Target identifying and positioning method for insulating elevator in distribution network live working scene

Abstract

The invention provides a target identifying and positioning method for an insulating elevator in a distribution network live working scene. The method comprises the following steps: S1, collecting laser point cloud data and image data of the scene and weighting the laser point cloud data and the image data; s2, fusing the first scene model and the second scene model according to the weight to obtain a scene model; s3, collecting regional point cloud data and regional image data of a region where the corresponding operation point location is located according to the scene type; and S4, target coordinates are determined according to the point location model. The method has the beneficial effects that the scene is firstly recognized by combining the laser point cloud and the image data, then the approximate position of the operation target is found according to the scene, then the target is accurately scanned, and the target is aligned with the standard target model, so that the target recognition accuracy and the working efficiency are improved, the target coordinates are more accurate, and the working energy consumption is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of target recognition, and in particular relates to a target recognition and positioning method for an insulating lift in a live operation scenario of a distribution network. Background Art

[0002] In the live operation scenario of the distribution network, an insulated lift is used to maintain and inspect the distribution network. Because it is in a live state, unmanned recognition technology is required to automatically identify the target and locate the target's coordinates. The insulated lift is moved to the vicinity of the target for maintenance and inspection. The current recognition method is to identify the target and calibrate the target coordinate data through laser point cloud technology. Since the operating targets of the distribution network vary in size, laser radar is required to scan from multiple angles and positions, resulting in low work efficiency. In addition, intensive scanning is required to ensure recognition accuracy, resulting in high power consumption. Summary of the invention

[0003] In view of this, the present invention aims to propose a target identification and positioning method for an insulating lift in a live operation scenario of a distribution network, in order to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above object, the technical solution of the present invention is achieved as follows: A first aspect of the present invention provides a method for identifying and locating a target for an insulating lift in a live working scenario of a distribution network, characterized in that it comprises the following steps: S1, collect laser point cloud data and image data of the scene and perform weighting; S2, constructing a first scene model through the laser point cloud data of the scene, constructing a second scene model according to the image data of the scene, and fusing the first scene model and the second scene model according to the weight to obtain a scene model; S3, inputting the scene model into the scene recognition model to identify the scene type, and collecting the regional point cloud data and regional image data of the area where the corresponding operation point is located according to the scene type; S4. Locate the work point through regional point cloud data and regional image data, build a point model based on the laser point cloud data of the work point, input the image data of the work point into the target recognition model, identify the target type, and determine the target coordinates based on the point model.

[0005] Further, S2 includes the following steps: S21: Constructing a first scene model using laser point cloud data includes the following steps: Preprocessing the collected scene laser point cloud data, and using the preprocessed scene laser point cloud data to construct a first scene model; S22: Constructing a second scene model according to the scene image data comprises the following steps: Preprocessing the collected scene image data, and using the preprocessed scene image data to construct a second scene model; S23: Fusion of the first scene model and the second scene model includes the following steps: The first scene model and the second scene model are aligned, and the first scene model and the second scene model are fused through a multimodal fusion algorithm to generate a three-dimensional scene model.

[0006] Furthermore, the scene model includes coordinate data, and the insulated elevator plans a moving path through the scene model coordinate data.

[0007] Furthermore, S3 includes the following steps: S31. Extract features from the scene model.

[0008] S32, using a pre-trained scene recognition model to identify the scene type; S33. Analyze the area where the work point is located according to the recognition result, move the insulated lift to the corresponding position and collect regional point cloud data and regional image data of the corresponding area.

[0009] Further, S4 includes the following steps: S41, extracting geometric features from the regional point cloud data, and locating the operation point through the regional image data and the geometric features; Use the point cloud data of the operation point to build a three-dimensional model through surface reconstruction as the detection point model; S42, inputting the image data of the operation point into the pre-trained target recognition model to identify the target type; S43, searching for a corresponding standard target point model in a standard target point model library according to the target type; S44, using the detection point model as a basis and using the standard target point model to perform alignment correction, and complete the coordinate data of the detection point model; S45. Determine the target coordinates according to the calibrated detection point model.

[0010] Further, S44 includes the following steps: S441, extracting detection feature points from the detection point model, and extracting feature points identical to the detection feature points from the standard target point model; S442, using a feature matching algorithm to match the detected feature points with feature points of a standard target point model; S443, using an alignment algorithm to align the standard target point model with the detection point model, and calculating an alignment transformation matrix by minimizing the distance between the point clouds; The alignment transformation matrix includes rotation and translation, which is used to align the standard target point model to the detection point model.

[0011] S444: applying a transformation matrix to the coordinate data of the standard target point model to convert it into the coordinate system of the detection point model.

[0012] Furthermore, in S1 and S3, laser point cloud data of the scene is obtained through the laser radar, and image data of the scene is obtained through the camera, and an adaptive weighting mechanism is constructed to adjust the weights of the two devices.

[0013] A second aspect of the present invention provides an electronic device, comprising a processor and a memory communicatively connected to the processor and used to store instructions executable by the processor, wherein the processor is used to execute the method described in the first aspect above.

[0014] The third aspect of the present invention provides a server, comprising at least one processor, and a memory communicatively connected to the processor, the memory storing instructions executable by the at least one processor, the instructions being executed by the processor so that the at least one processor executes the method described in the first aspect.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method described in the first aspect.

[0016] Compared with the prior art, the target identification and positioning method for an insulating lift in a live distribution network operation scenario described in the present invention has the following beneficial effects: The target recognition and positioning method for an insulated lift in a live operation scenario of a distribution network described in the present invention first recognizes the scene by combining laser point cloud and image data, then finds the approximate position of the operation target according to the scene, and then accurately scans the target. By aligning with a standard target model, the accuracy of target recognition and work efficiency are improved, and the target coordinates are more accurate, thereby reducing working energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0018] Figure 1 The figure is a flow chart of the identification and positioning method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0020] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and the like are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second", and the like may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0021] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood by specific circumstances.

[0022] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0023] Embodiment 1: like Figure 1 As shown, a target identification and positioning method for an insulating lift in a live operation scenario of a distribution network includes the following steps: S1. Collect laser point cloud data and image data of the scene and perform weighting.

[0024] In some embodiments, the specific implementation process of step S1 includes: The laser point cloud data of the scene is obtained by the laser radar, and the image data of the scene is obtained by the camera, and an adaptive weighting mechanism is constructed to adjust the weights of the two devices. The adaptive weighting mechanism is specifically as follows: S11. Real-time monitoring of the working environment and performance indicators of the equipment. The working environment is determined by detecting the ambient lighting conditions and weather conditions (such as rain and fog) to determine the camera and lidar. The performance indicators are evaluated using the real-time data output of the sensor. The real-time performance of the sensor is determined by calculating the error of the sensor measurement value (for example, the distance measurement error of the lidar and the image clarity of the camera).

[0025] S12. Adaptively adjust the prior probability based on the real-time environment perception and performance evaluation results: ; in, is the prior probability at time t, is the performance evaluation of sensor i at time t, α is the adjustment coefficient used to control the weight between historical prior and real-time evaluation; If the device performs well in a particular environment, its prior probability is increased; if it performs poorly, its prior probability is decreased.

[0026] S13. Construct a likelihood function to obtain the probability that the laser radar observes the current data under the current environmental conditions and time: ; in, is the expected value of the measurement, is the standard deviation of the measurement error.

[0027] S14. Update the posterior probability by combining the adaptively adjusted prior probability and likelihood function: ; The denominator is the weighted sum of the likelihood functions and prior probabilities of all hypotheses.

[0028] S15. Calculate the weight of the sensor data using the updated posterior probability. At each moment, adjust the weight according to the latest posterior probability. The weight indicates the relative importance of the device: .

[0029] S16. Optimize the prior probability and likelihood function through the feedback mechanism, compare the fusion results with the real environment data (for example, through ground truth data verification), and gradually adjust the model parameters: ; in, is the adjustment amount of the prior probability, R is the actual result, F is the fusion result, and β is the learning rate.

[0030] S2. Construct a first scene model through the laser point cloud data of the scene, construct a second scene model according to the image data of the scene, and obtain a scene model by fusing the first scene model and the second scene model according to the weight.

[0031] The S2 comprises the following steps: S21: Constructing a first scene model using laser point cloud data includes the following steps: Preprocessing the collected scene laser point cloud data, and using the preprocessed scene laser point cloud data to construct a first scene model; S22: Constructing a second scene model according to the scene image data comprises the following steps: Preprocessing the collected scene image data, and using the preprocessed scene image data to construct a second scene model; S23: Fusion of the first scene model and the second scene model includes the following steps: The first scene model and the second scene model are aligned, and the first scene model and the second scene model are fused through a multimodal fusion algorithm to generate a three-dimensional scene model.

[0032] In some embodiments, the process of preprocessing the scene laser point cloud data and the scene image data is as follows: The first cost function is obtained by using the scene image data for reprojection error analysis, and the second cost function is obtained by using the scene laser point cloud data for laser iterative closest point analysis. The first cost function and the second cost function are coupled according to the device weight and coupled positioning algorithm, and the position and posture of the device at the current moment are calculated through a nonlinear optimization algorithm.

[0033] The process of reprojection error analysis includes: Extract key feature points from the image. These feature points can be corners, edges or other salient points, usually extracted using algorithms such as SIFT, ORB or SURF, and match these feature points between multiple images. The matching process can be performed using a distance metric based on descriptors (such as Euclidean distance);

[0034] For each pair of matched feature points, the predicted position of the 3D point on the image plane is calculated based on the current estimated camera pose and internal parameters. The difference between the predicted position and the actual observed position is calculated. This difference is the reprojection error. The square sum of the reprojection errors of all matching feature points is used to construct the first cost function.

[0035] The process of the laser iterative closest point analysis includes: Get two sets of point cloud data (usually collected at consecutive moments) and perform preliminary alignment. The initial alignment can rely on the estimation of the inertial sensor or the pose obtained in the previous step. For each point in the source point cloud, find the nearest point in the target point cloud. This process is usually accelerated using data structures such as KD trees. The Euclidean distance between each pair of nearest points is calculated as the error metric. This distance reflects the spatial inconsistency between the two sets of point clouds, and the sum of the squared distance errors of all paired points is used as the second cost function.

[0036] The process of coupling the first cost function and the second cost function according to the weight of the device and the coupling positioning algorithm is to multiply the cost function with the corresponding weight and add them to obtain the total cost function: ; in, and is the corresponding weight, is the image reprojection error, is the laser point cloud error.

[0037] Furthermore, the process of calculating the position and posture of the device at the current moment through the nonlinear optimization algorithm is as follows: Constructing initial pose parameters , get the initial cost function C(x0), where x, y, z are the position coordinates of the device in three-dimensional space; α, β, γ are three parameters describing the rotation, which can be Euler angles or other representations (such as rotation vectors or quaternions); Calculate the first-order partial derivative matrix J of the total cost function with respect to the pose parameters and calculate the increment Δx: , where r is the residual vector, λ is the damping factor, and J T represents the transpose of matrix J, I is the identity matrix; Update the pose parameters, = + , get the new total cost function value C (x new ); Continue to iterate the above process until Less than a preset threshold, which is determined according to actual conditions, and is generally 10 -4 as a threshold.

[0038] In some embodiments, the construction process of the first scene model and the second scene model is as follows: According to the position and posture of the device at different times, the scene laser point cloud data at different times and the scene image data at different times are spliced ​​to obtain a complete scene laser dot matrix (first scene model) and a scene map (second scene model).

[0039] The scene model includes coordinate data, and the insulated elevator plans the moving path through the scene model coordinate data.

[0040] S3. Input the scene model into the scene recognition model to identify the scene type, and collect regional point cloud data and regional image data of the area where the corresponding operation point is located according to the scene type.

[0041] The S3 comprises the following steps: S31. Extract features from the scene model.

[0042] S32, using a pre-trained scene recognition model to identify the scene type; S33. Analyze the area where the work point is located according to the recognition result, move the insulated lift to the corresponding position and collect laser point cloud data and image data of the corresponding area.

[0043] In some embodiments, the specific execution process of step S32 is as follows: The scene recognition model is used to identify a relatively obvious object that is most relevant to the target object in the current scene model, and the relatively obvious object is used to acquire the target object region according to a preset position relationship.

[0044] S4. Locate the work point through regional point cloud data and regional image data, build a point model based on the laser point cloud data of the work point, input the image data of the work point into the target recognition model, identify the target type, and determine the target coordinates based on the point model.

[0045] The S4 comprises the following steps: S41, extracting geometric features from the regional point cloud data, and locating the operation point through the regional image data and the geometric features; Use the point cloud data of the operation point to build a three-dimensional model through surface reconstruction as the detection point model; S42, inputting the image data of the operation point into the pre-trained target recognition model to identify the target type; S43, searching for a corresponding standard target point model in a standard target point model library according to the target type; S44, using the detection point model as a basis and using the standard target point model to perform alignment correction, and complete the coordinate data of the detection point model; S45. Determine the target coordinates according to the calibrated detection point model.

[0046] In some embodiments, a specific execution process of step S41 is as follows: The regional image data is input into the neural network model to obtain the outline of the target object. According to the position of the outline in the regional image, the specific range of the working point is analyzed, the geometric features corresponding to the range of the working point are selected, and the working point is located according to the geometric features.

[0047] Furthermore, the process of extracting geometric features from regional point cloud data is as follows: S411, select the geometric features to be extracted (such as plane, spherical surface, cylindrical surface, etc.); S412, randomly selecting a group of points in the point cloud, assuming that they belong to the target geometric features, fitting a model (such as a plane equation) using the selected points according to the selected geometric features, and calculating how many points in the point cloud conform to the fitted model, that is, calculating the number of inliers that conform to the model; repeating step S412 until the best model is found, that is, the model with the largest number of inliers; S413. Select inliers from the model with the largest number of inliers as initial seed points. For each seed point, search for points in its neighborhood to determine whether they meet predefined similarity criteria (such as normal vector similarity, distance threshold), add qualified neighborhood points to the current feature region, and use these points as new seed points. Continue searching and expanding. When no new points can be added to the feature region, stop expanding.

[0048] Furthermore, the process of locating the working point according to the geometric features is as follows: Calculate the centroid (center point) of a point cloud: , where N is the total number of points, x i ,y i , z i is the coordinate of each point, subtract the coordinate of the centroid from the coordinate of all points, so that the centroid becomes the origin, thus centralizing the data; Calculate the covariance matrix of the centralized point cloud data: , where V i is the coordinate vector of each centered point; Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors: , where λ i is the eigenvalue, e i is the corresponding eigenvector; Arrange the eigenvalues ​​in descending order. The eigenvalues ​​reflect the variance of the data in the direction of the corresponding eigenvector. Select the eigenvector corresponding to the largest eigenvalue as the main direction of the point cloud. This vector represents the main extension direction of the point cloud in space. The eigenvector corresponding to the second largest eigenvalue usually represents the secondary direction. The third eigenvector is orthogonal to the first two directions.

[0049] The centroid is the geometric center of the work target. The spatial orientation of the work target can be defined through the characteristic vectors of the main direction and the secondary direction. The work point can be located using the center point and the spatial orientation.

[0050] The S44 comprises the following steps: S441, extracting detection feature points from the detection point model, and extracting feature points identical to the detection feature points from the standard target point model; S442, using a feature matching algorithm to match the detected feature points with feature points of a standard target point model; S443, using an alignment algorithm to align the standard target point model with the detection point model, and calculating an alignment transformation matrix by minimizing the distance between the point clouds; The alignment transformation matrix includes rotation and translation, which is used to align the standard target point model to the detection point model.

[0051] S444: applying a transformation matrix to the coordinate data of the standard target point model to convert it into the coordinate system of the detection point model.

[0052] Beneficial effects: By combining laser point cloud and image data, the scene is first identified, and then the approximate location of the work target is found based on the scene. The target is then accurately scanned and aligned with the standard target model to improve the accuracy of target recognition and work efficiency. The target coordinates are more accurate and the working energy consumption is reduced.

[0053] Embodiment 2: An electronic device comprises a processor and a memory which is communicatively connected to the processor and is used to store instructions executable by the processor, wherein the processor is used to execute the method described in the first embodiment above.

[0054] Embodiment three: A server includes at least one processor and a memory communicatively connected to the processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the processor so that the at least one processor executes the method described in Example 1.

[0055] Embodiment 4: A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in embodiment 1 is implemented.

[0056] Finally, it should be noted that 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A target identification and positioning method for an insulating elevator in a live working scenario of a distribution network, characterized in that: The following steps are involved: S1, collect laser point cloud data and image data of the scene and perform weighting; S2, constructing a first scene model through the laser point cloud data of the scene, constructing a second scene model according to the image data of the scene, and fusing the first scene model and the second scene model according to the weight to obtain a scene model; S3, inputting the scene model into the scene recognition model to identify the scene type, and collecting the regional point cloud data and regional image data of the area where the corresponding operation point is located according to the scene type; S4. Locate the work point through regional point cloud data and regional image data, build a point model based on the laser point cloud data of the work point, input the image data of the work point into the target recognition model, identify the target type, and determine the target coordinates based on the point model.

2. According to the target identification and positioning method for an insulating lift in a live working scenario of a distribution network according to claim 1, it is characterized in that: The S2 comprises the following steps: S21: Constructing a first scene model using laser point cloud data includes the following steps: Preprocessing the collected scene laser point cloud data, and using the preprocessed scene laser point cloud data to construct a first scene model; S22: Constructing a second scene model according to the scene image data comprises the following steps: Preprocessing the collected scene image data, and using the preprocessed scene image data to construct a second scene model; S23: Fusion of the first scene model and the second scene model includes the following steps: The first scene model and the second scene model are aligned, and the first scene model and the second scene model are fused through a multimodal fusion algorithm to generate a three-dimensional scene model.

3. According to claim 2, a method for identifying and locating a target for an insulating lift in a live working scenario of a distribution network, characterized in that: The scene model includes coordinate data, and the insulated elevator plans the moving path through the scene model coordinate data.

4. According to claim 1, a method for identifying and locating a target for an insulating lift in a live working scenario of a distribution network, characterized in that: The S3 comprises the following steps: S31, extracting features from the scene model; S32, using a pre-trained scene recognition model to identify the scene type; S33. Analyze the area where the work point is located according to the recognition result, move the insulated lift to the corresponding position and collect regional point cloud data and regional image data of the corresponding area.

5. The target identification and positioning method for an insulating lift in a live distribution network operation scenario according to claim 1 is characterized in that: The S4 comprises the following steps: S41, extracting geometric features from the regional point cloud data, and locating the operation point through the regional image data and the geometric features; Use the point cloud data of the operation point to build a three-dimensional model through surface reconstruction as the detection point model; S42, inputting the image data of the operation point into the pre-trained target recognition model to identify the target type; S43, searching for a corresponding standard target point model in a standard target point model library according to the target type; S44, using the detection point model as a basis and using the standard target point model to perform alignment correction, and complete the coordinate data of the detection point model; S45. Determine the target coordinates according to the calibrated detection point model.

6. The target identification and positioning method for an insulating lift in a live distribution network operation scenario according to claim 5 is characterized in that: The S44 comprises the following steps: S441, extracting detection feature points from the detection point model, and extracting feature points identical to the detection feature points from the standard target point model; S442, using a feature matching algorithm to match the detected feature points with feature points of a standard target point model; S443, using an alignment algorithm to align the standard target point model with the detection point model, and calculating an alignment transformation matrix by minimizing the distance between the point clouds; The alignment transformation matrix includes rotation and translation, which is used to align the standard target point model to the detection point model; S444: applying a transformation matrix to the coordinate data of the standard target point model to convert it into the coordinate system of the detection point model.

7. The target identification and positioning method for an insulating lift in a live distribution network operation scenario according to claim 1 is characterized in that: In S1 and S3, laser point cloud data of the scene is obtained by laser radar, image data of the scene is obtained by camera, and an adaptive weighting mechanism is constructed to adjust the weights of the two devices.

8. An electronic device, comprising a processor and a memory connected to the processor for storing instructions executable by the processor, characterized in that: The processor is used to execute the method described in any one of claims 1 to 7.

9. A server, characterized in that: It includes at least one processor and a memory communicatively connected to the processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the processor so that the at least one processor executes the method according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images