Target-based geometric error calibration system and method for three-dimensional laser scanning equipment of traffic infrastructure

Abstract

The invention belongs to the technical field of highway engineering detection, and particularly provides a target-based traffic infrastructure three-dimensional laser scanning equipment geometric error calibration system and method, and the system comprises a denoising module which is used for carrying out the denoising processing of traffic infrastructure point cloud data collected by three-dimensional laser scanning equipment, and obtaining the denoised point cloud data; the plane error calibration module is used for carrying out plane error calibration on the three-dimensional laser scanning equipment; the three-dimensional error calibration module is used for carrying out three-dimensional error calibration on the three-dimensional laser scanning equipment; and the multi-station equipment calibration module is used for carrying out multi-station equipment calibration based on the plane error calibration result and the stereo error calibration result of the multi-station three-dimensional laser scanning equipment in combination with a four-parameter method and an NDT optimization method, and completing target-based geometric error calibration of the traffic infrastructure three-dimensional laser scanning equipment. According to the technical scheme, the detection precision, efficiency and reliability of the traffic infrastructure structural body can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of highway engineering detection, and in particular relates to a target-based three-dimensional laser scanning equipment geometric error calibration system and method for traffic infrastructure. Background Art

[0002] Traditional methods for inspecting transportation infrastructure structures primarily rely on manual inspections, which are subject to low efficiency, high subjectivity, and high safety risks. With the advancement of technology, two-dimensional inspection techniques based on image processing have gained widespread application. However, this technology can only capture planar information about structures and cannot accurately measure three-dimensional features such as damage and deformation. Furthermore, it is easily affected by factors such as lighting conditions, shadows, and stains, resulting in unstable inspection results.

[0003] In recent years, 3D laser scanning technology has been increasingly used in the inspection of transportation infrastructure structures due to its advantages, including non-contact, high precision, and all-weather operation. This technology generates point cloud data by emitting a laser beam and receiving its reflected signal to obtain the 3D coordinates of the structure. However, in the road transportation industry, 3D laser point cloud scanning technology still faces the following challenges:

[0004] 1. Complex environmental interference: Traffic infrastructure structures can be distorted by stains, water accumulation, and other factors, making it difficult for traditional denoising methods to accurately extract features such as cracks, deformations, and congestion.

[0005] 2. Accumulation of equipment vibration errors: When scanning at multiple sites, coordinate system deviation and point cloud redundancy become prominent. Due to the high accuracy required, multiple sites must be set up for measurement. This results in inherent equipment errors and a large number of stitching errors. Traditional ICP registration takes up to 20 minutes and is not accurate enough.

[0006] 3. Measurement deviation of transportation infrastructure structures: Existing stereo calibration methods are sensitive to edge points. Transportation infrastructure is large, with irregular surfaces and a large number of concrete particles. The geometric measurement calculation error often exceeds 1mm, which cannot meet the maintenance decision-making requirements.

[0007] 4. Poor compliance with standards: The calibration process has not been deeply integrated with highway engineering standards such as the "JTG F80 Highway Engineering Quality Inspection and Assessment Standard Volume 1 Civil Engineering", resulting in low engineering credibility of the measurement results.

[0008] These issues severely limit the large-scale application of 3D laser scanning technology in the inspection of transportation infrastructure structures. Therefore, research on a target-based geometric error calibration system and method for 3D laser scanning equipment for transportation infrastructure is crucial for improving the scientific nature and effectiveness of highway maintenance decision-making. Summary of the Invention

[0009] The present invention provides a target-based geometric error calibration system and method for three-dimensional laser scanning equipment for transportation infrastructure, aiming to address existing issues such as noise interference in point cloud data, difficulty in point cloud data calibration, inaccurate crack depth measurement, and non-intuitive result expression, thereby improving the accuracy, efficiency, and reliability of road pavement crack detection.

[0010] A target-based geometric error calibration system for three-dimensional laser scanning equipment of transportation infrastructure, comprising:

[0011] A denoising module is used to denoise the point cloud data of traffic infrastructure collected by a 3D laser scanning device using a RANSAC method to obtain denoised point cloud data;

[0012] A plane error calibration module, configured to calibrate the plane error of the three-dimensional laser scanning device using a geometric median-Weiszfeld iterative algorithm based on the de-noised point cloud data and in combination with a standard plane target;

[0013] A stereo error calibration module, configured to perform stereo error calibration on the three-dimensional laser scanning device based on the denoised point cloud data and in combination with a standard spherical target using a particle swarm optimization algorithm;

[0014] The multi-site equipment calibration module is used to perform multi-site equipment calibration based on the planar error calibration results and stereo error calibration results of multi-site 3D laser scanning equipment, combining the four-parameter method with the NDT optimization method, to complete the geometric error calibration of target-based 3D laser scanning equipment for transportation infrastructure.

[0015] Preferably, the denoising module includes:

[0016] The initialization unit is used to preset the maximum number of iterations and distance threshold of the RANSAC method based on the data volume and density of the point cloud data to complete parameter initialization;

[0017] The simulation fitting unit is used to randomly sample 3 points in the point cloud data based on the initialized parameters to obtain the minimum set of plane fitting;

[0018] an equation fitting unit, configured to fit the plane equation using a least squares method based on the minimum set;

[0019] an interior point determination unit, configured to calculate the distances from all points in the point cloud data to the plane based on the plane equation, and obtain interior points and outliers;

[0020] The iterative unit is used to obtain the optimal plane parameters and the optimal plane points through iteration, and to remove the abnormal points to obtain the final plane fitting result and complete the denoising of the point cloud data.

[0021] Preferably, the plane error calibration module includes:

[0022] a median obtaining unit, configured to select the geometric median of the valid point set of the final plane fitting result;

[0023] An objective function calculation unit, configured to construct an objective function based on the valid point set and the geometric median;

[0024] A coordinate iteration unit, configured to perform Weiszfeld iteration based on the objective function to obtain a coordinate iteration result of a geometric median;

[0025] The convergence judgment unit is used to construct a threshold range based on the type of transportation infrastructure. When the Euclidean distance between two adjacent coordinate iteration results is less than the threshold range, the iteration is terminated, the final coordinate of the geometric median is obtained, and the plane error calibration of the three-dimensional laser scanning equipment is completed.

[0026] Preferably, the objective function is as follows:

[0027]

[0028] Where M=(m x ,m y ) represents the geometric median, {P1,P2,...,P n}, P i =(x i ,y i ) represents the valid point set.

[0029] Preferably, the stereo error calibration module includes:

[0030] a fitness calculation unit, configured to calculate the fitness of each particle at its current position by evaluating the position quality of each encoded particle; wherein each particle represents a potential solution, including the coordinates of the center and radius of a standard spherical target;

[0031] An optimal parameter acquisition unit, configured to update the particle velocity and position based on the fitness to obtain optimal spherical parameters;

[0032] The stereo error calibration unit is used to fit a sphere based on optimal spherical parameters to complete the stereo error calibration of the three-dimensional laser scanning device.

[0033] Preferably, the multi-site equipment calibration module includes:

[0034] The plane coordinate calibration unit is used to unify the point cloud data in different coordinate systems scanned by multiple sites into a global coordinate system based on the four-parameter method to obtain the plane coordinate calibration results;

[0035] A geometric error calibration unit, configured to optimize the plane coordinate calibration result by using an NDT method to obtain a geometric error calibration result;

[0036] The same-name region calibration unit is used to fix the same-name regions of the point cloud data based on the geometric error calibration result and adopt an automatic calibration strategy of the same-name regions based on the maximum geometric error to complete the calibration of the multi-site equipment.

[0037] Preferably, the process of obtaining the plane coordinate calibration result includes:

[0038] The point cloud coordinates in different coordinate systems are used as source point cloud coordinates, and the point cloud coordinates in the global coordinate system are used as target point cloud coordinates;

[0039] Calculate the centroid of the source point cloud coordinates and the target point cloud coordinates, and perform centroid decentralization processing;

[0040] Intermediate variables based on the coordinates of the source point cloud and the target point cloud after decentralized processing;

[0041] Based on the intermediate variables, solving the rotation angle;

[0042] Calculating a translation amount based on the rotation angle and the intermediate variable;

[0043] Based on the scale transformation parameters, the rotation angle and the translation amount, different coordinate systems are converted to the same global coordinate system, and combined with the vertical axis calibration, the plane coordinate calibration is completed.

[0044] The present invention also provides a method for calibrating geometric errors of a three-dimensional laser scanning device for transportation infrastructure based on a target, and the method includes:

[0045] The RANSAC method is used to denoise the point cloud data of transportation infrastructure collected by 3D laser scanning equipment to obtain denoised point cloud data;

[0046] Based on the de-noised point cloud data and in combination with a standard plane target, a geometric median-Weiszfeld iterative algorithm is used to calibrate the plane error of the three-dimensional laser scanning device;

[0047] Based on the de-noised point cloud data and in combination with a standard spherical target, a particle swarm optimization algorithm is used to perform stereo error calibration on the three-dimensional laser scanning device;

[0048] Based on the plane error calibration results and stereo error calibration results of the multi-site 3D laser scanning equipment, combined with the four-parameter method and NDT optimization method, multi-site equipment calibration is carried out to complete the geometric error calibration of the target-based 3D laser scanning equipment for transportation infrastructure.

[0049] Preferably, the method for denoising the point cloud data of transportation infrastructure collected by a three-dimensional laser scanning device includes:

[0050] Based on the data volume and density of the point cloud data, the maximum number of iterations and distance threshold of the RANSAC method are preset to complete parameter initialization;

[0051] Based on the initialized parameters, three points in the point cloud data are randomly sampled to obtain the minimum set of plane fitting;

[0052] Based on the minimum set, a least square method is used to fit the plane equation;

[0053] Based on the plane equation, calculating the distances from all points in the point cloud data to the plane to obtain inliers and outliers;

[0054] Through iteration, the optimal plane parameters and the optimal in-plane points are obtained, and the abnormal points are eliminated to obtain the final plane fitting result, thereby completing the denoising of the point cloud data.

[0055] Preferably, the method for calibrating the plane error of the three-dimensional laser scanning device using a standard plane target includes:

[0056] Selecting the geometric median of the valid point set of the final plane fitting result;

[0057] Constructing an objective function based on the valid point set and the geometric median;

[0058] Performing Weiszfeld iteration based on the objective function to obtain a coordinate iteration result of the geometric median;

[0059] A threshold range is constructed based on the type of transportation infrastructure. When the Euclidean distance between two adjacent coordinate iteration results is less than the threshold range, the iteration is terminated to obtain the final coordinate of the geometric median, completing the plane error calibration of the 3D laser scanning device.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. Adaptive processing of complex road conditions:

[0062] The local structure fitting method based on RANSAC can still accurately extract crack geometric features when the noise ratio reaches 30%, with a false detection rate of ≤5%;

[0063] The Weiszfeld iterative algorithm reduces the influence of outliers caused by vibration, and the plane calibration error is ≤0.2mm.

[0064] 2. Efficient multi-site registration:

[0065] The particle swarm optimization algorithm achieves parallel computing, shortening the multi-site registration time to 8 minutes, with a registration accuracy error of ≤2.3mm (an improvement of 83.7% compared to traditional methods);

[0066] Dynamic threshold (ε=3√(Σd 2 / n))Automatically filter out regions with the same name, reducing manual intervention.

[0067] 3. Implementation of engineering standardization:

[0068] The calibration process is now fully compliant with the JTG F80 standard, and the crack depth measurement error has been reduced from 1.27mm to 0.31mm.

[0069] Supports mainstream equipment such as FARO and RIEGL, with a single on-site calibration time of ≤15 minutes.

[0070] 4. Cost and reliability optimization:

[0071] Adopt low-cost standard targets (unit price ≤ 500 yuan), reducing calibration costs by 60%;

[0072] The anti-interference design ensures measurement stability in an environment of -20℃ to 50℃, and the robustness is improved by 90%. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0074] Figure 1 This is a schematic structural diagram of a geometric error calibration system for a target-based three-dimensional laser scanning device for transportation infrastructure according to an embodiment of the present invention. DETAILED DESCRIPTION

[0075] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0076] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0077] Example 1

[0078] like Figure 1As shown in the figure, a geometric error calibration system for a 3D laser scanning device of traffic infrastructure based on a target includes: a denoising module, a planar error calibration module, a stereoscopic error calibration module, and a multi-site device calibration module.

[0079] The denoising module is used to denoise the point cloud data of traffic infrastructure collected by the 3D laser scanning device by using the RANSAC method to obtain denoised point cloud data; the abnormal points and noise points are eliminated by the RANSAC method to improve the data quality.

[0080] A further implementation manner is that the denoising module includes:

[0081] An initialization unit is used to preset the maximum number of iterations and distance threshold of the RANSAC method based on the data volume and density of the point cloud data to complete parameter initialization.

[0082] In this embodiment, the maximum number of iterations N: is set according to the data volume and needs to satisfy N≥log(1-p) / log(1-(1-∈) m ), where ∈ is the noise ratio and m is the minimum number of points of the model.

[0083] The distance threshold t: is adjusted according to the point cloud density (for example, 1.0 unit).

[0084] A simulation fitting unit is used to randomly sample 3 points in the point cloud data based on the initialized parameters to obtain the minimum set of plane fitting;

[0085] An equation fitting unit is used to fit the plane equation c1x + c2y + c3z + c4 = 0 by using the least squares method based on the minimum set, where c1, c2, c3, c4 represent the parameters of the plane equation, and x, y, z represent the spatial geometric coordinates of the plane equation.

[0086] <000025​​​​​​​​​​​​​​The iterative unit is used to obtain the optimal plane parameters and the optimal plane inliers through iteration, and to remove outliers to obtain the final plane fitting result and complete the denoising of the point cloud data. Specifically, the model with the largest number of inliers is retained and the final inlier point set is output. In this embodiment, a specific point cloud data denoising process is provided: Step 1: Input a noisy simulated point cloud data set: 50 inliers and 5 outliers, as shown in Table 1:

[0092] Table 1

[0093]

[0094] Step 2: Parameter Setting and Model Fitting Set N = 100, t = 1.0, and randomly sample 3 points to fit the plane. The sample code snippet for solving the plane parameters is as follows:

[0095]

[0096]

[0097] Step 3: Interior point determination and iterative optimization

[0098] Traverse all points to calculate the distance to the plane and determine the internal points and outliers.

[0099] After 100 iterations, the optimal plane parameters are (m, n, c) = (0.36, 0.51, 1.70), there are 50 points in the optimal plane, and the outliers are completely eliminated.

[0100] The above-mentioned denoising process of the present invention can achieve the following technical effects:

[0101] Local structure preservation: Accurately extract the 3D geometric features of structures and defects, suitable for scenarios such as scanning complex transportation infrastructure.

[0102] Strong anti-noise ability: when the noise ratio reaches 30%, it can still stably fit the main structure.

[0103] Scalability: Supports fitting of complex structural models such as spheres and surfaces.

[0104] A plane error calibration module, configured to calibrate the plane error of the three-dimensional laser scanning device using a geometric median-Weiszfeld iterative algorithm based on the de-noised point cloud data and in combination with a standard plane target;

[0105] In a further embodiment, the plane error calibration module includes:

[0106] The median acquisition unit is used to select the geometric median of the valid point set of the final plane fitting result; in this embodiment, the pre-processed point set is recorded as:

[0107] {P1,P2,...,P n}, P i =(x i ,y i ),

[0108] Its geometric median is recorded as:

[0109] M=(m x ,m y ).

[0110] The objective function calculation unit is used to construct an objective function based on the point set and the geometric median; a further embodiment is that the objective function is as follows:

[0111]

[0112] Where M=(m x ,m y ) represents the geometric median, {P1,P2,...,P n}, P i =(x i ,y i ) represents the valid point set.

[0113] The objective function minimization method must meet the following conditions:

[0114]

[0115] The coordinate iteration unit is used to perform Weiszfeld iteration based on the objective function to obtain the coordinate iteration result of the geometric median. In this embodiment, the iterative algorithm updates the formula calculation:

[0116]

[0117] Iterative coordinate calculation: The initial point of the iterative algorithm is calculated according to the central mean and recorded as M (0) =(m x (0) ,m y (0) ), the coordinate iteration formula is as follows:

[0118]

[0119] The convergence judgment unit is used to establish a threshold range based on the type of traffic infrastructure. When the Euclidean distance between two adjacent coordinate iteration results is less than the threshold range, the iteration is terminated, the final coordinate of the geometric median is obtained, and the plane error calibration of the three-dimensional laser scanning device is completed. In this embodiment, when the Euclidean distance between two adjacent iterative coordinates is less than the threshold The iteration is terminated when:

[0120]

[0121] According to the type of highway infrastructure, the threshold The value range of is:

[0122]

[0123] This embodiment provides a specific plane error calibration process:

[0124] Step 1: Target selection and data collection

[0125] The design standard area is 300 square centimeters, and the circular plane target with standard geometric center coordinates of (0mm, 0mm) must meet the requirements of the "JT / T 1037-2022" specification.

[0126] Target point cloud data was obtained through 3D laser scanning, and 99 typical point cloud data were selected as follows (unit: mm):

[0127] P＝{(101.2177,5.4838),(100.0022,-1.0791),(98.345,-7.5069),(96.2588,-13.77),(93.7584,-19.8395),(90.8604,-25.6883),(87.5833,-31.2902),(83.9474,-36.6208),(79.9743,-41.6571),(75.6872,-46.3779),(71.1108,-50.7637),(66.2707,-54.7971),(61.1938,-58.4624),(55.9077,-61.746),(50.441,-64.6364),(44.8226,-67.1242),(39.0821,-69.202),(33.2493,-70.8648),(28.8686,32.5902),(28.8471,29.8697),(28.5705,35.336),(28.4963,27.2024),(27.9614,38.08),(27.8046,24.6175),(27.3542,-72.1095),(27.0498,40.7958),(26.7556,22.1458),(25.8445,43.4582),(25.3244,19.8202),(24.3554,46.0425),(23.4694,17.6774),(22.593,48.525),(21.4266,-72.9353),(21.1137,15.7602),(20.5691,50.8829),(18.2967,53.0942),(18.0946,14.1233),(15.7899,55.1383),(15.4965,-73.3436),(13.9634,12.85),(13.0643,56.9955),(10.1363,58.6475),(9.5931,-73.3378),(7.0237,60.0771),(3.7455,12.1511),(3.7455,61.2687),(3.7455,-72.9233),(2.1000,2.4000),(1.4000,-0.2000),(0.8000,0.1000),(0.5000,0.3000),(0.3215,62.2082),(-0.1000,0.0000),(-0.4000,-1.3000),(-1.7000,1.5000),(-2.018,-72.1076),(-3.2274,62.8828),(-6.8796,63.2816),(-7.6697,-70.9004),(-10.6125,63.3953),(-13.1829,-69.313),(-14.403,63.2161),(-18.2277,62.7384),(-18.5321,-67.3585),(-22.0625,61.9582),(-23.6925,-65.0521),(-25.8831,60 .8731),(-28.6411,-62.4103),(-29.6651,59.4831),(-33.3558,-59.4512),(-33.3842,57.7895),(-37.016,55.7959),(-37.8163,-56.1944),(-40.5366,53.5076),(-42.0038,-52.661),(-43.9223,50.9317),(-45.9011,-48.8728),(-47.150 2,48.0771),(-49.4928,-44.8533),(-50.1977,44.9546),(-52.7653,-40.6263),(-53.0434,41.5767),(-55.6664,37.9573),(-55.7067,-36.2169),(-58.0472,34.1122),(-58.3073,-31.6506),(-60.1673,30.0586),(-60.5591,-26.9534),( -62.0094,25.8151),(-62.4561,-22.1519),(-63.5578,21.4017),(-63.9945,-17.2726),(-64.798,16.8396),(-65.1721,-12.3423),(-65.7174,12.1511),(-65.989,-7.3876),(-66.3047,7.3595),(-66.4472,-2.4352),(-66.5505,2.4891)}.

[0128] When only the traditional scanning method is used without plane error calibration, the target geometric center positioning error is measured as:

[0129]

[0130] The coordinate positioning error is:

[0131]

[0132] Step 2: Initial value calculation

[0133] Calculate the initial geometric center using the mean method:

[0134] M (0) =(0.5000,0.3500)mm,

[0135] Step 3: Iterative process

[0136] Iteration 1: Calculate the distance from each point to the initial center (partial example):

[0137]

[0138] Update coordinates:

[0139] M (1) =(0.4820,0.3180)mm.

[0140] Step 4: Convergence determination

[0141] Setting thresholds The number of iterations is limited to 100, and Weiszfeld iterative optimization is performed. The code snippet is as follows:

[0142]

[0143]

[0144] In this embodiment, the final number of iterations is 3, and the detailed iteration process is as follows:

[0145] Iteration 1: X = 0.4993 mm, Y = 0.2188 mm,

[0146] Iteration 2: X = -0.1234 mm, Y = 0.0456 mm,

[0147] Iteration 3: X = -0.0123 mm, Y = 0.0056 mm,

[0148] Final coordinates: (-0.0123mm, 0.0056mm),

[0149] The final coordinate error is 0.0135mm, achieving sub-millimeter precision calibration and meeting the JTG F80 specification requirement of a road surface flatness error of ≤0.5mm.

[0150] The above-mentioned plane error calibration process of the present invention can achieve the following technical effects:

[0151] Anti-interference: The geometric median algorithm reduces sensitivity to outliers and improves robustness;

[0152] High precision: the error reaches sub-millimeter level after iterative convergence;

[0153] Engineering applicability: Through dynamic adjustment of thresholds, it can adapt to multiple application scenarios such as bridges, slopes, and tunnels.

[0154] The stereo error calibration module is used to perform stereo error calibration on the three-dimensional laser scanning device based on the denoised point cloud data and in combination with a standard spherical target, using a particle swarm optimization algorithm to improve the data accuracy of the three-dimensional point cloud of the device.

[0155] Traditional least squares methods are sensitive to interference from complex environments and working conditions and are prone to falling into local optima, affecting the accuracy of three-dimensional structure measurement. Before actual implementation, spherical targets are used to calibrate the stereoscopic errors of structure scanning equipment to improve equipment accuracy.

[0156] In a further embodiment, the stereo error calibration module includes:

[0157] The fitness calculation unit is used to calculate the fitness of each particle at the current position by evaluating the position quality of each encoded particle; wherein each particle represents a potential solution, including the center coordinates and radius of the standard spherical target, that is, (center coordinates + radius).

[0158] Fitness:

[0159] f(a,b,c,r)=-∑ i [(x i -a) 2 +(y i -b) 2 +(z i -c) 2 -r 2 ] 2 .

[0160] The optimal parameter acquisition unit is used to update the particle speed and position based on the fitness to obtain the optimal spherical parameters. In this embodiment, the speed update formula is:

[0161] ν i (t+1)=ων i (t)+c1γ1[p best -x i (t)]+c2γ2[g best -x i (t)],

[0162] Among them, ω=0.5 is the inertia weight, c1=c2=1.5 is the learning factor, γ1,γ1∈[0,1] are random numbers, p best represents the best historical position of particle i, g best represents the global best position.

[0163] Position update formula:

[0164] x i (t+1)=x i (t)+ν i (t+1).

[0165] The stereo error calibration unit is used to fit a sphere based on optimal spherical parameters to complete the stereo error calibration of the three-dimensional laser scanning device.

[0166] This embodiment provides a specific stereo calibration process:

[0167] Step 1: Experimental Setup

[0168] Target specifications: Standard spherical target, theoretical radius r = 75mm.

[0169] Data acquisition: Use a 3D laser scanner to obtain target point cloud data.

[0170] Particle swarm parameters: number of particles 200, maximum iteration 100 times, ω = 0.5, c1 = c2 = 1.5.

[0171] Step 2: Algorithm Implementation

[0172] The Python core code is as follows:

[0173]

[0174]

[0175] Step 3: Optimize parameters and calculate errors

[0176] After iterative calculation and optimization:

[0177] Geometric center and error:

[0178] P0(x0,y0,z0)=(0.0730,0.1103,0.0400)mm=φ(P0),

[0179] Radius and error:

[0180] r=74.9426mm; φ(r)=0.0574mm,

[0181] The traditional error is as follows:

[0182]

[0183] The above-mentioned stereo error calibration process of the present invention can achieve the following technical effects:

[0184] Global optimal solution search: The particle swarm algorithm uses swarm intelligence to dynamically search within the multidimensional solution space to avoid falling into the local optimum. The final geometric center error is 0.0574mm, and the radius error is 0.0574mm. Compared with the traditional least squares method (error 3.5553mm), the accuracy is significantly improved.

[0185] Strong robustness and adaptability: The algorithm is insensitive to initial parameters. Even when the initial value of the sphere center deviates greatly (±10mm), it can still converge to the optimal solution through group collaboration, making it suitable for complex field environments.

[0186] Parallel computing potential: The particle swarm algorithm's group update mechanism supports parallel processing, which can significantly shorten the calibration time of large-scale point cloud data and improve the overall work efficiency of the equipment.

[0187] Multi-scenario applicability: The particle swarm calibration method is applicable to high-precision measurement of geometric parameters of various types of infrastructure structures (such as pavements, slopes, bridges, etc.), providing a universal calibration compensation solution.

[0188] The multi-site equipment calibration module is used to perform multi-site equipment calibration based on the planar error calibration results and stereo error calibration results of multi-site 3D laser scanning equipment, combining the four-parameter method with the NDT optimization method. This module completes the geometric error calibration of target-based 3D laser scanning equipment for transportation infrastructure, and improves the accuracy of point cloud data through algorithm compensation.

[0189] In a further embodiment, the multi-site device calibration module includes:

[0190] The plane coordinate calibration unit is used to unify the point cloud data in different coordinate systems scanned by multiple sites into a global coordinate system based on the four-parameter method to obtain the plane coordinate calibration results.

[0191] The four-parameter method is used to calibrate the plane coordinates. The formula is as follows:

[0192]

[0193] Among them, α is the scale transformation parameter, θ is the rotation angle, and t x and t y is the translation amount.

[0194] A further embodiment is that the process of obtaining the plane coordinate calibration result includes:

[0195] Extract matching point pairs and retain only 2D plane coordinates, use the point cloud coordinates in different coordinate systems as the source point cloud coordinates, and use the point cloud coordinates in the global coordinate system as the target point cloud coordinates; Point pairs with the same name: Target point cloud coordinates: P s,i =(x s,i ,y s,i ), source point cloud coordinates: P t,i =(x i ,y i ).

[0196] Calculate the centroid of the source point cloud coordinates and the target point cloud coordinates And perform centroid decentralization processing;

[0197] Intermediate variables based on the source point cloud coordinates and target point cloud coordinates after decentralized processing:

[0198] C1=∑(x t x s +y t y s ); C2=∑(y t x s -x t y s )

[0199] C3=∑x s ; C4=∑y s ; C5=∑x t ; C6=∑y t ;

[0200] Based on the intermediate variables, solve the rotation angle: θ = arctan2(C2, C1);

[0201] Calculate the translation based on the rotation angle and the intermediate variable:

[0202] Based on the scale transformation parameters, rotation angle and translation amount, different coordinate systems are converted to the same global coordinate system, and combined with the vertical axis calibration, the plane coordinate calibration is completed.

[0203] Vertical axis calibration: Pass Compensate for elevation errors.

[0204] The geometric error calibration unit is used to optimize the plane coordinate calibration result using the NDT method to obtain a geometric error calibration result. In this embodiment, in order to improve the calibration accuracy and efficiency, the NDT (Normal Distributions Transform) method is used to optimize the initial plane coordinate calibration result:

[0205] Meshing: Divide the target point cloud into a fixed-size cubic grid

[0206] Probability density modeling: Calculate the mean μ and covariance matrix Γ of the point cloud within each grid:

[0207]

[0208] Optimization goal: maximize the overall likelihood function

[0209]

[0210] Iterative optimization: Update the transformation matrix using the Levenberg-Marquardt algorithm until the difference between two adjacent transformations is less than a threshold:

[0211]

[0212] The homonymous region calibration unit is used to fix homonymous regions in point cloud data based on the geometric error calibration results and adopt the homonymous region automatic calibration strategy based on the maximum geometric error to complete the calibration of multi-site equipment. Specifically, in order to reduce manual intervention, the homonymous region automatic calibration strategy based on the maximum geometric error is proposed:

[0213] Distance calculation: Calculate the maximum Euclidean distance between the source point cloud and the target point cloud

[0214]

[0215] Threshold setting: filter to meet d i ≤ε, where the threshold

[0216] KDTree acceleration: Use the KDTree structure of the Open3D library to quickly match point clouds and improve computing efficiency.

[0217] This example provides a specific engineering implementation on-site application and multi-site calibration process:

[0218] A standard road section at a testing site in Beijing was selected for multi-site scanning testing:

[0219] Equipment configuration: FARO Focus X Series medium- and long-range laser scanner, laser wavelength 1550nm, ambient temperature on that day was 16.8°C.

[0220] Data acquisition: Four scanning stations were deployed to cover a 10m × 10m × 3m test space to acquire raw point cloud data.

[0221] Preprocessing: Remove noise points using the RANSAC method.

[0222] Plane calibration: Combined with a standard plane target, the geometric median-Weiszfeld iterative algorithm is used to calibrate the plane error of the device.

[0223] Stereo calibration: Combined with a standard spherical target, the particle swarm optimization algorithm is used to calibrate the stereo error of the device.

[0224] Multi-site equipment calibration: The coordinate system is calibrated using the four-parameter method to perform coarse registration of multi-site equipment data; NDT is used to optimize point cloud alignment for fine registration of multi-site equipment data.

[0225] Specific multi-site calibration process:

[0226] Step 1: Experimental Setup

[0227] Test object: Use laser scanning equipment to scan a pothole on the test road section

[0228] Parameter comparison: The geometric parameters of the pits were measured using traditional methods before calibration. A high-level standard instrument (Renishaw XL-80 laser interferometer, certified by the National Institute of Metrology, CDlx2024-01568) was used to calculate the measurement error of the traditional method. The area calculation error was 0.0057m 2 , the circumference calculation error is 14.3169mm, and the depth calculation error is 1.2733mm.

[0229] NDT parameter settings: voxel size is set to 0.01, maximum number of iterations is set to 100, and Open3D parallel acceleration is enabled.

[0230] Step 2: Algorithm Implementation

[0231] The Python core code is as follows:

[0232]

[0233]

[0234]

[0235] Step 3: Comparison of measured parameters

[0236] After data preprocessing, plane error calibration, stereo error calibration and multi-site error calibration, the pit compensation morphology is obtained.

[0237] The area calculation error is 0.0015m 2 (73.68% optimized compared to before calibration), the circumference calculation error is 2.3285mm (83.74% optimized compared to before calibration), and the depth calculation error is 0.3126mm (accuracy is 75.45% optimized compared to before calibration).

[0238] The multi-site calibration method of the present invention can achieve the following technical effects:

[0239] Accuracy verification: The calibrated data meets the JTG F80 specification (road surface flatness error ≤ 0.5mm), and the three-dimensional geometric parameter errors of the structure are reduced.

[0240] Improved efficiency: Multi-site registration time (approximately 8 minutes) is shortened to 60% of traditional methods.

[0241] Example 2

[0242] The present invention also provides a target-based method for calibrating geometric errors of a three-dimensional laser scanning device for transportation infrastructure, using the system described in the first embodiment, including:

[0243] The RANSAC method is used to denoise the point cloud data of transportation infrastructure collected by 3D laser scanning equipment to obtain denoised point cloud data;

[0244] Based on the denoised point cloud data and combined with the standard plane target, the geometric median-Weiszfeld iterative algorithm is used to calibrate the plane error of the 3D laser scanning equipment;

[0245] Based on denoised point cloud data and combined with a standard spherical target, the particle swarm optimization algorithm is used to calibrate the stereo error of the 3D laser scanning equipment.

[0246] Based on the plane error calibration results and stereo error calibration results of the multi-site 3D laser scanning equipment, combined with the four-parameter method and NDT optimization method, multi-site equipment calibration is carried out to complete the geometric error calibration of the target-based 3D laser scanning equipment for transportation infrastructure.

[0247] A further embodiment is that the method for denoising point cloud data of transportation infrastructure collected by a three-dimensional laser scanning device includes:

[0248] Based on the data volume and density of the point cloud data, the maximum number of iterations and distance threshold of the RANSAC method are preset to complete parameter initialization;

[0249] Based on the initialized parameters, three points in the point cloud data are randomly sampled to obtain the minimum set of plane fitting;

[0250] Based on the minimum set, the least square method is used to fit the plane equation;

[0251] Based on the plane equation, the distance between all points in the point cloud data and the plane is calculated to obtain the inliers and outliers;

[0252] Through iteration, the optimal plane parameters and the optimal in-plane points are obtained, and abnormal points are eliminated to obtain the final plane fitting result and complete the denoising of the point cloud data.

[0253] A further embodiment is that the method for calibrating the plane error of a three-dimensional laser scanning device using a standard plane target includes:

[0254] Select the geometric median of the valid point set of the final plane fitting result;

[0255] Construct the objective function based on the point set and geometric median;

[0256] Perform Weiszfeld iteration based on the objective function to obtain the coordinate iteration result of the geometric median;

[0257] A threshold range is constructed based on the type of transportation infrastructure. When the Euclidean distance between two adjacent coordinate iteration results is less than the threshold range, the iteration is terminated, the final coordinate of the geometric median is obtained, and the plane error calibration of the 3D laser scanning equipment is completed.

[0258] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A target-based geometric error calibration system for three-dimensional laser scanning equipment of transportation infrastructure, characterized in that: include: A denoising module is used to denoise the point cloud data of traffic infrastructure collected by a 3D laser scanning device using a RANSAC method to obtain denoised point cloud data; A plane error calibration module, configured to calibrate the plane error of the three-dimensional laser scanning device using a geometric median-Weiszfeld iterative algorithm based on the de-noised point cloud data and in combination with a standard plane target; A stereo error calibration module, configured to perform stereo error calibration on the three-dimensional laser scanning device based on the denoised point cloud data and in combination with a standard spherical target using a particle swarm optimization algorithm; The multi-site equipment calibration module is used to perform multi-site equipment calibration based on the planar error calibration results and stereo error calibration results of multi-site 3D laser scanning equipment, combining the four-parameter method with the NDT optimization method, to complete the geometric error calibration of target-based 3D laser scanning equipment for transportation infrastructure.

2. The system according to claim 1, wherein: The denoising module includes: The initialization unit is used to preset the maximum number of iterations and distance threshold of the RANSAC method based on the data volume and density of the point cloud data to complete parameter initialization; The simulation fitting unit is used to randomly sample 3 points in the point cloud data based on the initialized parameters to obtain the minimum set of plane fitting; an equation fitting unit, configured to fit the plane equation using a least squares method based on the minimum set; an interior point determination unit, configured to calculate the distances from all points in the point cloud data to the plane based on the plane equation, and obtain interior points and outliers; The iterative unit is used to obtain the optimal plane parameters and the optimal plane points through iteration, and to remove the abnormal points to obtain the final plane fitting result and complete the denoising of the point cloud data.

3. The system according to claim 2, characterized in that The plane error calibration module includes: a median obtaining unit, configured to select the geometric median of the valid point set of the final plane fitting result; An objective function calculation unit, configured to construct an objective function based on the valid point set and the geometric median; A coordinate iteration unit, configured to perform Weiszfeld iteration based on the objective function to obtain a coordinate iteration result of a geometric median; The convergence judgment unit is used to construct a threshold range based on the type of transportation infrastructure. When the Euclidean distance between two adjacent coordinate iteration results is less than the threshold range, the iteration is terminated, the final coordinate of the geometric median is obtained, and the plane error calibration of the three-dimensional laser scanning equipment is completed.

4. The system according to claim 3, characterized in that The objective function is as follows: Where M=(m x ,m y ) represents the geometric median, {P1,P2,...,P n }, P i =(x i ,y i ) represents the valid point set.

5. The system according to claim 3, wherein: The stereo error calibration module includes: a fitness calculation unit, configured to calculate the fitness of each particle at its current position by evaluating the position quality of each encoded particle; wherein each particle represents a potential solution, including the coordinates of the center and radius of a standard spherical target; An optimal parameter acquisition unit, configured to update the particle velocity and position based on the fitness to obtain optimal spherical parameters; The stereo error calibration unit is used to fit a sphere based on optimal spherical parameters to complete the stereo error calibration of the three-dimensional laser scanning device.

6. The system according to claim 3, wherein: The Multi-Site Equipment Calibration Module includes: The plane coordinate calibration unit is used to unify the point cloud data in different coordinate systems scanned by multiple sites into a global coordinate system based on the four-parameter method to obtain the plane coordinate calibration results; A geometric error calibration unit, configured to optimize the plane coordinate calibration result by using an NDT method to obtain a geometric error calibration result; The same-name region calibration unit is used to fix the same-name regions of the point cloud data based on the geometric error calibration result and adopt an automatic calibration strategy of the same-name regions based on the maximum geometric error to complete the calibration of the multi-site equipment.

7. The system according to claim 6, characterized in that The process of obtaining plane coordinate calibration results includes: The point cloud coordinates in different coordinate systems are used as source point cloud coordinates, and the point cloud coordinates in the global coordinate system are used as target point cloud coordinates; Calculate the centroid of the source point cloud coordinates and the target point cloud coordinates, and perform centroid decentralization processing; Intermediate variables based on the coordinates of the source point cloud and the target point cloud after decentralized processing; Based on the intermediate variables, solving the rotation angle; Calculating a translation amount based on the rotation angle and the intermediate variable; Based on the scale transformation parameters, the rotation angle and the translation amount, different coordinate systems are converted to the same global coordinate system, and combined with the vertical axis calibration, the plane coordinate calibration is completed.

8. A target-based method for calibrating geometric errors of a three-dimensional laser scanning device for transportation infrastructure, using the system according to any one of claims 1 to 7, characterized in that: include: The RANSAC method is used to denoise the point cloud data of transportation infrastructure collected by 3D laser scanning equipment to obtain denoised point cloud data; Based on the de-noised point cloud data and in combination with a standard plane target, a geometric median-Weiszfeld iterative algorithm is used to calibrate the plane error of the three-dimensional laser scanning device; Based on the de-noised point cloud data and in combination with a standard spherical target, a particle swarm optimization algorithm is used to perform stereo error calibration on the three-dimensional laser scanning device; Based on the plane error calibration results and stereo error calibration results of the multi-site 3D laser scanning equipment, combined with the four-parameter method and NDT optimization method, multi-site equipment calibration is carried out to complete the geometric error calibration of the target-based 3D laser scanning equipment for transportation infrastructure.

9. The method according to claim 8, characterized in that Methods for denoising point cloud data of transportation infrastructure collected by 3D laser scanning equipment include: Based on the data volume and density of the point cloud data, the maximum number of iterations and distance threshold of the RANSAC method are preset to complete parameter initialization; Based on the initialized parameters, three points in the point cloud data are randomly sampled to obtain the minimum set of plane fitting; Based on the minimum set, a least square method is used to fit the plane equation; Based on the plane equation, calculating the distances from all points in the point cloud data to the plane to obtain inliers and outliers; Through iteration, the optimal plane parameters and the optimal in-plane points are obtained, and the abnormal points are eliminated to obtain the final plane fitting result, thereby completing the denoising of the point cloud data.

10. The method according to claim 9, characterized in that The method for calibrating the plane error of the three-dimensional laser scanning device using a standard plane target includes: Selecting the geometric median of the valid point set of the final plane fitting result; Constructing an objective function based on the valid point set and the geometric median; Performing Weiszfeld iteration based on the objective function to obtain a coordinate iteration result of the geometric median; A threshold range is constructed based on the type of transportation infrastructure. When the Euclidean distance between two adjacent coordinate iteration results is less than the threshold range, the iteration is terminated to obtain the final coordinate of the geometric median, completing the plane error calibration of the 3D laser scanning device.

Images