Platform gauge detection method

Abstract

The invention discloses a platform gauge detection method. The method comprises the following steps: S1, collecting data; s2, carrying out time synchronization and space synchronization on the station data; s3, point cloud fusion; s4, denoising the point cloud; s5, carrying out platform rough matching based on matrix transformation; s6, platform fine matching based on an improved particle swarm optimization algorithm; s7, platform limit detection; and S8, displaying in real time, and uploading a result. According to the method, matching of the platform and the design contour can be achieved, the change of the platform is obtained, and platform gauge detection, platform height extraction and side gauge value extraction are completed. According to the method, the point cloud data of the platform are integrated, synchronized and collected through multiple sensors, data processing and analysis are fully automatically carried out on the basis of a low-cost scanner through a high-precision platform intelligent detection algorithm, platform limit detection is rapidly completed, and the requirement for measurement precision can be met at the same time.

Description

Technical Field

[0001] The present invention relates to the technical field of railway and urban rail transit platform detection, and in particular to a platform clearance detection method. Background Art

[0002] By 2024, the operating mileage of my country's railways and urban rail transit will be nearly 200,000 kilometers, and the demand for operation and maintenance inspections will become increasingly greater. The Guangdong Railway Group, Beijing Bureau and other railway bureaus require platform limit inspections to be carried out at least once every quarter, which has formed a huge inspection market demand.

[0003] At present, most manual measurement methods in platform clearance detection use total stations or platform clearance measuring rulers. These single-point measurement methods are not continuous, the efficiency needs to be improved, and there may also be leakage points. In the intelligent detection method, the Taiyuan Institute of Science and Technology has developed a portable railway building clearance detector, which uses laser ranging to measure, and calculates the distance data based on the calibrated measurement angle, controllable optical path switch and the number of rotation steps of the stepper motor. However, the measurement accuracy of the system is not high and the detection time is relatively long; the large-scale vehicle-mounted building clearance detection system that can be used for continuous measurement has a high single-time operation efficiency and can meet the accuracy requirements. This type of system is generally integrated with high-precision scanners such as German Z+F and Leica, and line point cloud data collection is realized through mobile measurement. After processing and analysis, the clearance detection is completed, and the rapid operation detection effect on long lines is better. However, multi-platform detection faces the problem of repeated online and offline. In this type of platform clearance intelligent detection, the detection systems integrated by various manufacturers are generally heavy and relatively expensive, and the efficiency of multi-platform clearance detection is not high.

[0004] The use of low-cost portable equipment integrated detection systems can improve work efficiency and reduce detection costs. The use of low-cost laser radars has lower accuracy than high-precision laser radars, and requires improvement and optimization in the algorithm to meet the tolerance requirements of the detection specifications. Currently, there is little research and development of similar detection result optimization algorithms at home and abroad. Summary of the invention

[0005] In order to solve the problems in the background technology, the present invention proposes a platform limit detection method with high matching accuracy and strong timeliness.

[0006] To this end, the present invention adopts the following technical solutions:

[0007] A platform clearance detection method comprises the following steps:

[0008] S1, data collection:

[0009] Collecting platform data at different times on the rails of the platform in the area to be inspected, wherein the platform data includes inclination data, platform point cloud data and mileage data;

[0010] S2, performing time synchronization and space synchronization on the station data;

[0011] S3, point cloud fusion:

[0012] The synchronized station data obtained in S2 are merged to generate cross-section point cloud data after the inclination angle correction at the corresponding mileage; the cross-section point cloud data includes cross-section point cloud frames at different times;

[0013] S4, point cloud denoising:

[0014] S41, calculate the spatial weight G s (p i ,p j ), where i and j are the point numbers, p j For p i Neighborhood N(p i ) in any point, the neighborhood N(p i ) is p i The cross-section point cloud frame is represented by p i is the center of the circle, R is the radius of the area, the neighborhood N(p i ) does not include point p i , the R is set according to the actual situation;

[0015] S42, calculate the intensity weight G r ;

[0016] S43, calculate the midpoint p of the cross-section point cloud data obtained in S3 i The normal vector is as follows:

[0017]

[0018] Among them, C i For point p i The covariance matrix of Represents all points to point p in the neighborhood i The average distance of , T is the transposition operator;

[0019] Then, for the C i Perform eigenvalue decomposition to obtain several eigenvalues ​​λ and corresponding eigenvectors v, and take the smallest eigenvector as point p i The normal vector n i ;

[0020] S44, calculate the normal vector weight G n (n i ,n j ), as shown below:

[0021]

[0022] Among them, point p j The normal vector n j Solved by S43 method, n i *n j is the inner product of the normal vectors, which is used to represent the similarity between normal vectors, σ n n i and n j The standard deviation of the difference;

[0023] S45, calculate the curvature weight function G c :

[0024] S451, calculate the curvature of the midpoint of the cross-section point cloud data, as shown in the following formula:

[0025]

[0026] Among them, λ max,i is the maximum value among several eigenvalues ​​λ in S43, λ sum,i is the sum of all eigenvalues ​​calculated in S43, c i For point p i The curvature of

[0027] Then use the solution c i The same method is used to calculate the point p j The curvature c j :

[0028] S452, calculate the curvature weight function G c , as shown below:

[0029]

[0030] Among them, σ c c i and c j The standard deviation of the difference;

[0031] S46, denoising the cross-section point cloud data obtained in S3 using the following formula to obtain denoised point cloud data:

[0032]

[0033] Among them, p i ' is a point in the denoised point cloud data;

[0034] S5, platform coarse matching based on matrix transformation:

[0035] According to the design data of the platform in the area to be detected, a standard dimension diagram of the platform outline is obtained; then, a RANSAC algorithm is used to obtain the optimal transformation matrix of the denoised point cloud data according to the standard dimension diagram of the platform outline;

[0036] S6, station precise matching based on improved particle swarm optimization algorithm, includes the following steps:

[0037] S61, initializing all particles in the particle swarm optimization algorithm;

[0038] S62, determining the iterative method of the particle position and velocity, including the following sub-steps:

[0039] S621, determine the basic speed of the particle at each iteration, as follows:

[0040] v k ′(t+1)=ω(t)*v k (t)+a1r1(P k (t)-x k (t))+a2r2(g(t)-x k (t));

[0041] Among them, v k (t) represents the velocity of particle k after t iterations, v k ′(t+1) represents the basic velocity of particle k at the t+1th iteration; ω(t) is the inertia weight at the tth iteration; a1 and a2 are acceleration coefficients set according to the actual working conditions; r1 and r2 are random numbers in the range [0,1]; the local optimal solution P k (t) represents the best position that particle k has passed through before the tth iteration; the global optimal solution g(t) is the best position that all particles have passed through before the tth iteration, x k (t) is the position of particle k after t iterations;

[0042] S622, determine the velocity of the particle after each iteration:

[0043] Update the basic velocity obtained in S621 to obtain the velocity v of particle k after t+1 iterations k (t+1), we have the following formula:

[0044]

[0045] S623, determine the position of the particle after each update:

[0046] The position x of particle k after t+1 iterations is determined by the following formula k (t+1):

[0047] x k(t+1)=x k (t)+v k (t+1);

[0048] S63, evaluating the position of each particle by the following formula, calculating the objective function value of the position of each particle after t iterations, and recording the particle position with the best objective function value;

[0049] MinF(x k (t))=(f1(x k (t)), f2(x k (t)), f3(x k (t)));

[0050] Where MinF is the objective function, f1 is the root mean square error evaluation function, f2 is the maximum error evaluation function, and f3 is the matching time;

[0051] S64, terminate the iteration and obtain the optimal solution:

[0052] When the optimal particle position in S63 no longer changes, it is applied as the final result to the denoised point cloud data obtained in S46, so that the denoised point cloud data is accurately aligned with the platform outline standard size diagram, and the final cross-section point cloud data is obtained to complete the precise matching;

[0053] S7, check platform limits:

[0054] The platform limit detection results include the platform height value and the minimum limit value in the horizontal direction;

[0055] The vertical coordinates of the platform corner points in the final cross-section point cloud data in step S6 are the platform height values;

[0056] The minimum absolute value of the horizontal coordinates of the points on the side of the platform in the final cross-section point cloud data in S6 is obtained as the minimum limit value in the horizontal direction of the platform limit detection.

[0057] Step S2 is specifically as follows: time-synchronize the platform data at the same time obtained in S1; spatially synchronize the inclination data and mileage data in the time-synchronized vehicle coordinate system and the platform point cloud data in the scanner coordinate system to the track coordinate system with the center of the track surface as the origin at the corresponding time to complete spatial synchronization.

[0058] Step S41 is specifically as follows: the spatial weight G is calculated by the following formula: s :

[0059]

[0060] Among them, ||p i -p j || is point pi and point p j The Euclidean distance between s For p i and p j The standard deviation of the Euclidean distance.

[0061] Step S42 is specifically as follows: the intensity weight G is calculated by the following formula: r :

[0062]

[0063] Among them, I i and I j For point p i and p j The strength of r is the standard deviation of the intensity difference; the intensity of the point in the cross-section point cloud data is obtained through the corresponding point in the station data in S1.

[0064] Step S5 includes the following sub-steps:

[0065] S51, randomly selecting a certain number of points from the denoised point cloud data obtained in S4 as a coarse matching point set;

[0066] S52, estimating a transformation matrix using the coarse matching point set;

[0067] S53, rotating the denoised point cloud data using the transformation matrix calculated in S52, and calculating the errors between the points in the rotated denoised point cloud data and the points in the platform outline standard size diagram;

[0068] S54, selecting a threshold value according to the actual working condition. If the error obtained in S53 is within the threshold range, the point in the denoised point cloud data corresponding to the error is an inlier, and the number of all inliers obtained by the corresponding transformation matrix is ​​counted;

[0069] S55, repeating steps S51 to S54, and selecting the transformation matrix with the largest number of inliers as the optimal transformation matrix.

[0070] The initialization of all particles in S61 includes: the total number n of particles, the initial position of each particle, the initial speed of each particle, and the optimal transformation matrix obtained in S5 is used as the initial individual value of each particle.

[0071] Step S63 also includes the following steps:

[0072] S631, calculate convergence:

[0073] The convergence degree N(t) of the particle swarm after the tth iteration is calculated as shown in the following formula:

[0074]

[0075] S632, adjusting the inertia weight in S621, as shown in the following formula:

[0076]

[0077] Among them, ω(0) and ω min is the initial value and minimum value of all inertia weights, N(0) is the initial convergence degree;

[0078] S633, guide the global optimal solution:

[0079] In each iteration, the local optimal solution is solved by the following formula:

[0080] P k (t+1)=P k (t)+v k (t+1)+γ(g(t)-P k (t)) (14)

[0081] Among them, γ is the guiding factor, which is determined according to the actual working conditions.

[0082] Preferably, the initial value of the guiding factor γ is 2. When the convergence degree N(t) is low, the value of γ is increased by 0.1 each time to balance the local and global search capabilities of the particles.

[0083] Preferably, the platform clearance detection method of the present invention further comprises:

[0084] S8, real-time display, result upload:

[0085] During the actual detection, the detection results of the platform limit of the platform section at the current position are displayed in real time according to steps S1-S7 and uploaded to the host computer.

[0086] Compared with the prior art, the present invention has the following beneficial effects:

[0087] 1. The method of the present invention uses data output by multiple sensors integrated on the same track inspection vehicle; compared with the use of separate detection methods such as manual total stations and limit rulers, the method has strong continuity and high efficiency.

[0088] 2. The method of the present invention can complete data processing fully automatically and generate detection results quickly. Through the rough matching of platforms based on matrix transformation and the precise matching of platforms based on particle swarm optimization, the dynamic convergence control mechanism is introduced to adaptively adjust the ratio of global and local searches, inertia weights, etc., which significantly improves the matching accuracy and robustness of the algorithm; it can also effectively avoid premature convergence and improve the global optimization ability of the algorithm, thereby quickly completing platform extraction and limit detection.

[0089] 3. The method of the present invention can output the platform clearance detection results in real time and automatically upload the detection results. When the detection results are abnormal, on-site review can be carried out in time, which greatly improves the timeliness of the detection work and the effectiveness of the detection results.

[0090] 4. The method of the present invention introduces normal vector weights and curvature weights, and can adaptively adjust standard deviation parameters. Traditional bilateral filtering generally only considers the spatial and intensity information of point clouds. This method enhances the applicability and robustness of platform detection in different scenarios and ensures the accuracy of limit detection.

[0091] 5. The present invention is applicable to all railway and urban rail transit platform clearance detection, providing data support for the operation and maintenance department to ensure the safety of platform operation. At the same time, the platform matching results can also be used to determine the changes between the platform and the design size, whether it is damaged, etc., and can also provide a reference for the repair of the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 It is a flow chart of a platform clearance detection method of the present invention;

[0093] Figure 2 It is a structural schematic diagram of a track inspection system for running the method of the present invention;

[0094] Figure 3 Schematic diagram of the conversion of two spatial rectangular coordinate systems in step S2 of the present invention;

[0095] Figure 4 is a schematic diagram of a platform cross-section point cloud after point cloud fusion in an embodiment of the present invention;

[0096] Figure 5 It is a standard dimension diagram of the platform outline in the embodiment of the present invention;

[0097] Figure 6 It is a schematic diagram of precise station matching based on particle swarm optimization in an embodiment of the present invention.

[0098] In the figure: 1, scanner; 2, encoder; 3, inclinometer; 4, tablet computer; 5, scanner bracket; 6, push rod; 7, battery slot; 8, fixing bracket; 9, top wheel;

[0099] 10, platform surface; 11, platform side; 12, platform bottom edge. DETAILED DESCRIPTION

[0100] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and embodiments.

[0101] In order to implement the platform limit detection method of the present invention, the platform limit detection system must be able to simultaneously obtain the inclination data, platform point cloud data and mileage data at the current moment.

[0102] The scanner 1, encoder 2, inclinometer 3, tablet computer 4, and hand-push vehicle platform are integrated into a platform clearance detection system. The hand-push vehicle platform includes a scanner bracket 5, a hand push rod 6, a battery slot 7, a fixing frame 8, a top wheel 9, and a vehicle body. The vehicle platform is equipped with various sensors for mobile measurement, and uses a high-precision scanner to collect platform point cloud data, while obtaining the inclination data and mileage data at the current moment. The platform clearance detection system is shown in the figure below: Figure 2 , the specific composition is as follows:

[0103] (1) Driving structure:

[0104] 1. The vehicle body bracket adopts hollow square tube. Under the premise of sufficient structural strength, the wall thickness of the square tube is reduced and the weight is reduced; the vehicle body adopts modular assembly, and the T-type connection intersections are all equipped with quick-release screws, which is convenient for installation; the vehicle body platform is equipped with handles for easy cross-line transportation.

[0105] 2. The push rod adopts a rotatable and adjustable buckle mechanism, which can choose different heights and angles for pushing, suitable for different operators to push.

[0106] 3. The scanner is fixed on the vehicle platform using a scanner bracket and triangularly fixed using a fixing frame to ensure the stability of the scanner data acquisition.

[0107] 4. The encoder is embedded in the wheel, the inclinometer is installed inside the vehicle body, and each sensor is equipped with a wiring port plug, which is plug-in and correspondingly connected, which is quick to install and reduces the exposure of data cables.

[0108] 5. The wheels of the whole vehicle are made of nylon wheels, and the top wheels are made of ceramic wheels to ensure the insulation of the vehicle body.

[0109] (2) Tightening structure:

[0110] The tightening structure of the hand-push car body platform adopts a three-point tightening method to reduce the swing in the plane. The tightening wheel on the right end is installed with a spring. During the pushing process, it adaptively adjusts according to the changes in the track gauge to ensure the fit between the car body and the rail.

[0111] (3) System communication and storage:

[0112] The system uses lithium batteries to power each sensor. A single battery system can operate continuously for more than 3 hours. The battery is placed in the battery slot at the bottom of the scanner bracket. The battery powers the scanner, encoder, inclinometer and tablet computer; the data of each sensor is stored in the tablet computer.

[0113] The platform clearance detection system used in the present invention adopts lightweight materials and modular design, and uses a push-type vehicle platform to carry various sensors for mobile scanning and data collection. During the process, the scanner collects platform point cloud data; the encoder collects mileage data; the inclinometer collects inclination data; and the tablet computer is used to complete the method of the present invention. Figure 1 As shown, the method of the present invention specifically comprises the following steps:

[0114] S1, data collection:

[0115] The platform data at different times are obtained through the platform clearance detection system on the rails of the platform in the area to be detected. The platform data includes inclination data, platform point cloud data and mileage data. The platform point cloud data includes the coordinates of the platform point cloud and the intensity of the platform point cloud.

[0116] S2, time synchronization and space synchronization of station data:

[0117] The platform data obtained by S1 at the same time are synchronized in time; the inclination data and mileage data in the vehicle coordinate system after time synchronization, and the platform point cloud data in the scanner coordinate system are spatially synchronized to the track coordinate system with the center of the track surface as the origin at the corresponding time, completing the spatial synchronization. Specifically:

[0118] Time synchronization: The encoder and inclinometer data are stored in the tablet computer and the same system time is recorded. The encoder is directly connected to the scanner, and the collected scanner data carries the corresponding mileage information to complete the time synchronization of each sensor.

[0119] Spatial synchronization: Using the calibration method, the conversion relationship between different coordinate systems is established. a X a Y a Z a and O b X b Y b Z b Convert as Figure 3 The two coordinate axes are not parallel and the scales are not consistent. First, the two coordinate systems are translated to make the origin coincide, and then three rotations are performed to make the three coordinate axes parallel to each other. The relationship between the two coordinate systems is as follows:

[0120]

[0121] Where, X a , Y a , Z a is the coordinate before transformation, X b , Y b , Z bThe coordinates after conversion; X0, Y0, Z0 are translation parameters, δμ is the scale factor, R X R Y R z is the rotation matrix equation. The translation parameters, scale factors and rotation matrix equation are obtained according to the actual working conditions.

[0122] S3, point cloud fusion:

[0123] The platform point cloud data, inclination data and mileage data in the track coordinate system obtained by S2 are fused to generate the cross-section point cloud data after inclination correction at the corresponding mileage. The cross-section point cloud data includes cross-section point cloud frames at different times; the intensity of the point cloud in the cross-section point cloud data is the same as the intensity of the corresponding point in the platform point cloud.

[0124] S4, point cloud denoising:

[0125] The cross-sectional point cloud data with mileage information and inclination correction after S3 point cloud fusion often contains some noise, and these outliers are likely to affect the accuracy of platform detection. The surface and side of the platform are generally straight lines. The improved bilateral filtering method is used to preliminarily process the point cloud data. Compared with the traditional filtering method, the normal vector and curvature information are introduced to further improve the denoising effect. The specific steps include:

[0126] S41, calculate the spatial weight G s , as shown in formula (2):

[0127]

[0128] Among them, ||p i -p j || is point p i and point p j The Euclidean distance between j For p i Neighborhood N(p i ) in any point, σ s For p i and p j The standard deviation of the Euclidean distance of the neighborhood N(p i ) is p i The cross-section point cloud frame is represented by p i The area with center as circle and radius as R, and the neighborhood does not include point p i , R is set according to actual conditions.

[0129] S42, calculate the intensity weight G r , as shown in formula (3):

[0130]

[0131] Among them, Ii and I j For point p i and p j The strength of r is the standard deviation of the intensity difference.

[0132] S43, calculate the normal vector of the midpoint of the cross-section point cloud data:

[0133] For point p i , we can calculate its covariance matrix C i And calculate point p by judging the size of the characteristic vector i The normal vector n i , the specific calculation method is as follows:

[0134]

[0135] In the formula, Represents all points to point p in the neighborhood i The average distance.

[0136] The covariance matrix C i Perform eigenvalue decomposition to obtain several eigenvalues ​​λ and corresponding eigenvectors v, and take the smallest v among the eigenvectors v min As point p i The normal vector n i .v min For point p i The midpoint p of the cross-section point cloud frame i The direction with the least change.

[0137] S44, calculate the normal vector weight G n , as shown in formula (5):

[0138]

[0139] Among them, point p j The normal vector n j Solved by S43 method, n i *n j is the inner product of the normal vectors, which is used to represent the similarity between normal vectors, σ n n i and n j The standard deviation of the difference.

[0140] S45, calculate the curvature weight function G c :

[0141] S451, calculate the curvature of the point in the point cloud:

[0142] Point p i The curvature c iCalculated by formula (6):

[0143]

[0144] Among them, λ max,i is the maximum value among several eigenvalues ​​λ in S43, λ sum,i is the sum of all eigenvalues ​​calculated in S43.

[0145] Use the same method to calculate point p j The curvature c j :

[0146] S452, calculate the curvature weight function G c , as shown in formula (7):

[0147]

[0148] Among them, σ c c i and c j The standard deviation of the difference;

[0149] S46, denoising the cross-section point cloud data:

[0150] According to steps S41 to S45, the spatial weight, intensity weight, normal vector weight, and curvature weight of all points in the cross-section point cloud data are calculated, and then the cross-section point cloud data is denoised using equation (8) to obtain denoised point cloud data:

[0151]

[0152] Among them, p i ′ is a point in the denoised point cloud data.

[0153] S5, platform coarse matching based on matrix transformation:

[0154] According to the design data of the platform in the area to be inspected, obtain the standard dimension drawing of the platform outline.

[0155] The RANSAC algorithm is used to obtain the optimal transformation matrix through random sampling and model verification. The specific process is as follows:

[0156] S51, randomly selecting a certain number of points from the denoised point cloud data as a coarse matching point set;

[0157] S52, estimating a transformation matrix through a coarse matching point set;

[0158] S53, rotating the denoised point cloud data using the transformation matrix calculated in S52, and calculating the error between the points in the rotated denoised point cloud data and the points in the platform outline standard size diagram;

[0159] S54, selecting a suitable threshold value according to the actual working condition, if the error obtained in S53 is within the threshold range, the point is an interior point, and the number of all interior points obtained by the corresponding transformation matrix is ​​counted;

[0160] S55, repeating steps S51 to S54, selecting the transformation matrix with the largest number of inliers as the optimal transformation matrix;

[0161] S56, using the optimal transformation matrix to transform the denoised point cloud data to achieve a rough match between it and the standard size map of the platform outline.

[0162] S6, station precise matching based on improved particle swarm optimization algorithm:

[0163] The platform matching result obtained through rough matching in step S5 still has certain errors and cannot meet the accuracy requirements of platform detection. It is necessary to perform fine matching to further improve the matching accuracy between the platform section and the designed contour. Fine matching uses the particle swarm optimization algorithm (Particle Swarm Optimization, PSO) to search for the global optimal solution through individual collaboration and information sharing in the group. Design a particle with speed and position attributes, where the speed represents the speed of movement and the position represents the direction of movement. Each particle searches for the optimal solution of the transformation matrix in the search space separately, and records it as the current individual extreme value, and shares the individual extreme value with other particles in the entire particle swarm. The optimal individual extreme value is used as the current global optimal solution of the entire particle swarm. All particles in the particle swarm adjust their speed and position according to their current individual extreme value and the current global optimal solution shared by the entire particle swarm, and seek the optimal solution through iteration. The fine matching process based on particle swarm optimization is as follows: Figure 6 As shown, the specific process is as follows:

[0164] S61, initialize all particles, including:

[0165] (1) Initialize the number of particles: determine the total number of particles n;

[0166] (2) Initialize the particle position: Randomly initialize the particle swarm and determine the initial position of each particle. The position of each particle represents a possible solution.

[0167] (3) Initialize particle velocity: Initialize the particle velocity and determine the initial velocity of each particle. The particle velocity is used to determine the update direction and step size of the particle position in each iteration.

[0168] (4) The optimal transformation matrix obtained by S5 is used as the initial individual value of each particle.

[0169] S62, iterative method to determine the position and velocity of particles:

[0170] S621, determine the basic velocity of the particle at each iteration, and obtain it through formula (9):

[0171] v k ′(t+1)=ω(t)*v k (t)+a1r1(P k (t)-x k (t))+a2r2(g(t)-x k (t)) (9)

[0172] Among them, v k (t) represents the velocity of particle k after t iterations, v k ′(t+1) represents the basic speed of particle k at the t+1 iteration; ω(t) is the inertia weight at the t iteration. The initial value of the inertia weight is generally set to 0.9. A large inertia weight helps to enhance the global search capability; a1 and a2 are acceleration coefficients, which are set according to the actual working conditions; r1 and r2 are random numbers, and their value range is [0,1]; the local optimal solution P k (t) represents the best position that particle k has passed through before the tth iteration; the global optimal solution g(t) is the best position that all particles have passed through before the tth iteration, x k (t) is the position of particle k after t iterations. When t=0, it is the initial value set in S61.

[0173] S622, determine the velocity of the particle after each iteration:

[0174] The basic speed determined by S621 may have a local optimum, so a nonlinear adjustment factor is introduced into the speed update formula to improve the speed update mechanism and the speed calculation accuracy. The speed v of particle k after t+1 iterations is obtained. k (t+1), we have formula (10):

[0175]

[0176] S623, determine the position of the particle after each update:

[0177] The position x of particle k after t+1 iterations is determined by equation (11): k (t+1):

[0178] x k (t+1)=x k (t)+v k (t+1) (11)

[0179] S63, Dynamic convergence calculation of improved particle swarm algorithm:

[0180] After the particle swarm is initialized, optimal matching is required. We introduce an improved adaptive dynamic convergence control mechanism, which calculates the convergence state of the particle swarm in real time and dynamically adjusts the weights of global and local searches to avoid premature convergence and improve matching accuracy, thereby optimizing the convergence speed and accuracy of the algorithm.

[0181] S631, calculate convergence:

[0182] Calculate the convergence state of the current particle swarm, as shown in formula (12):

[0183]

[0184] Where N(t) is the convergence degree of the tth iteration.

[0185] S632, adjust the inertia weight:

[0186] At the same time, adaptive weights are introduced to dynamically adjust the number of iterations and improve the adaptive weight calculation as shown in formula (13).

[0187]

[0188] Among them, ω max and ω min is the initial value and minimum value of all inertia weights, and N(0) is the initial convergence degree. The inertia weight can be dynamically adjusted according to the global optimal and local optimal values ​​of the particle swarm, thereby balancing the capabilities of global search and local search. The initial convergence degree N(0) is calculated by the initial value of the particle set in S61.

[0189] When the convergence degree N(t) is large, ω(t) is close to ω max , to enhance the global search capability; when N(t) is small, ω(t) decreases, enhancing the local search capability.

[0190] S633, guide the global optimal solution:

[0191] In each iteration, the guiding role of g(t) is enhanced to prevent the algorithm from falling into the local optimum, as shown in formula (14):

[0192] P k (t+1)=P k (t)+v k (t+1)+γ(g(t)-P k (t)) (14)

[0193] Among them, v k(t) represents the velocity of particle i at the tth iteration. γ is the guiding factor, and its initial value is generally set to 2. When the convergence metric is low (i.e. close to convergence), the value of γ is increased by about 0.1 each time to balance the particle's local and global search capabilities.

[0194] S634, objective function evaluation integrating multiple matching errors:

[0195] By adopting the improved fusion multi-matching error evaluation method and optimizing the matching accuracy through a multi-objective optimization method, the extreme difference caused by a single evaluation index can also be avoided, as shown in formula (15):

[0196] MinF(x k (t))=(f1(x k (t)), f2(x k (t)), f3(x k (t))) (15)

[0197] Where MinF is the objective function, f1 is the root mean square error evaluation function, f2 is the maximum error evaluation function, and f3 is the matching time.

[0198] Use formula (15) to evaluate the position of each particle obtained in each iteration, calculate the objective function value of each particle at the position after iteration, and record the particle position with the best objective function value.

[0199] S64, terminate the iteration and obtain the optimal solution:

[0200] When the optimal particle position in S634 no longer changes, it is applied as the final result to the denoised point cloud data, so that it is accurately aligned with the standard dimension diagram of the platform outline to obtain the final cross-section point cloud data and complete the precise matching.

[0201] S7, check platform limits:

[0202] The platform limit detection results include the platform height value and the minimum limit value in the horizontal direction;

[0203] The final cross-sectional point cloud data obtained through precise matching in step S6 can be directly used for platform clearance detection and analysis. The vertical coordinates of the platform corner points of the matched standard platform contour are the platform height values ​​for platform clearance detection. The side point cloud is then processed and analyzed to obtain the minimum absolute value of the horizontal coordinates of the points on the side of the platform in the final cross-sectional point cloud data as the minimum limit value in the horizontal direction of the platform clearance detection.

[0204] S8, real-time display, result upload:

[0205] During the actual inspection, the inspection results of the platform limit of the current platform section can be displayed in real time for each frame of point cloud and uploaded to the host computer. When the inspection results are abnormal, they can also be reviewed in time, which greatly improves the inspection efficiency.

[0206] Example

[0207] In one embodiment of the present invention, the cross-sectional point cloud data of the platform to be inspected is obtained through steps S1-S3; Figure 4 The point cloud data of one frame in the cross-sectional point cloud data is shown in FIG. After performing point cloud denoising on this frame of point cloud data in step S4, step S5 is performed to compare the denoised point cloud frame with the Figure 5 The platform outline standard size diagram at the corresponding position shown in the figure is roughly matched; finally, the result of the rough matching is input into S6 for fine matching. The result of the fine matching of the above point cloud frame data is as follows Figure 6 shown.

[0208] Can be obtained from Figure 6 It can be seen that the point cloud frame data is basically consistent with the standard dimension diagram of the platform contour at the corresponding position, that is, this point cloud frame can be used as the platform contour at this position to detect the platform boundary.

Claims

1. A platform clearance detection method, characterized in that: The following steps are involved: S1, data collection: Collecting platform data at different times on the rails of the platform in the area to be inspected, wherein the platform data includes inclination data, platform point cloud data and mileage data; S2, performing time synchronization and space synchronization on the station data; S3, point cloud fusion: The synchronized station data obtained in S2 are merged to generate cross-section point cloud data after the inclination angle correction at the corresponding mileage; the cross-section point cloud data includes cross-section point cloud frames at different times; S4, point cloud denoising: S41, calculate the spatial weight G s (p i ,p j ), where i and j are the point numbers, p j For p i Neighborhood N(p i ) in any point, the neighborhood N(p i ) is p i The cross-section point cloud frame is represented by p i is the center of the circle, R is the radius of the area, the neighborhood N(p i ) does not include point p i , the R is set according to the actual situation; S42, calculate the intensity weight G r ; S43, calculate the midpoint p of the cross-section point cloud data obtained in S3 i The normal vector is as follows: Among them, C i For point p i The covariance matrix of Represents all points to point p in the neighborhood i The average distance of , T is the transposition operator; Then, for the C i Perform eigenvalue decomposition to obtain several eigenvalues ​​λ and corresponding eigenvectors v, and take the smallest eigenvector as point p i The normal vector n i ; S44, calculate the normal vector weight G n (n i ,n j ), as shown below: Among them, point p j The normal vector n j Solved by S43 method, n i *n j is the inner product of the normal vectors, which is used to represent the similarity between normal vectors, σ n n i and n j The standard deviation of the difference; S45, calculate the curvature weight function G c : S451, calculate the curvature of the midpoint of the cross-section point cloud data, as shown in the following formula: Among them, λ max,i is the maximum value among several eigenvalues ​​λ in S43, λ sum,i is the sum of all eigenvalues ​​calculated in S43, c i For point p i The curvature of Then use the solution c i The same method is used to calculate the point p j The curvature c j : S452, calculate the curvature weight function G c , as shown below: Among them, σ c c o and c j The standard deviation of the difference; S46, denoising the cross-section point cloud data obtained in S3 using the following formula to obtain denoised point cloud data: Among them, p i ′ is a point in the denoised point cloud data; S5, platform coarse matching based on matrix transformation: According to the design data of the platform in the area to be detected, a standard dimension diagram of the platform outline is obtained; then, a RANSAC algorithm is used to obtain the optimal transformation matrix of the denoised point cloud data according to the standard dimension diagram of the platform outline; S6, station precise matching based on improved particle swarm optimization algorithm, includes the following steps: S61, initializing all particles in the particle swarm optimization algorithm; S62, determining the iterative method of the particle position and velocity, including the following sub-steps: S621, determine the basic speed of the particle at each iteration, as follows: v k ′(t+1)=ω(t)*v k (t)+a1r1(P k (t)-x k (t))+a2r2(g(t)-x k (t)); Among them, v k (t) represents the velocity of particle k after t iterations, v k ′ (t+1) represents the basic velocity of particle k at the t+1th iteration; ω(t) is the inertia weight at the tth iteration; a1 and a2 are acceleration coefficients set according to the actual working conditions; r1 and r2 are random numbers in the range [0,1]; the local optimal solution P k (t) represents the best position that particle k has passed through before the tth iteration; the global optimal solution g(t) is the best position that all particles have passed through before the tth iteration, x k (t) is the position of particle k after t iterations; S622, determine the velocity of the particle after each iteration: Update the basic velocity obtained in S621 to obtain the velocity v of particle k after t+1 iterations k (t+1), we have the following formula: S623, determine the position of the particle after each update: The position x of particle k after t+1 iterations is determined by the following formula k (t+1): x k (t+1)=x k (t)+v k (t+1); S63, evaluating the position of each particle by the following formula, calculating the objective function value of the position of each particle after t iterations, and recording the particle position with the best objective function value; MinF(x k (t))=(f1(x k (t)),f2(x k (t)),f3(x k (t))); Where MinF is the objective function, f1 is the root mean square error evaluation function, f2 is the maximum error evaluation function, and f3 is the matching time; S64, terminate the iteration and obtain the optimal solution: When the optimal particle position in S63 no longer changes, it is applied as the final result to the denoised point cloud data obtained in S46, so that the denoised point cloud data is accurately aligned with the platform outline standard size diagram, and the final cross-section point cloud data is obtained to complete the precise matching; S7, check platform limits: The platform limit detection results include the platform height value and the minimum limit value in the horizontal direction; The vertical coordinates of the platform corner points in the final cross-section point cloud data in step S6 are the platform height values; The minimum absolute value of the horizontal coordinates of the points on the side of the platform in the final cross-section point cloud data in S6 is obtained as the minimum limit value in the horizontal direction of the platform limit detection.

2. The platform clearance detection method according to claim 1, characterized in that: Step S2 is specifically as follows: time-synchronize the platform data at the same time obtained in S1; spatially synchronize the inclination data and mileage data in the time-synchronized vehicle coordinate system and the platform point cloud data in the scanner coordinate system to the track coordinate system with the center of the track surface as the origin at the corresponding time to complete spatial synchronization.

3. The platform clearance detection method according to claim 1, characterized in that: Step S41 is specifically as follows: the spatial weight G is calculated by the following formula: s : Among them, ||p i -p j || is point p i and point p j The Euclidean distance between s For p i and p j The standard deviation of the Euclidean distance.

4. The platform clearance detection method according to claim 1, characterized in that: Step S42 is specifically as follows: the intensity weight G is calculated by the following formula: r : Among them, I i and I j For point p i and p j The strength of r is the standard deviation of the intensity difference; the intensity of the point in the cross-section point cloud data is obtained through the corresponding point in the station data in S1.

5. The platform clearance detection method according to claim 1, characterized in that: Step S5 The following steps are included: S51, randomly selecting a certain number of points from the denoised point cloud data obtained in S4 as a coarse matching point set; S52, estimating a transformation matrix using the coarse matching point set; S53, rotating the denoised point cloud data using the transformation matrix calculated in S52, and calculating the errors between the points in the rotated denoised point cloud data and the points in the platform outline standard size diagram; S54, selecting a threshold value according to the actual working condition. If the error obtained in S53 is within the threshold range, the point in the denoised point cloud data corresponding to the error is an inlier, and the number of all inliers obtained by the corresponding transformation matrix is ​​counted; S55, repeating steps S51 to S54, and selecting the transformation matrix with the largest number of inliers as the optimal transformation matrix.

6. The platform clearance detection method according to claim 1, characterized in that: The initialization of all particles in S61 includes: the total number n of particles, the initial position of each particle, the initial speed of each particle, and the optimal transformation matrix obtained in S5 is used as the initial individual value of each particle.

7. The platform clearance detection method according to claim 6, characterized in that: Step S63 also includes the following steps: S631, calculate convergence: The convergence degree N(t) of the particle swarm after the tth iteration is calculated as shown in the following formula: S632, adjusting the inertia weight in S621, as shown in the following formula: Among them, ω(0) and ω min is the initial value and minimum value of all inertia weights, N(0) is the initial convergence degree; S633, guide the global optimal solution: In each iteration, the local optimal solution is solved by the following formula: P k (t+1)=P k (t)+v k (t+1)+γ(g(t)-P k (t)) (14) Among them, γ is the guiding factor, which is determined according to the actual working conditions.

8. The platform clearance detection method according to claim 7, characterized in that: The initial value of the guiding factor γ is 2. When the convergence degree N(t) is low, the value of γ is increased by 0.1 each time to balance the local and global search capabilities of the particles.

9. The platform clearance detection method according to claim 1, characterized in that: Also includes: S8, real-time display, result upload: During the actual detection, the detection results of the platform limit of the platform section at the current position are displayed in real time according to steps S1-S7 and uploaded to the host computer.

Images