Pipeline butt joint positioning auxiliary guiding system based on laser radar

Abstract

The invention discloses a pipeline butt joint positioning auxiliary guiding system based on a laser radar, and belongs to the technical field of pipeline butt joint positioning. The pipeline butt joint positioning auxiliary guiding system based on the laser radar comprises a data acquisition module, a pipeline model module and an optimization module. The problem that in the prior art, when butt joint positioning is carried out, certain errors possibly exist, and consequently ideal precision is difficult to achieve in actual butt joint is solved, data collection is carried out through the laser radar, the precision can reach the millimeter level, the accuracy and stability of pipeline butt joint are ensured, the working efficiency is remarkably improved, and the labor intensity of workers is lowered. In the pipeline butt joint process, the positioning auxiliary guiding time can be greatly shortened, the laser radar can also be used for measuring the speed, the angle and other parameters, the laser radar can provide more comprehensive information support in pipeline butt joint positioning auxiliary guiding, the accuracy of pipeline butt joint positioning is further ensured, and the positioning accuracy is improved. And the ideal precision can be achieved during actual butt joint of the pipelines.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline docking positioning, and in particular to a pipeline docking positioning auxiliary guidance system based on laser radar. Background Art

[0002] Municipal sewage pipes are an important part of the urban infrastructure system. They are not only related to the quality of life and public health of urban residents, but also have a profound impact on environmental protection and social sustainable development. With the continuous advancement of technology and the improvement of environmental awareness, municipal sewage pipe construction technology will continue to innovate to adapt to the ever-changing urbanization and environmental challenges.

[0003] Chinese patent publication number CN113478128A discloses a pipe delivery docking and positioning system for long-distance pipeline installation, the system includes a pipeline docking and positioning device and a pipe delivery device placed on one side of the pipeline docking and positioning device; the pipeline docking and positioning device includes a first and a second pipeline adjustment device, a pipeline moving device, a supporting device, a pipeline positioning mechanism and a positioning base plate; the pipe delivery device includes a pipeline lifting mechanism and a pipeline transfer mechanism, the long-distance pipeline to be installed is lifted to the height of the pipeline moving device by the pipeline lifting mechanism, and the long-distance pipeline to be installed is transferred to the pipeline moving device of the pipeline docking and positioning device by the pipeline transfer mechanism, and the docking and positioning device is docked and positioned with the installed long-distance pipeline, which avoids a large number of workers from climbing to high altitudes to work, greatly reduces the high-altitude work risks and labor costs of manual operations, and improves the installation efficiency of the long-distance pipeline and the safety of the installation process.

[0004] In the actual use of the above-mentioned patent, the docking positioning device is docked with the installed long-distance pipeline. There may be certain errors in the design and manufacturing process of the docking positioning device, which makes it difficult to achieve the ideal accuracy during actual docking. Therefore, it does not meet the existing needs. In this regard, we propose a pipeline docking positioning auxiliary guidance system based on laser radar. Summary of the invention

[0005] The purpose of the present invention is to provide a pipeline docking positioning auxiliary guidance system based on laser radar, which uses laser radar for data collection with an accuracy of up to millimeter level, ensuring the accuracy and stability of pipeline docking, significantly improving work efficiency, and greatly shortening the time for positioning auxiliary guidance during pipeline docking. The laser radar can provide more comprehensive information support in pipeline docking positioning auxiliary guidance, further ensuring the accuracy of pipeline docking positioning, so that the pipeline can achieve ideal accuracy during actual docking, solving the problems raised in the above-mentioned background technology.

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a laser radar-based pipeline docking positioning auxiliary guidance system, comprising: The data acquisition module is used to calculate the position change of the laser radar, use the position change of the laser radar to obtain the point cloud data of the pipeline surface and process it, and reconstruct the pipeline surface after the processing; The pipeline model module is used to randomly extract sample data from the reconstructed pipeline surface point cloud data using the RANSAC algorithm, filter out abnormal values ​​that do not conform to the model, obtain the best fitting model parameters, and use the best fitting model parameters to model the pipeline; The optimization module is used to calculate the distance from the point to the surface of the cylindrical model, obtain the parameters of the initialized cylindrical model using the RANSAC algorithm, iterate the parameters of the initialized cylindrical model, and gradually optimize the fitting effect according to the updated results of the model parameters in each iteration.

[0007] Preferably, the data acquisition module comprises: The IMU module is used to test the changes in the speed and attitude of the laser radar and calculate the position change of the laser radar through integration; A data acquisition module is used to acquire point cloud data of the pipeline surface using the laser radar position change; The data processing module is used to pre-process the point cloud data in the working surface equation by using straight-through filtering, and to eliminate useless point cloud data according to the position range of the two pipelines relative to the laser radar; The surface reconstruction module is used to calculate the equation of the working surface using the processed point cloud data, and reconstruct the pipeline surface using the equation of the working surface.

[0008] Preferably, the pipeline model module includes: A sampling module, used to randomly extract sampling data from the reconstructed pipeline surface point cloud data using the RANSAC algorithm, and estimate the model based on the sampling data; The screening module is used to iteratively screen out the outliers that do not conform to the model in the random sampling data of the pipeline surface point cloud to obtain the best fitting model parameters; The model building module is used for modeling the pipeline by using the model parameters with the best fit to obtain the cylindrical model of the pipeline.

[0009] Preferably, the sampling module specifically includes: Select the number of iterations and threshold to randomly select the minimum number of points from the data set to fit the cylinder model; Estimate the parameters of the cylindrical model based on the selected points, calculate the fitting errors of other data points to the estimated cylindrical model, and determine the inliers that fall within a certain threshold range; If the number of points that fit the model is greater than the preset threshold, the model is considered a candidate model, otherwise the minimum number of points is randomly selected from the data set to fit the cylindrical model; If the specified number of iterations is reached or the number of points that meet the model is large enough, the iteration stops; When the stopping condition is reached, all the data points considered as inliers are used to re-estimate the final cylinder model parameters, which are the best fitting model parameters.

[0010] Preferably, the optimization module specifically includes: Segment the cylindrical model, and use each point in the segmented point cloud as a residual block. Use the residual block to calculate the error of each point under the current cylindrical model parameters, and obtain the distance from the point to the surface of the cylindrical model. Define the objective function, which is the sum of the squares of the distances from the point to the cylinder model; The RANSAC algorithm is used to calculate the initial values ​​of the center coordinates and radius of the cylindrical model to obtain the parameters of the initialized cylindrical model; The parameters of the initialized cylinder model are iterated, and the fitting effect is gradually optimized according to the updated results of the model parameters in each iteration.

[0011] Preferably, the step of gradually optimizing the fitting effect according to the update result of the model parameters in each iteration specifically includes: Calculate the gradient of the objective function with respect to the model parameters; Calculate the Hessian matrix of the objective function; The parameters are updated using Newton's method, and the fitting effect is gradually optimized according to the updated model parameters.

[0012] Preferably, the updating of parameters using Newton's method specifically includes: Define the convergence condition. When the change of the objective function value is less than a certain threshold or the change of the parameter is less than a certain threshold, the termination state of the iteration is determined. If the convergence condition is met, the iteration stops.

[0013] X new are the updated model parameters, X old are the model parameters before updating, f(X old ) For the function X old The function value at ; Solve the cylinder model equation, the formula is as follows:

[0014] In the above formula, x S , y S , z S is a point on the surface of the cylinder model, x 0 , y 0 , z 0 is the center coordinate of the cylindrical model, r 0 is the radius of the cylindrical model, a , b , c For the cylindrical model body X , Y , Z The direction vector of the axis, N is a positive integer, indicating the number of terms to be summed; The free cuboid fitting is used to provide Y-axis constraints for the cylindrical model, and the sliding window algorithm is used to process the point cloud in real time; The two sets of processed point clouds are used to solve the 3D Bounding Box respectively, and the Y-axis distance is calculated according to the calculation results to obtain the distance between the two pipes.

[0015] Preferably, the sliding window algorithm specifically includes: Set the 10-frame sequence output of the lidar to the window size; Use median filtering in the sliding window to reduce the impact of anomaly detection. When the data is smaller than the window size, the data will continue to accumulate. When the data is larger than the window size, the window will start to slide. Use the lidar's highest scan rate sliding window to add the latest observations to the queue.

[0016] Preferably, the laser radar-based pipeline docking positioning auxiliary guidance system further includes: The convolutional neural network model building module is used to design a convolutional neural network (CNN) model architecture suitable for point cloud data. The point cloud data for training is preprocessed by voxelization or point cloud convolution method, and converted into point cloud data in a format that can be processed by the convolutional neural network (CNN). The preprocessed point cloud data is used as the input data of the convolutional neural network (CNN) model for training, and the output is the characteristic parameters of the pipeline, which include the centerline coordinates, radius and axial vector. Then, through multi-layer convolution, pooling and full connection operations, the characteristic information in the point cloud data is extracted, and feature fusion and regression prediction are performed to obtain the pipeline model parameters. The generative adversarial network model construction module is used to construct a generative adversarial network (GAN) model including a generator and a discriminator; expand the training point cloud data and optimize the generative adversarial network (GAN) model through training. Specifically, random noise and pipeline feature parameters are input into the generator to output synthetic pipeline point cloud data; real point cloud data and synthetic pipeline point cloud data output by the generator are input into the discriminator to judge the authenticity of the synthetic pipeline point cloud data; during the training process, the generator continuously learns to generate synthetic point clouds that are closer to real data, and the discriminative ability of the discriminator is continuously improved through training; through adversarial training, the quality of the synthetic point cloud data generated by the generator is improved, and the quality of the synthetic point cloud data is supplemented into the training data set as additional training samples to enhance the generalization ability of the generative adversarial network model; Model deployment module, used to deploy the trained convolutional neural network (CNN) model and generative adversarial network (GAN) model into the pipeline docking positioning auxiliary guidance system; The model real-time optimization module is used to input the point cloud data collected by the real-time lidar into the convolutional neural network (CNN) model, quickly identify the pipeline characteristic parameters, and establish the pipeline model based on the pipeline characteristic parameters; according to the pipeline site conditions and docking requirements, the generative adversarial network (GAN) model is used to optimize and adjust the pipeline model in real time, and after optimization, a pipeline model that better meets the pipeline construction requirements is generated.

[0017] Preferably, the laser radar-based pipeline docking positioning auxiliary guidance system further includes: The maximum error storage module is used for the cylinder model optimized by the optimization module using the RANSAC algorithm, and the pipeline model optimized by the model real-time optimization module using the generative adversarial network (GAN) model. Each time it is used, the maximum errors of the two models are calculated and stored using the following formulas:

[0018] In the above formula, δ imax For the i The maximum error of the model, i is the serial number of the cylinder model or pipe model, n i For the i The number of output data of each model, x ik For the i The model k Output data, For the k The true value of the output data; The model accuracy evaluation module is used to calculate the reliability index of the two models respectively using the following formula when the cylinder model and the pipe model are inconsistent:

[0019] In the above formula, τ i For the i The reliability index of the model, m is the storage quantity of the maximum error, δ i1max For the i The maximum error of this model, σ imax For storage i The mean of the maximum error of the models; The model selection module is used to compare the reliability indexes of the cylinder model and the pipeline model, and select the model with a larger reliability index for auxiliary positioning of the pipeline docking.

[0020] Compared with the prior art, the present invention has the following beneficial effects: The present invention utilizes laser radar for data collection with an accuracy of up to millimeter level, thereby ensuring the accuracy and stability of pipeline docking and significantly improving work efficiency. During the pipeline docking process, the time for positioning auxiliary guidance can be greatly shortened. The laser radar can also be used to measure speed, angle and other parameters, so that the laser radar can provide more comprehensive information support in pipeline docking positioning auxiliary guidance, further ensuring the accuracy of pipeline docking positioning, so that the pipelines can achieve ideal accuracy during actual docking. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of a module of a pipeline docking positioning auxiliary guidance system based on laser radar of the present invention; Figure 2 It is a schematic diagram of the laser radar-based pipeline docking positioning auxiliary guidance system method of the present invention. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] In order to solve the problem that the existing technology is used in actual use, the docking positioning device is used to dock and position the installed long-distance pipeline. There may be certain errors in the design and manufacturing process of the docking positioning device, which makes it difficult to achieve the ideal accuracy in actual docking. Please refer to Figure 1-Figure 2, this embodiment provides the following technical solutions: The laser radar-based pipeline docking positioning auxiliary guidance system includes: The data acquisition module is used to calculate the position change of the laser radar, obtain the point cloud data of the pipeline surface using the position change of the laser radar, and process it. After the processing is completed, the pipeline surface is reconstructed. The laser radar is connected to the laptop through a network cable and driven by the ROS system (Robot Operating System); The pipeline model module is used to randomly extract sample data from the reconstructed pipeline surface point cloud data using the RANSAC algorithm (random sample consensus algorithm), filter out abnormal values ​​that do not conform to the model, obtain the best fitting model parameters, and use the best fitting model parameters to model the pipeline; The optimization module is used to calculate the distance from the point to the surface of the cylindrical model, use the RANSAC algorithm (random sample consensus algorithm) to obtain the parameters of the initialized cylindrical model, iterate the parameters of the initialized cylindrical model, and gradually optimize the fitting effect according to the updated results of the model parameters in each iteration.

[0024] Data acquisition module, including: The IMU (Inertial Measurement Unit) module is used to test the changes in the speed and attitude of the laser radar. The position change of the laser radar is calculated by integration, so the relative position change is measured; A data acquisition module is used to acquire point cloud data of the pipeline surface using the laser radar position change; The point cloud data is preprocessed using straight-through filtering. According to the possible position range of the two pipes relative to the laser radar, useless point cloud data is eliminated. The proportion of important data in the point cloud data after straight-through filtering is greatly increased, but these point cloud data still contain some noise, which will complicate the calculation of the geometric features of the local point cloud and even cause errors, leading to fitting failure. Outlier point filtering is used to calculate the average distance from each point to its neighboring points. The algorithm performs statistical analysis on the calculation results, and the calculation results are Gaussian distributions with mean values ​​and standard deviations. For points whose average distance is outside the interval defined by the mean value and standard deviation, they are noise and are eliminated from the point cloud.

[0025] The data processing module is used to pre-process the point cloud data in the working surface equation by using straight-through filtering, and to eliminate useless point cloud data according to the position range of the two pipelines relative to the laser radar; The surface reconstruction module is used to calculate the equation of the working surface using the processed point cloud data, and reconstruct the pipeline surface using the equation of the working surface.

[0026] Pipeline model module, including: A sampling module is used to randomly extract sampling data from the reconstructed pipeline surface point cloud data using the RANSAC algorithm (random sample consensus algorithm) and estimate the model based on the sampling data; The screening module is used to iteratively screen out the outliers that do not conform to the model in the random sampling data of the pipeline surface point cloud to obtain the best fitting model parameters. The outliers that do not conform to the model are screened out through an iterative process, and the points that conform to the model are attributed to the internal points of the model, and the points that do not conform to the model are attributed to the external points of the model to obtain the best fitting model parameters. The pipeline is modeled using the best fitting model parameters.

[0027] The model building module is used for modeling the pipeline by using the model parameters with the best fit to obtain the cylindrical model of the pipeline.

[0028] Specifically, the RANSAC algorithm (Random Sample Consensus Algorithm) is an iterative method for estimating mathematical model parameters. It is particularly suitable for situations where there are noise and outliers in the data. The RANSAC algorithm estimates model parameters by randomly selecting data points and screens out outliers that do not conform to the model through an iterative process, ultimately obtaining the best fitting model parameters.

[0029] Sampling module, including: Select the number of iterations and threshold to randomly select the minimum number of points from the data set to fit the cylinder model; Estimate the parameters of the cylinder model, such as center, radius, axis, etc., based on the selected points, calculate the fitting error of other data points to the estimated cylinder model, and determine the inliers that fall within a certain threshold range; If the number of points that fit the model is greater than the preset threshold, the model is considered a candidate model, otherwise the minimum number of points is randomly selected from the data set to fit the cylindrical model; If the specified number of iterations is reached or the number of points that meet the model is large enough, the iteration is stopped; When the stopping condition is reached, all the data points considered as inliers are used to re-estimate the final cylinder model parameters, which are the best fitting model parameters.

[0030] Optimization modules include: Segment the cylindrical model, and use each point in the segmented point cloud as a residual block. Use the residual block to calculate the error of each point under the current cylindrical model parameters, and obtain the distance from the point to the surface of the cylindrical model. Define the objective function, which is the sum of the squares of the distances from the point to the cylinder model; The RANSAC algorithm (random sample consensus algorithm) is used to calculate the initial values ​​of the center coordinates and radius of the cylinder model to obtain the parameters of the initialized cylinder model; The parameters of the initialized cylinder model are iterated, and the fitting effect is gradually optimized according to the updated results of the model parameters in each iteration. The fitting effect is gradually optimized according to the updated results of the model parameters in each iteration.

[0031] In each iteration, the fitting effect is gradually optimized according to the updated results of the model parameters, including: Calculate the gradient of the objective function with respect to the model parameters; Calculate the Hessian matrix of the objective function; The fitting effect is gradually optimized according to the updated model parameters.

[0032] Use Newton's method to update parameters, including: Define the convergence condition. When the change of the objective function value is less than a certain threshold or the change of the parameter is less than a certain threshold, the termination state of the iteration is determined. If the convergence condition is met, the iteration stops.

[0033] X new are the updated model parameters, X old are the model parameters before updating, f(X old ) For the function X old The function value at ; Solve the cylinder model equation, the formula is as follows:

[0034] In the above formula, x S , y S , z S is a point on the surface of the cylinder model, x 0 , y 0 , z 0 is the center coordinate of the cylindrical model, r 0 is the radius of the cylindrical model, a , b , c For the cylindrical model body X , Y , ZThe direction vector of the axis, N is a positive integer, indicating the number of terms to be summed; The free cuboid fitting is used to provide Y-axis constraints for the cylindrical model. At the same time, the sliding window algorithm is used to process the point cloud in real time to ensure the density of the point cloud and the accuracy of calculation. The sliding window algorithm is a technology used to process data series or time series. It processes and analyzes data by moving a fixed-size window on the data series, allowing continuous data subsets to be processed in the data stream and performing specific operations on each subset. The two groups of processed point clouds are used to solve the 3D Bounding Box respectively, and the Y-axis distance is calculated according to the calculation results to obtain the distance between the two pipes.

[0035] Sliding window algorithm, specifically including: Set the 10-frame sequence output of the lidar to the window size; Use median filtering in the sliding window to reduce the impact of anomaly detection. When the data is smaller than the window size, the data will continue to accumulate. When the data is larger than the window size, the window will start to slide. Use the lidar's highest scan rate sliding window to add the latest observations to the queue.

[0036] Working principle: When using the laser radar-based pipeline docking positioning auxiliary guidance system of the present invention, according to Figure 1 and Figure 2 , including the following steps: Step 1: Install the laser radar so that the X-axis is nearly perpendicular to the pipe wall and the Y-axis is nearly parallel to the pipe wall. Test the changes in the speed and posture of the laser radar. Calculate the position change of the laser radar by integration, and use the position change of the laser radar to obtain the point cloud data of the pipe surface. Step 2: Use straight-through filtering to pre-process the point cloud data in the working surface equation, remove useless point cloud data, use the processed point cloud data to calculate the working surface equation, and use the working surface equation to reconstruct the pipeline surface; Step 3: Use the RANSAC algorithm (random sample consensus algorithm) to randomly extract sample data from the reconstructed pipeline surface point cloud data, and estimate the model based on the sample data. Through an iterative process, outliers that do not conform to the model are screened out, and points that conform to the model are attributed to internal points of the model, and points that do not conform to the model are attributed to external points of the model, so as to obtain the best model parameters for fitting; Step 4: Model the pipeline using the best fitting model parameters to obtain a cylindrical model of the pipeline; Step 5: Segment the cylindrical model, and use each point in the segmented point cloud as a residual block. Use the residual block to calculate the error of each point under the current cylindrical model parameters to obtain the distance from the point to the surface of the cylindrical model. Step 6: Define the objective function, use the RANSAC algorithm (random sample consensus algorithm) to calculate the initial values ​​of the center coordinates and radius of the cylinder model, obtain the parameters of the initialized cylinder model, iterate the parameters of the initialized cylinder model, and gradually optimize the fitting effect according to the updated results of the model parameters in each iteration; Step 7: According to the optimized fitting effect, adjust the device by operating the remote control to gradually reduce X , Y , Z The data of the axis is taken over.

[0037] In summary, in the laser radar-based pipeline docking positioning auxiliary guidance system of the present invention, the laser radar can measure the shape, size, position and other parameters inside and outside the pipeline with an accuracy of up to millimeter level. This high precision is crucial for pipeline docking positioning and can ensure the accuracy and stability of docking. The laser radar uses a laser beam for measurement and does not need to directly contact the pipeline surface. The laser radar can quickly scan the pipeline surface and quickly detect large-area pipelines, which significantly improves work efficiency. During the pipeline docking process, the positioning auxiliary guidance time can be greatly shortened. In addition to measuring distance, the laser radar can also be used to measure speed, angle and other parameters, so that the laser radar can provide more comprehensive information support in pipeline docking positioning auxiliary guidance.

[0038] Based on the above-mentioned embodiment, the laser radar-based pipeline docking positioning auxiliary guidance system further includes: The convolutional neural network model building module is used to design a convolutional neural network (CNN) model architecture suitable for point cloud data. The point cloud data for training is preprocessed by voxelization or point cloud convolution method, and converted into point cloud data in a format that can be processed by convolutional neural networks (CNN). The preprocessed point cloud data is used as the input data of the convolutional neural network (CNN) model for training, and the output is the characteristic parameters of the pipeline, which include the centerline coordinates, radius and axial vector. Then, through multi-layer convolution, pooling and full connection operations, the characteristic information in the point cloud data is extracted, and feature fusion and regression prediction are performed to obtain the pipeline model parameters. The generative adversarial network model construction module is used to construct a generative adversarial network (GAN) model including a generator and a discriminator; expand the training point cloud data and optimize the generative adversarial network (GAN) model through training. Specifically, random noise and pipeline feature parameters are input into the generator to output synthetic pipeline point cloud data; real point cloud data and synthetic pipeline point cloud data output by the generator are input into the discriminator to judge the authenticity of the synthetic pipeline point cloud data; during the training process, the generator continuously learns to generate synthetic point clouds that are closer to real data, and the discriminative ability of the discriminator is continuously improved through training; through adversarial training, the quality of the synthetic point cloud data generated by the generator is improved, and the quality of the synthetic point cloud data is supplemented into the training data set as additional training samples to enhance the generalization ability of the generative adversarial network model; Model deployment module, used to deploy the trained Convolutional Neural Networks (CNN) model and Generative Adversarial Network (GAN) model into the pipeline docking positioning auxiliary guidance system; The model real-time optimization module is used to input the point cloud data collected by the real-time lidar into the Convolutional Neural Networks (CNN) model, quickly identify the pipeline characteristic parameters, and establish the pipeline model based on the pipeline characteristic parameters; according to the pipeline site conditions and docking requirements, the Generative Adversarial Network (GAN) model is used to optimize and adjust the pipeline model in real time, and after optimization, a pipeline model that better meets the pipeline construction requirements is generated.

[0039] Specifically, preprocessing the data, such as normalization and data enhancement, and expanding the sample size can improve the generalization ability of the model; utilizing the generative adversarial network (GAN) generation characteristics to optimize the pipeline model, and by adjusting the input parameters of the generator, generating pipeline models of different shapes, sizes, and postures, to provide more reference solutions for pipeline docking; in complex pipeline scenarios, being able to quickly and accurately identify pipeline features will improve the automatic recognition and optimization capabilities of the pipeline model, establish a high-precision pipeline model, and optimize and adjust the model based on real-time feedback, which will greatly improve the efficiency and accuracy of pipeline docking, reduce manual intervention, and realize intelligent and automated pipeline construction, thereby improving the technical level of the entire pipeline construction industry.

[0040] Based on the above-mentioned embodiment, the laser radar-based pipeline docking positioning auxiliary guidance system further includes: The maximum error storage module is used for the cylinder model optimized by the optimization module using the RANSAC algorithm (random sample consensus algorithm), and the pipeline model optimized by the model real-time optimization module using the Generative Adversarial Network (GAN) model. Each time it is used, the maximum errors of the two models are calculated and stored using the following formulas:

[0041] In the above formula, δ imax For the i The maximum error of the model, i is the serial number of the cylinder model or pipe model, n i For the i The number of output data of each model, x ik For the i The model k Output data, For the k The true value of the output data; The model accuracy evaluation module is used to calculate the reliability index of the two models respectively using the following formula when the cylinder model and the pipe model are inconsistent:

[0042] In the above formula, τ i For the i The reliability index of the model, m is the storage quantity of the maximum error, δ i1max For the i The maximum error of this model, σ imax For storage i The mean of the maximum error of the models; The model selection module is used to compare the reliability indexes of the cylinder model and the pipeline model, and select the model with a larger reliability index for auxiliary positioning of the pipeline docking.

[0043] Specifically, the maximum error of the two models is calculated and stored each time they are used. By calculating the maximum error, the performance of the model or system in the worst case can be understood, ensuring the stability and reliability of the model or system in practical applications. For each maximum error, the reliability index of the model is calculated using the above formula, and the reliability indexes of the cylinder model and the pipeline model are compared. The model with a larger reliability index is selected for this pipeline docking positioning assistance, thereby improving the positioning accuracy of the current positioning assistance.

[0044] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only the elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0045] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. The laser radar-based pipeline docking positioning auxiliary guidance system is characterized by: include: The data acquisition module is used to calculate the position change of the laser radar, use the position change of the laser radar to obtain the point cloud data of the pipeline surface and process it, and reconstruct the pipeline surface after the processing; The pipeline model module is used to randomly extract sample data from the reconstructed pipeline surface point cloud data using the RANSAC algorithm, filter out abnormal values ​​that do not conform to the model, obtain the best fitting model parameters, and use the best fitting model parameters to model the pipeline; The optimization module is used to calculate the distance from the point to the surface of the cylinder model, obtain the parameters of the initialized cylinder model using the RANSAC algorithm, iterate the parameters of the initialized cylinder model, and gradually optimize the fitting effect according to the updated results of the model parameters in each iteration.

2. The laser radar-based pipeline docking positioning auxiliary guidance system according to claim 1 is characterized in that: The data acquisition module comprises: The IMU module is used to test the changes in the speed and attitude of the laser radar and calculate the position change of the laser radar through integration; A data acquisition module is used to acquire point cloud data of the pipeline surface using the laser radar position change; The data processing module is used to pre-process the point cloud data in the working surface equation by using straight-through filtering, and to eliminate useless point cloud data according to the position range of the two pipelines relative to the laser radar; The surface reconstruction module is used to calculate the equation of the working surface using the processed point cloud data, and reconstruct the pipeline surface using the equation of the working surface.

3. The laser radar-based pipeline docking positioning auxiliary guidance system according to claim 1 is characterized in that: The pipeline model module includes: A sampling module, used to randomly extract sampling data from the reconstructed pipeline surface point cloud data using the RANSAC algorithm, and estimate the model based on the sampling data; The screening module is used to iteratively screen out the outliers that do not conform to the model in the random sampling data of the pipeline surface point cloud to obtain the best fitting model parameters; The model building module is used for modeling the pipeline by using the model parameters with the best fit to obtain the cylindrical model of the pipeline.

4. The laser radar-based pipeline docking positioning auxiliary guidance system according to claim 3 is characterized in that: The sampling module specifically includes: Select the number of iterations and threshold to randomly select the minimum number of points from the data set to fit the cylinder model; Estimate the parameters of the cylindrical model based on the selected points, calculate the fitting errors of other data points to the estimated cylindrical model, and determine the inliers that fall within a certain threshold range; If the number of points that fit the model is greater than the preset threshold, the model is considered a candidate model, otherwise the minimum number of points is randomly selected from the data set to fit the cylindrical model; If the specified number of iterations is reached or the number of points that meet the model is large enough, the iteration stops; When the stopping condition is reached, all the data points considered as inliers are used to re-estimate the final cylinder model parameters, which are the best fitting model parameters.

5. The laser radar-based pipeline docking positioning auxiliary guidance system according to claim 1 is characterized in that: The optimization module specifically includes: Segment the cylindrical model, use each point in the segmented point cloud as a residual block, use the residual block to calculate the error of each point under the current cylindrical model parameters, and obtain the distance from the point to the surface of the cylindrical model; Define the objective function, which is the sum of the squares of the distances from the point to the cylinder model; The RANSAC algorithm is used to calculate the initial values ​​of the center coordinates and radius of the cylindrical model to obtain the parameters of the initialized cylindrical model; The parameters of the initialized cylinder model are iterated, and the fitting effect is gradually optimized according to the updated results of the model parameters in each iteration.

6. The laser radar-based pipeline docking positioning auxiliary guidance system according to claim 5 is characterized in that: The step of gradually optimizing the fitting effect according to the update result of the model parameters in each iteration specifically includes: Calculate the gradient of the objective function with respect to the model parameters; Calculate the Hessian matrix of the objective function; The parameters are updated using Newton's method, and the fitting effect is gradually optimized according to the updated model parameters.

7. The laser radar-based pipeline docking positioning auxiliary guidance system according to claim 6 is characterized in that: The method of updating parameters by using Newton's method specifically includes: Define the convergence condition. When the change of the objective function value is less than a certain threshold or the change of the parameter is less than a certain threshold, the termination state of the iteration is determined. If the convergence condition is met, the iteration stops. ； X new are the updated model parameters, X old are the model parameters before updating, f(X old ) For the function X old The function value at ; Solve the cylinder model equation, the formula is as follows: ； In the above formula, x S , y S , z S is a point on the surface of the cylinder model, x 0 , y 0 , z 0 is the center coordinate of the cylindrical model, r 0 is the radius of the cylindrical model, a , b , c For the cylindrical model body X , Y , Z The direction vector of the axis, N is a positive integer, indicating the number of terms to be summed; The free cuboid fitting is used to provide Y-axis constraints for the cylindrical model, and the sliding window algorithm is used to process the point cloud in real time; The two sets of processed point clouds are used to solve the 3D Bounding Box respectively, and the Y-axis distance is calculated according to the calculation results to obtain the distance between the two pipes.

8. The laser radar-based pipeline docking positioning auxiliary guidance system according to claim 7 is characterized in that: The sliding window algorithm specifically includes: Set the 10-frame sequence output of the lidar to the window size; Use median filtering in the sliding window to reduce the impact of anomaly detection. When the data is smaller than the window size, the data will continue to accumulate. When the data is larger than the window size, the window will start to slide. Use the lidar's highest scan rate sliding window to add the latest observations to the queue.

9. The laser radar-based pipeline docking positioning auxiliary guidance system according to claim 1 is characterized in that: Also includes: The convolutional neural network model building module is used to design a convolutional neural network (CNN) model architecture suitable for point cloud data. The point cloud data for training is preprocessed by voxelization or point cloud convolution method, and converted into point cloud data in a format that can be processed by the convolutional neural network (CNN). The preprocessed point cloud data is used as the input data of the convolutional neural network (CNN) model for training, and the output is the characteristic parameters of the pipeline, which include the centerline coordinates, radius and axial vector. Then, through multi-layer convolution, pooling and full connection operations, the characteristic information in the point cloud data is extracted, and feature fusion and regression prediction are performed to obtain the pipeline model parameters. The generative adversarial network model construction module is used to construct a generative adversarial network (GAN) model including a generator and a discriminator; expand the training point cloud data and optimize the generative adversarial network (GAN) model through training. Specifically, random noise and pipeline feature parameters are input into the generator to output synthetic pipeline point cloud data; the real point cloud data and the synthetic pipeline point cloud data output by the generator are input into the discriminator to determine the authenticity of the synthetic pipeline point cloud data; During the training process, the generator continuously learns to generate synthetic point clouds that are closer to real data, and the discriminator's ability to distinguish is continuously improved through training; Through adversarial training, the quality of synthetic point cloud data generated by the generator is improved, and the quality of synthetic point cloud data is added to the training dataset as additional training samples to enhance the generalization ability of the generative adversarial network model; Model deployment module, used to deploy the trained convolutional neural network (CNN) model and generative adversarial network (GAN) model into the pipeline docking positioning auxiliary guidance system; The model real-time optimization module is used to input the point cloud data collected by the real-time lidar into the convolutional neural network (CNN) model, quickly identify the pipeline characteristic parameters, and establish the pipeline model based on the pipeline characteristic parameters; according to the pipeline site conditions and docking requirements, the generative adversarial network (GAN) model is used to optimize and adjust the pipeline model in real time, and after optimization, a pipeline model that better meets the pipeline construction requirements is generated.

10. The laser radar-based pipeline docking positioning auxiliary guidance system according to claim 9, characterized in that: Also includes: The maximum error storage module is used for the cylinder model optimized by the optimization module using the RANSAC algorithm, and the pipeline model optimized by the model real-time optimization module using the generative adversarial network (GAN) model. Each time it is used, the maximum errors of the two models are calculated and stored using the following formulas: ； In the above formula, δ imax For the i The maximum error of the model, i is the serial number of the cylinder model or pipe model, n i For the i The number of output data of each model, x ik For the i The model k Output data, For the k The true value of the output data; The model accuracy evaluation module is used to calculate the reliability index of the two models respectively using the following formula when the cylinder model and the pipe model are inconsistent: ； In the above formula, τ i For the i The reliability index of the model, m is the storage quantity of the maximum error, δ i1max For the i The maximum error of this model, σ imax For storage i The mean of the maximum error of the models; The model selection module is used to compare the reliability indexes of the cylinder model and the pipeline model, and select the model with a larger reliability index for auxiliary positioning of the pipeline docking.

Images