature extraction on the multi-view image to obtain a multi-view feature map; inputting the multi-view feature map into a depth multi-view stereoscopic model, and calculating to obtain a disparity map; generating second point cloud data according to the disparity map, and reconstructing a three-dimensional model of the power transmission corridor in combination with the first point cloud data and the second point cloud data; and performing obstacle detection and classification on the three-dimensional model of the power transmission corridor to obtain a detection result, and generating a risk assessment report according to the detection result. According to the method, the precision and efficiency of three-dimensional scene reconstruction of the power transmission corridor can be improved, the accuracy of obstacle detection and risk assessment is enhanced, the safety and management efficiency of operation and maintenance of the power transmission corridor can be improved, and the operation and maintenance cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method, system, device and medium for reconstructing a three-dimensional scene of a power transmission corridor. Background Technology

[0002] As the "main artery" of power transmission, power transmission corridors play a crucial role in ensuring power supply through their safety and stability. Traditional methods of inspection and maintenance of power transmission corridors, such as manual inspections and periodic remote sensing data collection, are not only inefficient and costly, but also fail to meet the stringent requirements for power supply reliability. Against this backdrop, the synergistic development of drones, remote sensing technology, and 3D reconstruction technology, utilizing drones equipped with various sensors to collect data and combining it with advanced 3D reconstruction algorithms, has become a new direction for monitoring and maintaining power transmission corridors, bringing new possibilities for improving the efficiency of transmission corridor operation and maintenance and ensuring a safe power supply.

[0003] Existing 3D reconstruction and monitoring technologies for power transmission corridors still have many shortcomings. On the one hand, accuracy and adaptability are significant issues. In the complex environment of power transmission corridors, traditional multi-view stereo reconstruction (MVS) and LiDAR technologies struggle to balance accuracy and scene adaptability. When faced with occlusion, sparse textures, complex lighting, and weather conditions, traditional MVS methods are prone to errors in image matching and disparity estimation, leading to decreased reconstruction accuracy. While LiDAR technology can provide high-precision data, its high equipment cost and poor modeling of irregular or complex power facilities, trees, and other obstacles result in weak adaptability to specific environments. On the other hand, real-time performance, automation, and robustness are lacking. Most existing 3D reconstruction methods rely heavily on manual intervention and complex data processing workflows, making real-time monitoring and large-scale application impossible. When faced with environmental disturbances such as severe weather, lighting changes, and object occlusion, existing technologies lack robustness, resulting in unstable data acquisition and reconstruction results. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method and system for reconstructing a three-dimensional scene of a power transmission corridor, which solves the problems of insufficient accuracy, efficiency, and automation of traditional three-dimensional reconstruction methods in power transmission corridor applications, as well as poor adaptability to complex environments, high data processing costs, and lack of intelligent optimization.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for reconstructing a three-dimensional scene of a power transmission corridor, comprising:

[0008] Acquire multi-view images and first point cloud data of the power transmission corridor;

[0009] Feature extraction is performed on the multi-view images to obtain multi-view feature maps;

[0010] The multi-view feature map is input into the depth multi-view stereo model to calculate and obtain the disparity map;

[0011] Based on the disparity map, second point cloud data is generated, and combined with the first and second point cloud data, a three-dimensional model of the power transmission corridor is reconstructed.

[0012] Obstacle detection and classification are performed on the three-dimensional model of the power transmission corridor to obtain detection results, and a risk assessment report is generated based on the detection results.

[0013] This invention acquires multi-view images and point cloud data, combines them with deep convolutional neural networks and deep multi-view stereo models, generates disparity maps, optimizes and fuses point cloud data to reconstruct a high-precision 3D model; it can improve the accuracy and efficiency of 3D scene reconstruction of power transmission corridors.

[0014] As a preferred embodiment of the method for reconstructing a three-dimensional scene of a power transmission corridor according to the present invention, wherein: feature extraction is performed on the multi-view images to obtain a multi-view feature map, including:

[0015] A deep convolutional neural network model is used to perform convolutional pooling operations on the multi-view images to extract key feature points, and a multi-view feature map is generated based on the key feature points.

[0016] The beneficial effects of this preferred technical solution are as follows: by using a deep convolutional neural network model to perform convolutional pooling operations to extract key feature points and generate multi-view feature maps, it can automatically learn and accurately extract key features in the image, providing a more accurate and effective data foundation for subsequent disparity map calculations and other steps, thereby improving the accuracy and reliability of 3D reconstruction.

[0017] As a preferred embodiment of the method for reconstructing a three-dimensional scene of a power transmission corridor according to the present invention, the method includes: inputting the multi-view feature map into a depth multi-view stereo model and calculating and obtaining a disparity map, including:

[0018] The depth multi-view stereo model calculates the disparity value of each pixel in the input multi-view feature map to generate a preliminary disparity map. The preliminary disparity map is then geometrically optimized to obtain the final disparity map.

[0019] As a preferred embodiment of the method for reconstructing a three-dimensional scene of a power transmission corridor according to the present invention, wherein: generating second point cloud data based on the disparity map includes:

[0020] Based on the disparity map and combined with the acquisition device parameters, the disparity map is converted into a depth map;

[0021] The pixels in the depth map are converted into three-dimensional spatial points to generate second point cloud data. As a preferred embodiment of the method for reconstructing a three-dimensional scene of a power transmission corridor according to the present invention, the method involves: combining the first point cloud data and the second point cloud data to reconstruct a three-dimensional model of the power transmission corridor, including:

[0022] The second point cloud data was optimized using the Poisson surface reconstruction algorithm;

[0023] By combining the optimized second point cloud data and the first point cloud data, the point cloud data is converted into a 3D network using the triangulation method;

[0024] The three-dimensional network is smoothed to reconstruct a three-dimensional model of the power transmission corridor.

[0025] The beneficial effects of this preferred technical solution are as follows: using the Poisson surface reconstruction algorithm to optimize the second point cloud data, and combining it with the first point cloud data through triangulation and smoothing to reconstruct the three-dimensional model, can effectively reduce point cloud data noise, optimize the model surface, make the reconstructed three-dimensional model of the power transmission corridor more in line with the real scene, and improve the model quality and visualization effect.

[0026] As a preferred embodiment of the method for reconstructing a three-dimensional scene of a power transmission corridor according to the present invention, the method includes: performing obstacle detection and classification on the three-dimensional model of the power transmission corridor to obtain detection results, and generating a risk assessment report based on the detection results, including:

[0027] Point cloud data is segmented using Euclidean clustering algorithm to extract geometric features of the three-dimensional model of the power transmission corridor, and obstacles are identified based on these geometric features.

[0028] The obstacles are classified using a support vector machine classifier to obtain the detection results.

[0029] As a preferred embodiment of the method for reconstructing a three-dimensional scene of a power transmission corridor according to the present invention, it further includes:

[0030] Calculate the distance between obstacles and transmission lines, and assess the risk level using a weighted scoring method;

[0031] The risk levels are divided into low risk, medium risk, and high risk, and different risk assessment reports are automatically generated based on the risk level.

[0032] Secondly, the present invention provides a system for reconstructing a three-dimensional scene of a power transmission corridor, comprising: a data acquisition module for acquiring multi-view images and first point cloud data of the power transmission corridor;

[0033] The feature extraction module is used to extract features from the multi-view images to obtain multi-view feature maps;

[0034] The image calculation module is used to input the multi-view feature map into the depth multi-view stereo model and calculate and obtain the disparity map;

[0035] The 3D model generation module is used to generate second point cloud data based on the disparity map, and to reconstruct a 3D model of the power transmission corridor by combining the first point cloud data and the second point cloud data.

[0036] The prevention and detection module is used to detect and classify obstacles in the three-dimensional model of the power transmission corridor, obtain the detection results, and generate a risk assessment report based on the detection results.

[0037] Thirdly, the present invention provides a computer device, comprising:

[0038] Memory and processor;

[0039] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of a method for reconstructing a three-dimensional scene of a power transmission corridor.

[0040] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method for reconstructing a three-dimensional scene of a power transmission corridor.

[0041] Compared with existing technologies, the beneficial effects of this invention are as follows: By acquiring multi-view images and point cloud data, combining deep convolutional neural networks and deep multi-view stereo models, this invention generates disparity maps, optimizes and fuses point cloud data to reconstruct a high-precision 3D model; it utilizes Euclidean clustering algorithms and support vector machine classifiers to accurately detect and classify obstacles, and scientifically assesses risk levels and generates reports based on distance calculation and weighted scoring methods. This improves the accuracy and efficiency of 3D scene reconstruction in power transmission corridors, enhances the accuracy of obstacle detection and risk assessment, improves the safety and management efficiency of power transmission corridor operation and maintenance, and reduces operation and maintenance costs. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1This is a schematic diagram of the overall process of a method for reconstructing a three-dimensional scene of a power transmission corridor according to an embodiment of the present invention. Detailed Implementation

[0044] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0045] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for reconstructing a three-dimensional scene of a power transmission corridor is provided, comprising:

[0046] S100: Acquire multi-view images and first point cloud data of the power transmission corridor;

[0047] S102: Extract features from multi-view images to obtain multi-view feature maps; input the multi-view feature maps into a depth multi-view stereo model and calculate and obtain a disparity map;

[0048] S104: Based on the disparity map, generate the second point cloud data, and combine the first point cloud data and the second point cloud data to reconstruct the three-dimensional model of the power transmission corridor;

[0049] S106: Perform obstacle detection and classification on the 3D model of the power transmission corridor, obtain the detection results, and generate a risk assessment report based on the detection results.

[0050] It should be noted that, as a special application scenario, the 3D reconstruction of power transmission corridors requires consideration not only of point cloud accuracy, but also of data acquisition efficiency, real-time reconstruction results, and adaptability to different environmental factors. While existing depth multi-view stereo methods perform well in standard environments, they may still encounter problems such as slow data processing speed, insufficient accuracy, or inability to meet real-time monitoring requirements when dealing with long-distance, complex, and dynamically changing power transmission corridors.

[0051] Therefore, the application of dense point cloud reconstruction technology based on deep multi-view stereo in 3D power transmission corridor scenes urgently needs targeted optimization and innovation. By combining deep learning with the specific environmental features of power transmission corridors, the accuracy and efficiency of 3D reconstruction can be effectively improved, especially when facing complex power facilities, trees, and other obstacles, providing more accurate modeling results.

[0052] This embodiment proposes a method for reconstructing a 3D scene of a power transmission corridor based on a deep multi-view stereo dense point cloud. The aim is to address the shortcomings of traditional 3D reconstruction methods in terms of accuracy, efficiency, and automation through optimization and targeted improvements to existing deep multi-view stereo algorithms, particularly in the complex environment of power transmission corridors. By introducing deep learning algorithms, image feature extraction, disparity prediction, and point cloud optimization can be automated, overcoming problems caused by manual operation, data sparsity, and environmental interference in traditional methods. This invention's method can also adapt to dynamic changes under different environmental conditions, providing high-precision and high-efficiency 3D modeling results.

[0053] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the above embodiment, a method for reconstructing a three-dimensional scene of a power transmission corridor is provided.

[0054] In this embodiment of the invention, in step S100, a drone equipped with a high-definition camera and a LiDAR sensor is used to collect multi-view images and first point cloud data of the power transmission corridor, and then the multi-view images and first point cloud data are preprocessed.

[0055] Specifically, the high-definition camera is configured with a 12MP resolution; the LiDAR sensor has a scanning radius of 100 meters and a scanning accuracy of 2 points per square meter; the drone's flight altitude is set to 80 meters to ensure that the images and point cloud data cover the entire corridor area.

[0056] Multiple flight paths were established along the power transmission corridor, each spaced 10 meters apart with a 25% overlap. This ensured precise lateral and longitudinal coverage of various trees, obstacles, and power facilities. The drone captured a high-resolution image every 10 meters, ensuring a 50%-70% overlap between images and sufficient overlap between different viewpoints for subsequent image matching and 3D reconstruction. The drone's real-time differential GPS system recorded the precise locations of multi-view images and initial point cloud data. Thirty high-resolution images were acquired, covering the entire corridor area.

[0057] In one alternative implementation, preprocessing may include quality assessment, geometric correction, illumination adjustment of multi-view images, and denoising, registration, simplification, and densification of the first point cloud data.

[0058] For example, when preprocessing multi-view images, 10 high-quality images are selected as inputs for subsequent processing based on sharpness, exposure, and lighting consistency.

[0059] The selected multi-view images are then subjected to distortion correction using a formula, which is expressed as follows:

[0060]

[0061] Where x,y are the coordinates in the uncorrected multi-view image, x′,y′ are the coordinates in the corrected multi-view image, and K is the camera's intrinsic parameter matrix.

[0062] To eliminate the impact of uneven illumination on subsequent feature extraction, a histogram equalization algorithm is used to adjust the brightness and contrast of multi-view images. The histogram equalization formula is expressed as:

[0063]

[0064] Where H(r) is the cumulative histogram of the multi-view image, p(i) is the proportion of pixels with brightness value i in the multi-view image, and r is the brightness range of the multi-view image.

[0065] For example, the first point cloud data preprocessing includes noise filtering, point cloud simplification, point cloud alignment, and registration. Specifically, a neighborhood-based algorithm is used to denoise and simplify the point cloud, removing outliers with low density. Point cloud data from different flight paths are globally registered to ensure accurate alignment of the various datasets.

[0066] In this embodiment of the invention, step S102 involves extracting features from the multi-view images to obtain a multi-view feature map, inputting the multi-view feature map into a depth multi-view stereo model, and calculating the disparity map, which includes the following steps A1-A2:

[0067] A1: Use a deep convolutional neural network model to perform convolutional pooling operations on multi-view images, extract key feature points, and generate multi-view feature maps based on the key feature points.

[0068] A2: The depth multi-view stereo model calculates the disparity value of each pixel in the input multi-view feature map, generates a preliminary disparity map, and performs geometric optimization on the preliminary disparity map to obtain the final disparity map.

[0069] In this embodiment of the invention, the specific process of extracting key feature points using a deep convolutional neural network model includes:

[0070] ① First convolutional layer: basic feature extraction. The main function of the first convolutional layer is to extract low-level features such as edges and textures.

[0071] ② Multi-layer convolution: High-level feature extraction. A deep convolutional neural network structure is used to extract high-order features. The first few layers mainly extract local features, such as edges and corners; the middle layers focus on texture and shape information; and the deep layers extract semantic-level features, such as power lines, towers, and trees. Then, accelerated convergence is performed to prevent gradient vanishing.

[0072] ③ Pooling layer: Max pooling is used to reduce computation, filter out redundant information, and extract the most salient features. Through multi-scale feature fusion, it preserves both global information and local details, making it particularly suitable for obstacle detection and depth estimation.

[0073] ④ Multi-scale feature fusion layer: This layer fuses features extracted at different scales to improve the accuracy of depth prediction. It is represented as follows:

[0074] F fusion =αF low +βF mid +γF high

[0075] Among them, F low For low-level features, F mid For mid-level features, F high These are high-level features, and α, β, and γ are weighting coefficients.

[0076] ⑤ Output layer: After the above feature extraction process, the final output multi-view feature map includes edge features, corner features, texture features, geometric features, category features, etc. The multi-view feature map is used for the next step of disparity calculation and depth information generation.

[0077] In one alternative implementation, the method for generating multi-view feature maps can also be scale-invariant feature transformation (SMT). SMT is a method for detecting and describing local features in an image, possessing scale, rotation, and illumination invariance. Specifically, the SMT method first constructs a difference-of-Gaussian pyramid in the multi-view image, determining the location and scale of feature points by finding extrema in different scale spaces. Then, it calculates the gradient direction histogram in the neighborhood of each feature point to generate a descriptor for the feature point. Finally, these feature points and descriptors are combined to form the multi-view feature map.

[0078] In another alternative implementation, the method for generating multi-view feature maps can also be an accelerated robust feature algorithm. This algorithm first uses a matrix to detect the location of feature points in the multi-view image, then calculates the wavelet responses around the feature points to generate descriptors for those points, thereby obtaining the multi-view feature map. Compared to scale-invariant feature transform algorithms, accelerated robust feature algorithms offer faster computation speed and better real-time performance.

[0079] In this embodiment of the invention, the process of obtaining a disparity map using a depth multi-view stereo model includes:

[0080] Ten multi-view feature maps from different perspectives were used as input for the depth multi-view stereo model. Each multi-view feature map captured different objects such as trees, ground, and water from different perspectives.

[0081] The depth multi-view stereo model calculates the disparity value for each pixel based on these input multi-view feature maps. The disparity value represents the distance from the camera to the object; a larger disparity indicates that the object is closer to the camera, while a smaller disparity indicates that the object is farther away.

[0082] The formula for obtaining the preliminary disparity map is expressed as follows:

[0083] D(x,y)=DeepMVS(I1,I2…,I n )

[0084] Where D(x,y) represents the depth information in the disparity map, and I1,I2,…,I n The input consists of multiple feature maps from multiple viewpoints.

[0085] After the initial disparity map is generated, optimization is performed using the geometric consistency between images to reduce errors and improve the accuracy of depth information. Depth correction is performed using the known geometry in the initial disparity map to ensure that the output depth information is more accurate, especially in areas with indistinct textures such as trees and water.

[0086] In this embodiment of the invention, step S104, generating second point cloud data based on the disparity map, and reconstructing the three-dimensional model of the power transmission corridor by combining the first and second point cloud data, includes the following steps B1-B3:

[0087] B1: Based on the disparity map and combined with the parameters of the acquisition device, the disparity map is converted into a depth map; the pixels in the depth map are converted into three-dimensional spatial points to generate the second point cloud data.

[0088] B2: Optimize the second point cloud data using the Poisson surface reconstruction algorithm.

[0089] B3: Combining the optimized second point cloud data and the first point cloud data, the point cloud data is converted into a three-dimensional network using the triangulation method; the three-dimensional network is then smoothed to reconstruct the three-dimensional model of the power transmission corridor.

[0090] In this embodiment of the invention, based on the disparity map and combined with high-definition camera parameters, the disparity map is converted into a depth map, and the conversion formula is expressed as:

[0091]

[0092] Where Z(x,y) is the depth value, f is the focal length of the high-definition camera, B is the distance between the cameras, and D(x,y) is the depth information in the disparity map;

[0093] In this embodiment of the invention, depth information is obtained through a depth map, and combined with the camera's intrinsic and extrinsic parameters, triangulation is used to convert the depth information of each pixel into three-dimensional coordinates. The triangulation calculation formula is expressed as follows:

[0094]

[0095] Where P represents a point in the 3D point cloud, (x,y) are the pixel coordinates in the image, and c x c y f represents the principal point coordinates of the camera. x f y Let be the focal length of the camera, and D(x,y) be the depth information in the disparity map.

[0096] In an alternative implementation, the moving least squares method can also be used to optimize the second point cloud data. The moving least squares method estimates the local surface information of each point by performing a polynomial fit within its neighborhood. When processing the second point cloud data, for each point, the algorithm searches for points in its neighborhood, and then, based on the distribution of these neighborhood points, uses a polynomial function to fit the optimal position of the point, thereby smoothing the point cloud data, removing noise, and preserving the geometric features of the point cloud.

[0097] In another alternative implementation, the specific method for converting point cloud data into a 3D network using triangulation can be the wavefront method. The wavefront method first requires determining the boundary points of the point cloud data. Then, starting from the boundary, an initial triangle is used as the wavefront, and new triangles are continuously added to the leading edge of the wavefront until the entire point cloud region is covered by triangles. During the process of adding new triangles, appropriate points are selected to form new triangles based on the geometric information and topology of the point cloud, ensuring the quality of the generated 3D network, avoiding the formation of narrow or irregular triangles, and simultaneously achieving a good fit to the surface shape of the point cloud data, providing a high-quality mesh foundation for subsequent 3D model reconstruction.

[0098] In this embodiment of the invention, step S106 involves obstacle detection and classification of the three-dimensional model of the power transmission corridor to obtain detection results. Generating a risk assessment report based on these results includes the following steps C1-C3:

[0099] C1: The point cloud data is segmented using the Euclidean clustering algorithm to extract the geometric features of the 3D model of the power transmission corridor, and obstacles are identified based on the geometric features.

[0100] C2: Use a support vector machine classifier to classify obstacles and obtain detection results.

[0101] C3: Calculate the distance between obstacles and transmission lines, and use a weighted scoring method to assess the risk level; the risk level is divided into low risk, medium risk and high risk, and different risk assessment reports are automatically generated according to the risk level.

[0102] In this embodiment of the invention, the Euclidean clustering algorithm is expressed by the following formula:

[0103]

[0104] Wherein d(P i P j Let P be a point. i and point P j The distance between them, (x i ,y i ,z i Let P be a point. i The coordinates, (x j ,y j ,z j Let P be a point. j The coordinates;

[0105] If d(P) i P j If the values ​​are less than a set threshold, the two points are considered to belong to the same cluster. The point cloud is then divided into multiple point cloud clusters using the Euclidean clustering algorithm.

[0106] For each segmented point cloud cluster, obstacle recognition is performed by calculating the morphological features of each cluster, such as the volume, surface normal, and geometric features of the point cloud cluster.

[0107] Different objects have different geometric properties, so these features can be used to determine the object category of a point cloud cluster.

[0108] For example, point cloud clusters of trees typically exhibit significant height differences and relatively regular circular or elliptical shapes. By calculating the geometric features of the clusters, such as height, area, and convex hull, trees can be effectively identified.

[0109] For example, point cloud clusters of buildings and transmission towers often exhibit relatively regular straight-line characteristics, and these objects can be identified by calculating the surface normal direction and surface curvature of the point cloud clusters.

[0110] In this embodiment of the invention, the formula for calculating geometric features includes:

[0111] ① For each point cloud cluster, calculate the difference between the maximum and minimum heights of all points within the cluster, expressed as:

[0112] Δh = max(H) - min(H)

[0113] Where Δh represents the height difference, and H represents the height of all points within the point cloud cluster.

[0114] ② Calculate the convex hull area of ​​the point cloud cluster to assess the regularity of its shape. The convex hull area reflects the shape characteristics of the point cloud cluster. Generally, trees have a more rounded shape, while buildings and power transmission towers have regular geometric shapes.

[0115] In this embodiment of the invention, a support vector machine classifier is used to classify different types of obstacles based on the identified geometric features. The calculation formula is as follows:

[0116] f(x) =<w,x> +b

[0117] Where w is the normal vector of the hyperplane, x is the input feature, b is the bias, and f(x) is the predicted value.

[0118] Based on the extracted geometric features, such as the height difference of point cloud clusters, convex hull area, and surface normals, a support vector machine classifier is trained to classify obstacles such as trees, buildings, and power transmission towers.

[0119] After performing cluster analysis on the point cloud using the Euclidean clustering algorithm of this invention, a support vector machine classifier is then used to classify different types of obstacles. The specific locations of nine types of trees are successfully identified, and their distances from the power transmission lines are marked, with high classification accuracy.

[0120] In one alternative implementation, a deep learning-based semantic segmentation method can be used to identify and classify obstacles. Specifically, a convolutional neural network is used to process the point cloud data or corresponding 2D images of the 3D model of the power transmission corridor. First, a large amount of labeled data containing various obstacles, including trees, buildings, and transmission towers, is collected to train the deep learning model. During training, the model learns the feature patterns of different obstacles, extracts features through convolutional layers, reduces data dimensionality through pooling layers, restores spatial resolution through upsampling layers, and finally outputs the category label for each pixel or point. For new 3D model data of the power transmission corridor, it is input into the trained model, which automatically performs semantic segmentation on the data, identifying different types of obstacles and their locations.

[0121] In another alternative implementation, a region-growing-based method can be used to identify and classify obstacles. Specifically, suitable seed points are selected from the point cloud data of the 3D model of the power transmission corridor. Seed points can be points with distinctive features, such as points with abrupt changes in height or large curvature. Then, similarity criteria are defined, such as the similarity of features like distance, normal direction, and color between points. For each seed point, points in its neighborhood that satisfy the similarity criteria are searched, and these points are merged into the region where the seed point is located. Next, starting from the newly merged point, the search for points in the neighborhood continues, continuously expanding the region until no more points that meet the criteria can be merged. This process is repeated until all points are assigned to a region. Finally, the regions are classified according to their geometric features, thereby identifying different types of obstacles.

[0122] In this embodiment of the invention, the horizontal distance between each obstacle and the transmission line is calculated using the generated three-dimensional point cloud and the known location of the transmission line.

[0123] In one possible implementation, for vegetation such as trees, the growth of the trees needs to be considered to assess future risks. Based on information such as tree species, age, and season, the growth rate of the trees is predicted, and the potential encroachment on transmission corridor space is estimated. Tree growth models are used to predict the future height or diameter of the trees, thereby calculating the impact of tree growth on transmission lines.

[0124] For example, a tree growth model can be represented as:

[0125] H(t) = H0 + r*t

[0126] Where H(t) is the tree height at future time t, H0 is the initial height, r is the tree growth rate, and t is time.

[0127] In this embodiment of the invention, the degree of threat posed to the power transmission line by the obstacle is assessed based on the distance between the obstacle and the power transmission line, tree growth prediction, and characteristics such as the height and type of the obstacle.

[0128] In this embodiment of the invention, a weighted scoring method is used, assigning weights to each obstacle according to different parameters to obtain a comprehensive risk score. The risk level assessment formula is expressed as follows:

[0129] R risk =ω1*D height +ω2*D distance +ω3*D type

[0130] Among them, R risk For obstacle risk scoring, D height D is the height of the obstacle. distance D represents the distance between the obstacle and the power transmission line. type ω1, ω2, and ω3 represent the risk coefficients for obstacle types (such as trees, buildings, etc.), and ω1, ω2, and ω3 represent the weights of each parameter.

[0131] Based on the calculated risk score, obstacles are classified into low risk, medium risk, and high risk.

[0132] In one alternative implementation, low risk indicates that the obstacle is far from the transmission line and the trees are growing slowly. Medium risk indicates that the obstacle is close to the transmission line, but the trees are growing slowly. High risk indicates that the obstacle is close to the transmission line and the trees are growing rapidly, potentially causing future impacts.

[0133] For each identified obstacle, its distance from the transmission line was calculated. The calculations revealed that some trees were too close to the transmission line, posing a significant collision risk. A comprehensive risk score was calculated for each obstacle, showing that some trees posed a high risk; three high-risk obstacles were detected, requiring immediate pruning or relocation. Transmission towers and buildings posed a lower risk and generally did not require intervention.

[0134] In an optional implementation, step S106 involves obstacle detection and classification of the 3D model of the power transmission corridor to obtain detection results. The method for generating a risk assessment report based on these results can also be: utilizing a combination of point cloud feature histograms and a random forest classifier. Specifically, the point cloud data of the 3D model of the power transmission corridor is first processed to extract the feature histogram for each point cloud cluster. The feature histogram can encompass various information such as the spatial distribution, density variation, and normal vector direction distribution of the point cloud cluster, providing a more comprehensive description of its characteristics. For example, a normal vector direction histogram can be constructed by counting the number of points within different normal vector direction intervals, or a density histogram can be constructed by calculating the density of points within different spatial locations.

[0135] Then, these feature histograms are used as input data to train a random forest classifier. The random forest consists of multiple decision trees. During training, it randomly samples features and samples from the input data to build multiple decision tree models, and finally combines the results of these decision trees for classification prediction. After training, the feature histograms of the new 3D model point cloud data of the power transmission corridor are input into the trained random forest classifier, which can then identify different types of obstacles, such as trees, buildings, and power transmission towers, and output the detection results.

[0136] During the risk assessment phase, the distance between obstacles and transmission lines is calculated using the generated 3D point cloud and the known location of the transmission lines. For obstacles such as trees, in addition to considering distance, factors such as historical meteorological data and soil condition data can be combined to optimize the tree growth prediction model. For example, by incorporating information such as precipitation and sunshine duration from meteorological data, and parameters such as nutrient content and pH from soil data, a more accurate tree growth prediction model can be established through multiple linear regression or more complex machine learning algorithms.

[0137] In risk level assessment, fuzzy comprehensive evaluation is used instead of weighted scoring. Fuzzy comprehensive evaluation can better handle the fuzziness and uncertainty in the evaluation process. First, the set of evaluation factors is determined, including the distance between the obstacle and the transmission line, tree growth predictions, obstacle height, and obstacle type. Then, an evaluation level set is constructed, such as low risk, medium risk, and high risk. Next, the weights of each evaluation factor are determined, which can be done using methods such as the analytic hierarchy process (AHP). Then, a fuzzy relation matrix is ​​established based on expert experience or historical data to describe the degree of membership of each evaluation factor to different risk levels. Finally, the risk level of each obstacle is obtained through fuzzy synthesis operations, generating a risk assessment report.

[0138] Example 3 illustrates a method for reconstructing a three-dimensional scene of a power transmission corridor. It should be noted that the technical solution of this system for reconstructing a three-dimensional scene of a power transmission corridor is based on the same concept as the method described above. Details not described in detail in this embodiment can be found in the description of the method described above.

[0139] This embodiment also provides a system for reconstructing a three-dimensional scene of a power transmission corridor, including:

[0140] The data acquisition module is used to acquire multi-view images and first point cloud data of the power transmission corridor;

[0141] The feature extraction module is used to extract features from multi-view images to obtain multi-view feature maps;

[0142] The image calculation module is used to input multi-view feature maps into the depth multi-view stereo model and calculate and obtain disparity maps;

[0143] The 3D model generation module is used to generate second point cloud data based on the disparity map, and then combine the first point cloud data and the second point cloud data to reconstruct the 3D model of the power transmission corridor.

[0144] The prevention and detection module is used to detect and classify obstacles in the 3D model of the power transmission corridor, obtain the detection results, and generate a risk assessment report based on the detection results.

[0145] This embodiment also provides an electronic device suitable for reconstructing a three-dimensional scene of a power transmission corridor, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for reconstructing a three-dimensional scene of a power transmission corridor as proposed in the above embodiment.

[0146] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for reconstructing a three-dimensional scene of a power transmission corridor as proposed in the above embodiments.

[0147] The storage medium proposed in this embodiment and the method for reconstructing a three-dimensional scene of a power transmission corridor proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0148] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0149] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for reconstructing a three-dimensional scene of a power transmission corridor, characterized in that, include: Acquire multi-view images and first point cloud data of the power transmission corridor; Feature extraction is performed on the multi-view images to obtain multi-view feature maps; The multi-view feature map is input into the depth multi-view stereo model to calculate and obtain the disparity map; Based on the disparity map, second point cloud data is generated, and combined with the first and second point cloud data, a three-dimensional model of the power transmission corridor is reconstructed. Obstacle detection and classification are performed on the three-dimensional model of the power transmission corridor to obtain detection results, and a risk assessment report is generated based on the detection results.

2. The method for reconstructing a three-dimensional scene of a power transmission corridor as described in claim 1, characterized in that, Feature extraction is performed on the multi-view images to obtain multi-view feature maps, including: A deep convolutional neural network model is used to perform convolutional pooling operations on the multi-view images to extract key feature points, and a multi-view feature map is generated based on the key feature points.

3. The method for reconstructing a three-dimensional scene of a power transmission corridor as described in claim 2, characterized in that, The multi-view feature map is input into the depth multi-view stereo model to calculate and obtain the disparity map, including: The depth multi-view stereo model calculates the disparity value of each pixel in the input multi-view feature map to generate a preliminary disparity map. The preliminary disparity map is then geometrically optimized to obtain the final disparity map.

4. The method for reconstructing a three-dimensional scene of a power transmission corridor as described in claim 3, characterized in that, Based on the disparity map, second point cloud data is generated, including: Based on the disparity map and combined with the acquisition device parameters, the disparity map is converted into a depth map; The pixels in the depth map are converted into three-dimensional spatial points to generate second point cloud data.

5. The method for reconstructing a three-dimensional scene of a power transmission corridor as described in claim 4, characterized in that, By combining the first point cloud data and the second point cloud data, a three-dimensional model of the power transmission corridor is reconstructed, including: The second point cloud data was optimized using the Poisson surface reconstruction algorithm; By combining the optimized second point cloud data and the first point cloud data, the point cloud data is converted into a 3D network using the triangulation method; The three-dimensional network is smoothed to reconstruct a three-dimensional model of the power transmission corridor.

6. The method for reconstructing a three-dimensional scene of a power transmission corridor as described in claim 5, characterized in that, Obstacle detection and classification are performed on the three-dimensional model of the power transmission corridor to obtain detection results. A risk assessment report is generated based on the detection results, including: The point cloud data is segmented using the Euclidean clustering algorithm to extract the geometric features of the three-dimensional model of the power transmission corridor, and obstacles are identified based on the geometric features. The obstacles are classified using a support vector machine classifier to obtain the detection results.

7. The method for reconstructing a three-dimensional scene of a power transmission corridor as described in claim 6, characterized in that, Also includes: Calculate the distance between obstacles and transmission lines, and assess the risk level using a weighted scoring method; The risk levels are divided into low risk, medium risk, and high risk, and different risk assessment reports are automatically generated based on the risk level.

8. A system for reconstructing a three-dimensional scene of a power transmission corridor, using the method described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire multi-view images and first point cloud data of the power transmission corridor; The feature extraction module is used to extract features from the multi-view images to obtain multi-view feature maps; The image calculation module is used to input the multi-view feature map into the depth multi-view stereo model and calculate and obtain the disparity map; The 3D model generation module is used to generate second point cloud data based on the disparity map, and to reconstruct a 3D model of the power transmission corridor by combining the first point cloud data and the second point cloud data. The prevention and detection module is used to detect and classify obstacles in the three-dimensional model of the power transmission corridor, obtain the detection results, and generate a risk assessment report based on the detection results.

9. A computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method for reconstructing a three-dimensional scene of a power transmission corridor as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method for reconstructing a three-dimensional scene of a power transmission corridor as described in any one of claims 1 to 7.

Images