A non-contact method for detecting a friction coefficient of a pavement

Abstract

The application discloses a non-contact pavement friction coefficient detection method, which comprises the following steps: obtaining a plurality of pavement point clouds by performing three-dimensional texture scanning on a pavement through a laser profilometer and a visible light camera; performing three-dimensional texture reconstruction on the pavement point clouds to obtain a space curve; extracting an index set of apparent texture features from the space curve; collecting friction data through a pavement friction test vehicle; and establishing a mapping relationship between the index set and the friction coefficient. The detection method realizes three-dimensional reconstruction of pavement micro-texture and macro-texture through two non-contact detection methods of a laser profilometer and a visible light camera, and then establishes a mapping relationship between a friction coefficient value collected by a traditional pavement friction test vehicle and the index set. The index set and the mapping relationship obtained through the method can be used to directly obtain the friction coefficient of the pavement.

Description

Technical Field

[0001] This invention relates to the detection of the friction coefficient of pavement, specifically to a non-contact method for detecting the friction coefficient of pavement. Background Technology

[0002] The anti-skid capability of a road surface is not determined solely by the condition of the tire or the road surface itself, but rather by the interaction between the two. Factors affecting road surface anti-skid capability include: the road surface's inherent characteristics (texture), contact surface characteristics (dry / wet, road surface contamination), and tire condition (speed, tread pattern, tire pressure, and slip ratio). Among these, micro-texture (1μm–1mm) primarily affects the coefficient of friction in a dry road surface, while macro-texture (1mm–10mm) affects the coefficient of friction in a wet road surface. Accurately determining the coefficient of friction of a road surface based on these factors is a pressing technical problem that needs to be solved. Summary of the Invention

[0003] The purpose of this invention is to provide a non-contact method for detecting the coefficient of friction of pavement. This method uses two non-contact detection methods, namely laser construction depth measurement and visible light camera, to achieve three-dimensional reconstruction of the micro and macro textures of the pavement. Then, it establishes a mapping relationship between the friction coefficient values ​​collected by the traditional pavement friction test vehicle and the index set. The friction coefficient of the pavement can be directly obtained through the index set and mapping relationship obtained by the method of this invention.

[0004] The technical solution adopted to achieve the purpose of this invention is a non-contact method for detecting the coefficient of friction of pavement, which includes:

[0005] S1. Perform three-dimensional texture scanning on the pavement to obtain multiple pavement point clouds;

[0006] S2. Perform three-dimensional texture reconstruction on the point cloud of the track surface to obtain a spatial surface;

[0007] S3. Extract an index set of apparent texture features from the spatial surface;

[0008] S4. Collect friction data using a pavement friction test vehicle and establish a mapping relationship between the index set and the friction coefficient.

[0009] In the above technical solution, multiple track point clouds with micro-textures are obtained by scanning the track surface with a laser-constructed depth meter, and multiple track point clouds with macro-textures are obtained by scanning the track surface with a visible light camera.

[0010] In the above technical solution, the pavement point cloud is assembled into complete three-dimensional pavement information through a stitching algorithm.

[0011] In the above technical solution, a three-dimensional Delaunay triangulation surface reconstruction algorithm is used to reconstruct the three-dimensional information of the pavement to obtain a spatial surface.

[0012] In the above technical solution, feature analysis is performed on the point cloud data in the spatial curved surface to extract a set of indicators, including surface normal vector, MTD construction depth, average spacing of contour single peaks, and fractal dimension of three-dimensional texture morphology.

[0013] In the above technical solution, cluster analysis and regression analysis are performed on the index set and friction coefficient data to establish a mapping relationship between the index set and the friction coefficient.

[0014] Furthermore, the aforementioned non-contact pavement friction coefficient detection method also includes: using rough set analysis to target the correlation with the friction coefficient, and adjusting the cluster analysis algorithm parameters to minimize the sum of squared fitting errors.

[0015] Furthermore, the above-mentioned non-contact pavement friction coefficient detection method also includes: using the energy function of a Boltzmann machine to deeply learn the joint probability distribution of the index set and the friction coefficient, and judging the accuracy of the mapping relationship based on the joint probability distribution.

[0016] This invention first employs a mobile platform consisting of a pavement laser-based depth measurement system and a 2D visible light camera—both non-contact detection methods—to continuously scan the pavement. This allows the point cloud to be converted to the same coordinate system for data correction. The microscopic texture information from the laser-based depth measurement is then filtered for noise reduction using a Kalman filter, while the macroscopic texture information from the visible light camera scan is filtered using a fixed threshold to remove the influence of small grayscale values ​​on the pavement texture. Next, the corrected data from both sources are combined with the features and spatial relationships of the point cloud to perform precise registration and achieve 3D reconstruction. Finally, the 3D reconstructed pavement image is used to calculate key geometric parameters and extract a set of relevant feature indices, thus establishing a mapping relationship between the friction coefficient values ​​collected by traditional pavement friction testing vehicles and the indices. In the actual operation of the friction testing system, only the indices need to be obtained using this invention's method, and then the pavement friction coefficient can be calculated using the obtained mapping relationship.

[0017] Compared to traditional professional pavement friction testing vehicles that require on-site measurements by professionals and specialized equipment, the method of this invention only requires a laser-constructed depth gauge and a visible light camera. The required equipment is simple, economical, and easy to operate, eliminating the need for professional measurement personnel. This not only saves on the costs of traditional measurements but also ensures high accuracy. Attached Figure Description

[0018] Figure 1 This is a flowchart of a non-contact method for detecting the coefficient of friction of a road surface according to the present invention.

[0019] Figure 2 This is a schematic diagram of the camera's original displacement data.

[0020] Figure 3 To Figure 2 A schematic diagram of the displacement data after Kalman filtering correction.

[0021] Figure 4 This is a schematic diagram of the point cloud before registration.

[0022] Figure 5 This is a schematic diagram of the registered point cloud.

[0023] Figure 6 This is a schematic diagram of the Delaunay triangulation process.

[0024] Figure 7 This is a schematic diagram of the 3D reconstruction model.

[0025] Figure 8 This is a schematic diagram of the surface normal vector.

[0026] Figure 9 This is a schematic diagram of the average spacing between the single peaks of the profile. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0028] like Figure 1 As shown, the non-contact pavement friction coefficient detection method of the present invention includes:

[0029] S1. Perform a three-dimensional texture scan on the pavement.

[0030] This invention obtains multiple point clouds of the track surface with microscopic texture by scanning the track surface using a laser-based depth sensor, and obtains multiple point clouds of the track surface with macroscopic texture by scanning the track surface using a visible light camera. The two scanning methods are described below:

[0031] A laser texture depth gauge scans the pavement surface. This instrument uses laser ranging to measure the depth variations between and on the surface of material particles. Vehicle-mounted laser texture depth gauges are intelligent instruments suitable for determining pavement texture depth under normal driving conditions, without severe damage, water accumulation, snow accumulation, or mud. Currently, laser texture depth gauges on the market are relatively mature, with a measurement resolution of up to 0.02 mm, sufficient for measuring the microscopic texture of the pavement surface. The microscopic texture information obtained from the laser texture depth scan is then filtered by a Kalman filter to eliminate noise.

[0032] When scanning the pavement surface with a visible light camera to perform macroscopic texture scanning, the camera should first be fixed with its plane approximately parallel to the surface, ensuring a consistent distance between the camera and the surface. Considering camera accuracy, the camera is typically mounted at a height of 50-80 cm above the pavement. Generally, with a selected visible light camera and a fixed distance, if the quality of the acquired surface information is high, the scanning range will be small. In this case, continuous scanning by moving the camera and subsequent point cloud registration can be used to address the issue. To meet the needs of outdoor mobile acquisition of pavement surface 3D information, a visible light camera can be integrated into an embedded development board. Using an embedded development board not only reduces the size and power consumption of the scanning equipment, but also allows for better consistency and performance in programs developed based on it. Embedded development boards have a single function and a simple, consistent system architecture, making them ideal as edge computing devices for pavement 3D scanning. The macroscopic texture information scanned by the visible light camera is filtered using a fixed threshold to remove the influence of small grayscale values ​​on the pavement texture structure information.

[0033] S2. The spatial surface is obtained by reconstructing the three-dimensional texture of the point cloud obtained in S1.

[0034] S2.1. The pavement point cloud is assembled into complete three-dimensional pavement information using a stitching algorithm.

[0035] Because the scanning range is limited each time, multiple point clouds acquired from different locations need to be stitched together to form complete 3D information. The point clouds acquired in each frame are independent, and their 3D coordinates are calculated relative to the coordinate system at that moment. Therefore, the point cloud coordinates of each frame need to be transformed to the world coordinate system based on their spatial location before point cloud stitching can be performed. Taking macroscopic textures captured by the camera as an example, the changes in the camera's pose during the shooting process can be calculated from the changes in each frame of the video. During shooting, the world coordinate system is used by default at the first frame, and the displacement of each subsequent frame is the displacement of the camera's coordinate system relative to the world coordinate system at that moment.

[0036] The camera translation vector for each frame is t = [t1, t2, t3]. T And the Euler angle vector o = [r, p, y] T Transform to the world coordinate system and plot the camera's displacement curve, such as... Figure 2 As shown, the original displacement data contains many abrupt anomalies and cannot be directly used for coordinate transformation of point cloud data.

[0037] The camera pose satisfies the Markov property, meaning the pose at time k is only related to the poses at times k-1 and is independent of previous poses. To eliminate the influence of noise during observation, Kalman filtering can be used to optimally estimate the motion state. Kalman filtering is an algorithm that uses the state equations of a linear system, combined with system observation data, to optimally estimate the system state. The optimization results of the camera motion trajectory using Kalman filtering are shown below. Figure 3 As shown in the figure, it can be seen that the Kalman filter effectively filters noise.

[0038] Using the corrected displacement data, the point cloud can be transformed to the same coordinate system. Although the Kalman filter can effectively eliminate noise, the corrected point cloud still retains a small displacement deviation, such as... Figure 4 As shown, it is necessary to combine the characteristics of the point cloud to perform fine registration of the point cloud.

[0039] The ICP algorithm directly selects the two closest points in the point cloud to be matched as corresponding points through a straight-line search. This is a brute-force exhaustive algorithm, heavily reliant on the initial displacement of the point cloud. When the initial distance between the two point clouds is large, convergence is likely to occur. Because the features of plate point clouds are relatively simple, directly registering two original point clouds will also result in significant registration errors. This is why a coordinate transformation of the original point cloud is necessary before using the ICP algorithm. After point cloud registration using the ICP algorithm, as shown... Figure 5 As shown.

[0040] S2.2 The three-dimensional information of the pavement is reconstructed using a three-dimensional Delaunay triangulation surface reconstruction algorithm to obtain a spatial surface.

[0041] like Figure 6 As shown, the steps of using three-dimensional Delaunay triangulation in this embodiment include:

[0042] (1) Randomly select 4 non-coplanar points in the point cloud set P;

[0043] (2) Take any new point Pi∈P, and determine the positional relationship between Pi and the tetrahedron formed by the 4 points in step (1):

[0044] a) If Pi is outside the tetrahedron, calculate the distances from Pi to the four faces of the tetrahedron, and select the closest face to form a new tetrahedron with Pi, such as... Figure 6 As shown in (a);

[0045] b) If Pi is inside the tetrahedron, divide Pi and the four faces of the tetrahedron into four new tetrahedrons, such as... Figure 6 As shown in (b);

[0046] c) If Pi lies on one face of a tetrahedron, divide the triangle containing Pi into three new triangles, and then, together with another vertex outside the face, divide the tetrahedron into three new tetrahedrons. Figure 6 As shown in (c);

[0047] d) If Pi lies on one of the edges of the tetrahedron, divide the tetrahedron into two new tetrahedrons based on the edge containing Pi, such as... Figure 6 As shown in (d);

[0048] (3) Check if there are isolated points. If so, proceed to step (2) to continue; otherwise, end the process. The three-dimensional reconstructed spatial surface model is as follows: Figure 7 As shown.

[0049] S3. Extract the apparent texture features from the spatial surface;

[0050] Feature analysis is performed on the point cloud data in the spatial curved surface to extract a set of indicators, including surface normal vector, MTD construction depth, average spacing of contour unimodal peaks, and fractal dimension of three-dimensional texture morphology.

[0051] Surface normal vectors are important properties of road surfaces. Accurate 3D reconstruction of road surfaces also relies on normal vectors. A common method for calculating normal vectors is based on local surface fitting, assuming the sampling plane of the point cloud is smooth everywhere. Therefore, the local neighborhood of any point can be fitted using a local plane. The normal vector can be obtained by solving the eigenvalue vectors and principal component analysis of the eigenvalues ​​of the point cloud within the target point's neighborhood. Figure 8 As shown.

[0052] That is, a point cloud surface mesh is constructed based on the data of points in the 3D point cloud data and their neighborhood information, and then the surface normal is calculated from the mesh.

[0053] Construction Depth (MTD) is one of the important indicators reflecting the skid resistance of road surfaces in current specifications. Here, MTD refers to the arithmetic mean of the total distances from all points on the profile to the peak line of the profile within the sampling range, as shown in the following formula:

[0054]

[0055] In other words, it is the arithmetic mean of the elevation differences between each point within the sampling range and the highest point within the range in the three-dimensional point cloud data.

[0056] The average spacing S of single peaks in a two-dimensional profile is defined as the average of the single-peak spacing (the projected length of the distance between the highest points of two adjacent single peaks on the midline) within the sampling length of the profile. For three-dimensional space, the corresponding length average needs to be replaced with the area average, such as... Figure 9 As shown.

[0057] For the three-dimensional reconstruction model of the road surface, that is, within the unit sampling range, the average value of the single peak spacing within the range is calculated. The average single peak spacing of the profile is the main evaluation parameter of the transverse information of the profile surface, which characterizes the peak density of the profile and is of great significance for evaluating the stability and wear resistance of the profile.

[0058] The specific method for calculating the dimension of the fractal box in a 3D texture is as follows: First, construct some cubes with side length L and use them to cover the contour surface of the 3D shape. Calculate the number N of boxes that intersect the contour surface for different side lengths L. L For the 3D reconstruction model of the pavement, cube boxes with side length L are set on the plane of the perpendicular projection of the 3D curved surface. That is, the entire projection plane is divided into an L×L grid. Cubes are stacked on each grid until they just intersect the curved surface, and the number of cubes on each grid is calculated. Finally, the counts of all boxes are summed to obtain N. L .

[0059] Plot logN L -1 / logL double logarithmic curve, the fractal dimension D is the slope of the curve in the double logarithmic coordinate system.

[0060]

[0061] S4. Extract an index set based on the extracted surface texture features, collect friction data using a pavement friction test vehicle, and establish a mapping relationship between the index set and the friction coefficient.

[0062] This embodiment establishes a mapping relationship between the index set and the friction coefficient by performing cluster analysis and regression analysis on the index set and friction coefficient data.

[0063] The friction coefficient of a running track is related to the material of the contacting object, the smoothness of the surface, the degree of dryness or wetness, the surface temperature, and the relative speed. Two to three asphalt and cement running tracks with different functional areas were selected as test scenarios. Two measurement experiments were conducted simultaneously under different working conditions: one used a traditional pavement friction testing vehicle to collect friction coefficient values; the other used the method of this invention to analyze the apparent texture features at the same location and extract a set of indicators, including surface normal vector, MTD construction depth, average spacing of contour peaks, and fractal dimension of three-dimensional texture morphology. Using Python or MATLAB software, cluster analysis and regression analysis were performed on the indicator set and friction coefficient data to establish a mapping relationship between the indicator set and the friction coefficient. The specific operation process is as follows:

[0064] First, cluster analysis is applied to the indicator set data. Cluster analysis is an unsupervised learning process; during classification, it is not necessary to predefine a classification criterion. Cluster analysis can automatically classify data based on the sample data. Through cluster analysis, the distribution of the indicator set data is obtained, the characteristics of each cluster are observed, and further analysis is conducted on specific clusters.

[0065] For each cluster of indexes, a multiple regression model is established, with the dependent variable Y (measured pavement friction coefficient) and the k independent variables affecting the dependent variable being X1, X2, ..., X... k (Surface normal vector, MTD construction depth, average spacing of contour unimodal peaks, fractal dimension of 3D texture topography), assuming that the influence of each independent variable on the dependent variable Y is linear, that is, with other independent variables remaining constant, the mean of Y changes with the independent variable X. i The change is uniform, and we have an overall regression model, as shown in equation (3):

[0066] Y = β0 + β1X1 + β2X2 + ... + β k X k +ε (3)

[0067] Where β0,β1,β2,...,β k These are the regression parameters.

[0068] Regression analysis has three basic tasks: 1) estimating model parameters using sample data; 2) testing the model parameters for hypotheses; and 3) applying the regression model to predict the corresponding variable (surface friction coefficient).

[0069] Taking the expectation of both sides of equation (3), we get:

[0070] E(Y|X1,X2,...,X k )=β0+β1X1+β2X2+...+β k X k (4)

[0071] Equation (4) is called the overall regression equation, E(Y|X1,X2,...,X). k ) represents the situation where, given the independent variable X i The conditional mean of the observed value Y under given conditions. Where β0, β1, β2, ..., β k The population parameters are often unknown, therefore, estimates are given based on sample observations. At this point, we obtain the sample regression equation (5).

[0072]

[0073] in, It is E(Y|X1,X2,...,X) k The point estimate of ) is then obtained. The parameters are then estimated using least squares estimation, with the following settings:

[0074]

[0075] Q respectively for Taking the partial derivative and setting it to 0, we get:

[0076]

[0077] Solving the system of equations will yield estimates of the parameters.

[0078] Cluster analysis is used to identify potential, hard-to-observe relationships in the index set data. Index set data with similar characteristics are classified and preprocessed, and a multivariate regression model is established based on the classified data to achieve a rapid conversion from pavement texture information extraction to pavement friction coefficient prediction.

[0079] To ensure the accuracy of the model, this embodiment can also perform rough set analysis, taking the correlation r with the friction coefficient as the target, as shown in equations (8) and (8). By adjusting the parameters of the cluster analysis algorithm, the goal is to minimize the sum of squared fitting errors.

[0080]

[0081] In the formula, n represents the number of samples, x i This represents the friction coefficient test value of the i-th sample. y represents the average value. i This represents the estimated friction coefficient for the i-th sample. This represents its average value.

[0082] Rough set theory is a mathematical tool for characterizing incompleteness and uncertainty. It can effectively analyze various incomplete information, such as imprecise and inconsistent data, and can also analyze and reason about data to discover implicit knowledge and reveal potential patterns. It can quantify the ability of different elements in a set of indicators to predict the friction coefficient. The main idea is to derive decision or classification rules for a problem by reducing knowledge while maintaining classification ability. For complex problems, rough set theory does not require much prior knowledge; analysis based on the data information in the decision table itself is sufficient. This method can identify the important factors characterizing the friction coefficient and eliminate irrelevant factors.

[0083] Rough set analysis primarily aims to minimize the error between predicted and actual values, adjusting clustering algorithm parameters to ultimately refine the regression analysis model. Model refinement, based on measured data, has a clear objective and can effectively improve the accuracy and performance of the prediction model.

[0084] To verify the mapping relationship between the index set obtained by the method of the present invention and the friction coefficient values ​​collected using a traditional pavement friction test vehicle, this embodiment also provides a verification scheme. This verification scheme uses the energy function of a Boltzmann machine as a deep learning model for implementation, specifically including:

[0085] The joint probability distribution is defined using an energy function:

[0086]

[0087] Where E(X) is the energy function of the Boltzmann machine, and Z is the function that ensures ∑ x The partition function of P(x) = 1

[0088] E(x) = -x T Ux-b T x

[0089] x is a three-dimensional binary random vector, U is the weight matrix of the three-dimensional model parameters, and b is the bias vector.

[0090] A three-dimensional binary random vector is established by taking a set of training samples from the friction coefficient values ​​collected by a traditional pavement friction test vehicle and the corresponding index set obtained by this invention. The joint probability distribution of the variables is obtained by learning the Boltzmann machine based on maximum likelihood. Finally, the similarity between the joint probability distribution and the mapping relationship function is compared. If the similarity between the two is within a specified threshold range, the mapping relationship can more realistically reflect the relationship between the actual friction coefficient value and the index set. The threshold can be determined by calculating multiple sets of data and comparing them with the actual test data. Therefore, the friction coefficient can be calculated using the index set and mapping relationship obtained by the method of this invention, and the calculated friction coefficient is close to the actual friction coefficient.

Claims

1. A non-contact pavement friction coefficient detection method, characterized by, Comprise: S1, three-dimensional texture scanning of the pavement to obtain a plurality of pavement point clouds; S2, three-dimensional texture reconstruction of the pavement point cloud to obtain a spatial surface; S3, extracting the index set of apparent texture features from the spatial surface, specifically including feature analysis of point cloud data in the spatial surface, extracting the index set of surface normal vector, MTD, constructed depth, profile unimodal average distance, and three-dimensional texture topography fractal dimension; S4, collecting friction coefficient data by pavement friction test vehicle, and establishing the mapping relationship between the index set and the friction coefficient.

2. The non-contact pavement friction coefficient detection method according to claim 1, characterized in that: Through laser constructed depth instrument scanning pavement to obtain a plurality of pavement point clouds with micro texture, and through visible light camera scanning pavement to obtain a plurality of pavement point clouds with macro texture.

3. The non-contact pavement friction coefficient detection method according to claim 2, characterized in that: Through splicing algorithm, the pavement point clouds are combined into complete pavement three-dimensional information.

4. The non-contact pavement friction coefficient detection method according to claim 3, characterized in that: The three-dimensional Delaunay The triangular division surface reconstruction algorithm reconstructs the three-dimensional information of the track surface to obtain a spatial surface.

5. The non-contact pavement friction coefficient detection method of claim 1, wherein: Cluster analysis and regression analysis of the index set and the friction coefficient data are performed to establish the mapping relationship between the index set and the friction coefficient.

6. The non-contact pavement friction coefficient detection method according to claim 1, characterized in that: It further comprises: through rough set analysis, the correlation with the friction coefficient is taken as the target, the fitting error sum of squares is pursued to be minimum by adjusting the clustering analysis algorithm parameters.

7. The non-contact pavement friction coefficient detection method according to claim 5 or 6, characterized in that: The joint probability distribution of the index set and the friction coefficient is learned through the energy function of the Boltzmann machine, and the accuracy of the mapping relationship is judged according to the joint probability distribution.

Images