A method and system for monitoring roadbed compaction based on airborne lidar

Abstract

The present invention discloses a roadbed compaction monitoring method and system based on airborne laser radar, belonging to the technical field of intelligent monitoring of roadbed compaction. The monitoring method includes using a drone to collect roadbed data, and constructing a historical database and a navigation database; according to the timestamps of the data in the historical database and the navigation database, all timestamps in the two databases are merged, duplicated, and constructed to obtain a complete time axis; according to the three-dimensional point cloud data corrected by the navigation deviation parameters, the roadbed compaction related parameters are calculated, and a related parameter database is constructed; the related parameter data in the related parameter database and the three-dimensional point cloud data in the historical database are fitted to obtain a roadbed compaction characterization parameter model; finally, combined with the two-dimensional image data in the historical database, the roadbed compaction change image under different time periods is generated. The present invention can effectively realize the intelligent monitoring of roadbed compaction.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring of roadbed compaction, and in particular to a roadbed compaction monitoring method and system based on airborne laser radar. Background Art

[0002] As the basic structure of the road, the roadbed is the foundation that supports the entire road structure and the traffic load on the road. For the construction of roadbed in road engineering, especially the construction of compaction, the whole process is relatively cumbersome. In the construction of roadbed in road engineering, the compaction process is relatively cumbersome. In order to ensure the quality of construction, each step must be standardized, combining theoretical knowledge with practical work and applying it to the construction to ensure the smooth progress of the roadbed project. Construction companies need to pay attention to the improvement of the construction effect of road engineering, adopt reasonable and effective roadbed compaction construction technology, lay a solid foundation for the overall project quality, and thus improve the driving quality of traffic pavement.

[0003] The rolling of the roadbed is an important part of the roadbed and pavement compaction technology, and the compaction degree is a key indicator to measure the density of the roadbed material. High compaction means that the roadbed structure is tighter, which can provide better bearing capacity, reduce the settlement and deformation of the roadbed under load, and thus ensure the long-term stability of the road. In the existing monitoring of the compaction degree of roadbed and pavement, random inspections are usually adopted, which is a point-to-surface method, and the overall compaction situation is not comprehensively judged; at the same time, since the roadbed filling is not in place at one time, it will involve multiple layers of repeated filling, which is difficult to manage. The manual recording method is inefficient, and the recording method is highly operational, which is prone to data falsification, resulting in irreversible engineering quality risks. Summary of the invention

[0004] One of the purposes of the present invention is to provide a roadbed compaction monitoring method based on airborne laser radar to solve the shortcoming of the prior art that there is a lack of intelligent monitoring of roadbed compaction.

[0005] The present invention is implemented by the following technical scheme. A method for monitoring the compaction of a roadbed based on an airborne laser radar comprises the following steps: S100, using a drone to collect data, and recording the roadbed data, navigation data and navigation data of the drone collected in each data collection, stamping the roadbed data, navigation data and navigation data of the drone obtained in the same data collection with the same timestamp, and constructing a historical database and a navigation database, the historical database is used to store the collected roadbed data, and the navigation database is used to store the navigation data and navigation data of the drone, the roadbed data is three-dimensional point cloud data and two-dimensional image data collected by an airborne laser radar and a camera installed under the drone, and the airborne laser radar and the camera are coupled to ensure that the laser radar and the camera collect data at the same time; S200, according to the timestamps of the data in the historical database and the navigation database, all timestamps in the two databases are merged, duplicated, and a complete timeline is constructed, based on the constructed timeline, for each timestamp, the closest roadbed data, navigation data and navigation data are found, the navigation data and navigation data of two adjacent timestamps on the timeline are compared, and the navigation deviation parameters are calculated to correct the roadbed data , use the ICP algorithm combined with the navigation deviation parameter to align the three-dimensional point cloud data of two adjacent time stamps on the time axis, ensure the accurate alignment of each layer of data, and ensure that the three-dimensional point cloud data of two adjacent time stamps are accurately compared in the same coordinate system, and finally obtain the three-dimensional point cloud data corrected by the navigation deviation parameter; S300, calculate the relevant parameters of the roadbed compaction degree according to the three-dimensional point cloud data corrected by the navigation deviation parameter in step S200, and build a relevant parameter database; S400, fit the relevant parameter data in the relevant parameter database and the three-dimensional point cloud data in the historical database to obtain the roadbed The compaction degree characterization parameter model takes the roadbed compaction degree characterization parameter model as a sample, which is input into a two-layer convolutional network layer to extract the implicit connection between the parameters in the roadbed compaction degree characterization parameter model on the time axis; S500, the feature information of the implicit connection between the parameters on the time axis is collected as feature data, which is input into the RBF neural network for classification, and then the hidden layer selection rule of the RBF neural network is used to select the change of the compaction degree of the roadbed in different time periods from the classified feature data, and combined with the two-dimensional image data in the historical database, to generate the change image of the roadbed compaction degree in different time periods.

[0006] Furthermore, the data collection includes initial scanning, scanning during construction and final scanning; the initial scanning includes using an airborne laser radar and a camera to scan the highway roadbed in its initial state to obtain initial three-dimensional point cloud data and initial image data, aligning the scanned initial three-dimensional point cloud data and initial image data to construct an initial three-dimensional model of the roadbed filling site.

[0007] Furthermore, the alignment operation includes calibrating the internal parameters of the camera, wherein the internal parameters of the camera include focal length (f x , f y ) and the optical center (c x , c y ); Determine the camera internal parameter matrix K through the camera internal parameters,

[0008] ,

[0009] Determine the camera external parameters, which are the relative position and posture between the camera and the laser radar, that is, the rotation matrix R and the translation vector T. The external parameter matrix is ​​[R|T]. The point cloud data is converted from the laser radar coordinate system to the camera coordinate system using the following formula:

[0010] ,

[0011] Where (x l , y l , z l ) is the point in the laser radar coordinate system, (x c , y c , z c ) is a point in the camera coordinate system; the camera's intrinsic parameter matrix K is used to project the converted three-dimensional point cloud data onto the camera's two-dimensional image plane using the following formula:

[0012] ,

[0013] Among them, (u, v) is the camera image coordinate. According to the camera image coordinate (u, v) obtained by projection, color information is extracted from the image and assigned to the corresponding point cloud point; the point cloud with color information is used to reconstruct the three-dimensional environment.

[0014] Furthermore, the parameters related to the compaction degree of the roadbed include the filling elevation change data of the roadbed after compaction, the loose laying thickness and thickness after rolling of the filler, and the compaction density of each layer.

[0015] Furthermore, the filling elevation change data of the compacted roadbed is obtained by comparing and calculating the three-dimensional point cloud data of two adjacent time stamps on the time axis, including the following formula:

[0016] Dh = h t2 - h t1 ,in, Dh is the height difference of the roadbed; h t2is the roadbed height at the latest timestamp on the timeline, h t1 It is the roadbed height of the timestamp before the latest timestamp on the timeline;

[0017] Furthermore, the loose laying thickness and the thickness after rolling of the filler are obtained by the following steps:

[0018] The three-dimensional point cloud data after the filler is laid on the roadbed is extracted according to the timestamp, and the three-dimensional point cloud data after the filler is compacted is compared between the two sets of data to calculate the loose laying thickness and the thickness after rolling of the filler.

[0019] Among them, the loose laying thickness is passed. , calculated, where T loose is the loose thickness of the filler, N is the total number of points in the point cloud data obtained by scanning, Dh i is the height difference of the i-th point in the point cloud data;

[0020] After rolling, the thickness is , calculated, where T compacted is the thickness after rolling; N compacted To satisfy Dh i A point below the threshold used to distinguish whether the paved fill is compacted.

[0021] Furthermore, the compaction density of each layer is obtained by comparing the 3D point cloud data of two adjacent time stamps and calculating the elevation difference of each point cloud point. Δz , Δz=z B2 -z B1 ,in, z B2 is the point cloud point at the next timestamp on the time axis, z B1 is the point cloud point of the previous timestamp on the time axis; the elevation difference of the calculated point cloud points is gridded with a unit grid of 1m×1m, and the average elevation difference Δz of each grid is calculated at the same time i , where i represents the i-th grid, so that the volume difference V in each grid i It is calculated by multiplying the elevation difference by the grid area, V i= Δz i 1m 2 , where V i is the volume difference of the i-th grid; Δz i is the average elevation difference of the ith grid, 1m 2is the grid area; the volume difference of all grids is summed up to get the volume difference of the entire filling layer, , where V 总 is the total volume difference of the entire filling layer, and n is the total number of grids; the filling mass is calculated according to the density of the filling material, and finally the compaction density of each layer is obtained according to the filling mass.

[0022] Further, in step S500, the changes in compaction degree of the roadbed in different time periods are selected from the classified feature data using the hidden layer selection rules of the RBF neural network, including the following sub-steps: S510, pre-processing the relevant parameter data in the relevant parameter database and the three-dimensional point cloud data in the historical database to ensure data quality and consistency; S520, sorting the roadbed compaction related parameters according to the time axis and collecting them into a feature data set of roadbed compaction data, which covers the changes in the roadbed compaction related parameters in different time periods; S530, determining the RBF neural network structure, setting the initial number of nodes to twice the number of features, and selecting the number of nodes with the best performance through experiments with different numbers of nodes, and using the K-means clustering algorithm to determine the number of center points, which will be used as hidden layer nodes; the selection rules of the hidden layer nodes include, according to the formula, To determine the initial number of hidden layer nodes of the RBF neural network, where α is the 3D point cloud data adjustment parameter; l is the number of RBF network input nodes; r is the number of hidden layer nodes; k is the number of RBF network output layer nodes; e is a constant coefficient with a value range of 10-20; S540, cluster analysis is performed on the training set data to determine the center position of the hidden layer nodes, calculate the radial basis function parameters of the RBF neural network, and construct the interpolation matrix of the RBF neural network, which is calculated based on the distance between the input data and the center position of the hidden layer; S550, use the least squares method to train the weights of the RBF neural network so that the network outputs a true compaction value.

[0023] On the other hand, the present invention also provides a roadbed compaction monitoring system based on an airborne laser radar, the system includes a processor; a memory storing a computer program, and when the computer program is executed by the processor, the roadbed compaction monitoring method based on the airborne laser radar as described above is implemented.

[0024] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0025] 1. The present invention uses drones for data collection, which improves the efficiency and coverage of data acquisition, especially in difficult-to-reach or dangerous areas. It also ensures the simultaneous acquisition of high-precision three-dimensional point cloud data and two-dimensional image data through the combination of airborne laser radar and camera, provides rich roadbed information, and stamps the roadbed data, navigation data and navigation data with the same timestamp to ensure data consistency and synchronization.

[0026] 2. The present invention corrects the roadbed data by calculating the navigation deviation parameters, thereby improving the accuracy and reliability of the data. The ICP (iterative closest point) algorithm is used to collect navigation data for registration of three-dimensional point cloud data, thereby ensuring accurate alignment and comparison of data layers at adjacent timestamps.

[0027] 3. The present invention calculates the relevant parameters of the roadbed compaction according to the corrected three-dimensional point cloud data, and constructs a relevant parameter database to provide basic data for further analysis of the roadbed compaction. By fitting the relevant parameter data and the three-dimensional point cloud data, a roadbed compaction characterization parameter model is established, and the convolutional network layer is used to extract the implicit relationship between the parameters in the model. At the same time, combined with the two-dimensional image data in the historical database, an image of the roadbed compaction change is generated, providing intuitive and easy-to-understand visualization results, which is convenient for engineering management and decision support. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0029] Figure 1 This is a flowchart provided for Example 1 of the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0031] Example 1

[0032] This embodiment discloses a roadbed compaction monitoring method based on airborne laser radar, and the method flow chart is as follows: Figure 1 As shown, the roadbed compaction monitoring method includes the following steps:

[0033] Step 1: Use drones to collect data, and record the roadbed data, navigation data and navigation data collected in each data collection, stamp the roadbed data, navigation data and navigation data of the drone with the same timestamp, and build a historical database and a navigation database. The historical database is used to store the collected roadbed data, and the navigation database is used to store the navigation data and navigation data of the drone. The roadbed data is obtained by collecting three-dimensional point cloud data and two-dimensional image data from the airborne laser radar and camera installed under the drone. The airborne laser radar and camera are coupled to ensure that the laser radar and camera collect data at the same time. Data collection can include using the airborne laser radar and camera to scan the highway roadbed in the initial state to obtain initial three-dimensional point cloud data and initial image data, aligning the scanned initial three-dimensional point cloud data and initial image data, constructing an initial three-dimensional model of the roadbed filling site, and updating the initial three-dimensional model using the three-dimensional point cloud data and image data with the latest timestamp according to the data timestamp to reflect the changes of the roadbed on the time axis.

[0034] Specifically, before the first scanning modeling, the debris, tree roots, etc. in the construction area to be scanned are first cleaned, and the surface is leveled. After the leveling is completed, the dry soil is properly sprinkled with water to moisten it to reach the optimal moisture content, and the highway subgrade in the most initial state is obtained. The drone is operated to perform an initial scan of the highway subgrade in the most initial state to obtain initial data. It should be noted that the drone height should be controlled to 10~15m during scanning. It is best to make the airborne laser radar and camera on the drone just cover the width of the subgrade that needs to be monitored. The drone is operated to navigate along the subgrade section that needs to be monitored, scan the subgrade and save the scanned data. An airborne laser radar and camera are installed at the bottom of the drone, where the airborne laser radar is used to scan the subgrade to obtain high-resolution road surface 3D mapping laser point cloud data; the camera is used to capture corresponding image data while the airborne radar is scanning. The airborne laser radar and camera are coupled in hardware. When the airborne laser radar is started, the camera will also be started. When the airborne laser radar is turned off, the two will also be turned off at the same time. The coupling of the two can achieve simultaneous start and stop, thereby ensuring that the laser point cloud data and image data obtained by scanning can be coupled and connected in time, which is convenient for subsequent data processing and comparison. At the same time, a timestamp is added to the data of each scan to facilitate subsequent data processing. When the drone is scanned for the first time, the scanning path of the drone is recorded, and the navigation data of the drone is recorded to form a drone navigation database. The scanning path is fixed, and in subsequent scans, the drone is operated to scan along the path of the first scan, so as to ensure the consistency of each scan data. At the same time, the deviation of each path from the initial path is calculated based on the data in the navigation database, and the deviation data is recorded in the drone navigation database, and each navigation database is also stamped with a corresponding timestamp.

[0035] Align the initial laser point cloud data and image data obtained by scanning. The specific alignment operation includes:

[0036] The coupling of the airborne laser radar and the camera ensures that the laser radar and the camera collect data at the same time. In other words, the data has been aligned on the time axis through hardware coupling. It only needs to be spatially aligned. Before spatial alignment, the internal parameters of the camera are calibrated first, including the focal length f. x , f y and optical center c x , c y , as well as radial distortion and tangential distortion; the camera internal parameter matrix K is determined by the internal parameters,

[0037]

[0038] Then determine the external parameters, which are the relative position and posture between the camera and the lidar, namely the rotation matrix R and the translation vector T. Get the external parameter matrix: [R|T].

[0039] Then, the point cloud data is converted from the lidar coordinate system to the camera coordinate system using the external parameter matrix, which includes the following formula:

[0040] ,

[0041] Where (x l , y l , z l ) is the point in the laser radar coordinate system, (x c , y c , z c ) is a point in the camera coordinate system.

[0042] Then use the camera's intrinsic parameter matrix K to project the converted point cloud data 3D points onto the camera's 2D image plane, which is converted using the following formula:

[0043] ,

[0044] Where (u, v) is the camera image coordinate.

[0045] Finally, according to the camera image coordinates (u, v) obtained by projection, the color information is extracted from the image and assigned to the corresponding point cloud points.

[0046] Then the point cloud with color information is used to reconstruct the 3D environment and obtain the initial 3D model of the roadbed filling site.

[0047] After each roadbed compaction operation, the drone is started to scan the roadbed, and then the above steps are repeated based on the latest scanning data to update the three-dimensional model of the roadbed filling site.

[0048] Each time after completing the 3D environment reconstruction, the collected initial laser point cloud data and image data will be stored in the historical database to facilitate subsequent data calls.

[0049] Step 2: According to the timestamps of the data in the historical database and the navigation database, all timestamps in the two databases are merged, duplicated, and a complete timeline is constructed. Based on the constructed timeline, for each timestamp, the closest roadbed data, navigation data, and navigation data are found.

[0050] Then, the navigation data and navigation data of two adjacent timestamps on the time axis are compared, and the navigation deviation parameters are calculated to correct the roadbed data. The ICP algorithm is combined with the navigation deviation parameters to align the three-dimensional point cloud data of two adjacent timestamps on the time axis to ensure the accurate alignment of each layer of data and the accurate comparison of the three-dimensional point cloud data of two adjacent timestamps in the same coordinate system, and finally the three-dimensional point cloud data corrected by the navigation deviation parameters are obtained.

[0051] Step 3: Calculate the relevant parameters of roadbed compaction based on the three-dimensional point cloud data corrected by the navigation deviation parameters in step 2, and build a relevant parameter database.

[0052] Specifically, the parameters related to the compaction degree of the roadbed calculated include the change data of the filling elevation of the roadbed after compaction, the loose laying thickness and thickness after rolling of the filler, and the compaction density of each layer.

[0053] Among them, the filling elevation change data of the compacted roadbed is obtained by comparing the three-dimensional point cloud data of two adjacent time stamps on the time axis, and is calculated using the following formula:

[0054] Dh = h t2 - h t1 ,in, Dh is the height difference of the roadbed; h t2 is the roadbed height at the latest timestamp on the timeline, h t1 It is the roadbed height of the timestamp before the latest timestamp on the timeline.

[0055] The loose thickness and thickness after rolling of the filler are obtained by the following steps:

[0056] The three-dimensional point cloud data after the filler is laid on the roadbed is extracted according to the timestamp, and the three-dimensional point cloud data after the filler is compacted is compared between the two sets of data to calculate the loose laying thickness and the thickness after rolling of the filler.

[0057] Among them, the loose laying thickness is through, , calculated.

[0058] In the above formula, T loose is the loose thickness of the filler, N is the total number of points in the scanned point cloud data, Dh i is the height difference of the i-th point in the point cloud data;

[0059] The threshold is used to distinguish whether the paved filler is compacted. The specific threshold setting can be determined according to the construction situation. , calculated.

[0060] In the above formula, T compacted is the thickness after rolling; N compacted To satisfy Dh i The point that is less than the set threshold value, that is, the point that is used to distinguish whether the paving filler is compacted.

[0061] The compaction density of each layer is obtained by the following steps:

[0062] Compare the 3D point cloud data of two adjacent timestamps and calculate the elevation difference of each point cloud point Δz.

[0063] Δz=z B2 -z B1 ,in, z B2 is the point cloud point at the next timestamp on the time axis, z B1 It is the point cloud point of the previous timestamp on the timeline.

[0064] Then the elevation differences of the calculated point cloud points are gridded with 1m×1m grids, and the average elevation difference Δz of each grid is calculated. i , where i represents the i-th grid, so that the volume difference V in each grid i Calculated by multiplying the elevation difference by the grid area.

[0065] V i= Δz i 1m 2 , where V iis the volume difference of the i-th grid; Δz i is the average elevation difference of the ith grid, 1m 2 is the grid area.

[0066] Finally, the volume difference of all grids is summed up to obtain the volume difference of the entire filling layer.

[0067] , where V 总 is the total volume difference of the entire filling layer, and n is the total number of grids.

[0068] The filling quality is calculated according to the density of the filling material, and finally the compaction density of each layer is obtained according to the filling quality. Specifically, the density of the filling material is designed and calculated according to the specific construction materials.

[0069] Step 4: Fit the relevant parameter data in the relevant parameter database and the three-dimensional point cloud data in the historical database to obtain a roadbed compaction characterization parameter model, and use the roadbed compaction characterization parameter model as a sample, input it into the two-layer convolutional network layer, and use it to extract the implicit connection between the parameters in the roadbed compaction characterization parameter model on the time axis. After obtaining the implicit connection, the feature information of the implicit connection between the parameters on the time axis is collected as feature data, input into the RBF neural network for classification, and use the hidden layer selection rules of the RBF neural network to select the changes in the compaction degree of the roadbed in different time periods from the classified feature data.

[0070] Specifically, using RBF neural network to extract and analyze the changes in roadbed compaction degree in different time periods from the classified feature data includes the following sub-steps:

[0071] Pre-process the relevant parameter data in the relevant parameter database and the 3D point cloud data in the historical database to ensure data quality and consistency.

[0072] The parameters related to roadbed compaction are sorted according to the time axis and collected into a characteristic data set of roadbed compaction data, which covers the changes of parameters related to roadbed compaction in different time periods.

[0073] Then determine the RBF neural network structure and set the initial number of nodes to twice the number of features. And through experiments with different numbers of nodes, select the number of nodes with the best performance. Finally, use the K-means clustering algorithm to determine the number of center points, which will be used as hidden layer nodes.

[0074] Specifically, the hidden layer selection rules include, according to the formula, To determine the initial number of hidden layer nodes of the RBF neural network, where α is the 3D point cloud data adjustment parameter; l is the number of RBF network input nodes; r is the number of hidden layer nodes; k is the number of RBF network output layer nodes; e is a constant coefficient, ranging from 10 to 20.

[0075] Perform cluster analysis on the training set data to determine the center position of the hidden layer nodes.

[0076] Calculate the radial basis function parameters of the RBF neural network and finally construct the interpolation matrix of the RBF neural network, which is calculated based on the distance between the input data and the center position of the hidden layer.

[0077] Use the least squares method to train the weights of the RBF neural network so that the network outputs the true compaction value.

[0078] Finally, combined with the two-dimensional image data in the historical database, the change images of roadbed compaction degree in different time periods are generated.

[0079] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A roadbed compaction monitoring method based on airborne laser radar, characterized in that: The solidity monitoring method comprises: S100, using a drone to collect data, and recording the roadbed data collected in each data collection, the navigation data and the navigation data of the drone, and stamping the roadbed data, the navigation data and the navigation data obtained in the same data collection with the same timestamp, and building a historical database and a navigation database, the historical database is used to store the collected roadbed data, and the navigation database is used to store the navigation data and the navigation data of the drone, The roadbed data is three-dimensional point cloud data and two-dimensional image data obtained by collecting the airborne laser radar and camera installed under the drone. The airborne laser radar and camera are coupled to ensure that the laser radar and the camera collect data at the same time; S200, according to the timestamps of the data in the history database and the navigation database, all the timestamps in the two databases are merged, duplicates are removed, and a complete timeline is constructed. Based on the constructed timeline, for each timestamp, the closest roadbed data, navigation data, and navigation data are found. Compare the navigation data and the navigation data of two adjacent time stamps on the time axis, calculate the navigation deviation parameter to correct the roadbed data, use the ICP algorithm combined with the navigation deviation parameter to align the three-dimensional point cloud data of two adjacent time stamps on the time axis, ensure the accurate alignment of each layer of data, ensure that the three-dimensional point cloud data of two adjacent time stamps are accurately compared in the same coordinate system, and finally obtain the three-dimensional point cloud data corrected by the navigation deviation parameter; S300, calculating the roadbed compaction related parameters according to the three-dimensional point cloud data corrected by the navigation deviation parameters in step S200, and constructing a related parameter database, The parameters related to the compaction degree of the roadbed include the filling elevation change data of the roadbed after compaction, the loose laying thickness and thickness after rolling of the filler, and the compaction density of each layer; S400, fitting the relevant parameter data in the relevant parameter database and the three-dimensional point cloud data in the historical database to obtain a roadbed compaction degree characterization parameter model, taking the roadbed compaction degree characterization parameter model as a sample, inputting it into a two-layer convolutional network layer to extract the implicit connection between the parameters in the roadbed compaction degree characterization parameter model on the time axis; S500, the feature information of the implicit connection between the parameters on the time axis is collected as feature data, which is input into the RBF neural network for classification, and then the hidden layer selection rule of the RBF neural network is used to select the change of the compaction degree of the roadbed in different time periods from the classified feature data, and combined with the two-dimensional image data in the historical database, an image of the change of the compaction degree of the roadbed in different time periods is generated.

2. The method for monitoring roadbed compaction based on airborne laser radar according to claim 1, characterized in that: The data acquisition includes scanning the highway roadbed in the initial state using an airborne laser radar and a camera to obtain initial three-dimensional point cloud data and initial image data, aligning the scanned initial three-dimensional point cloud data and initial image data to construct an initial three-dimensional model of the roadbed filling site, The initial three-dimensional model is updated using the three-dimensional point cloud data and image data with the latest timestamp according to the data timestamp to reflect the changes of the roadbed on the time axis.

3. The method for monitoring roadbed compaction based on airborne laser radar according to claim 2 is characterized in that: The alignment operation includes: Calibrate the camera internal parameters, including focal length (f x , f y ) and the optical center (c x , c y ); The camera internal parameter matrix K is determined by the camera internal parameters. ， Determine the camera external parameters, which are the relative position and posture between the camera and the laser radar, that is, the rotation matrix R and the translation vector T. The external parameter matrix is , Use the external parameter matrix to transform the point cloud data from the lidar coordinate system to the camera coordinate system using the following formula: ， Where (x l , y l , z l ) is the point in the laser radar coordinate system, (x c , y c , z c ) is a point in the camera coordinate system; Use the camera's intrinsic parameter matrix K to project the converted three-dimensional points of the point cloud data onto the camera's two-dimensional image plane using the following formula: ， Where (u, v) is the camera image coordinate, According to the camera image coordinates (u, v) obtained by projection, color information is extracted from the image and assigned to the corresponding point cloud points; 3D environment reconstruction using point clouds with color information.

4. The method for monitoring roadbed compaction based on airborne laser radar according to claim 1, characterized in that: The filling elevation change data of the compacted roadbed is obtained by comparing and calculating the three-dimensional point cloud data of two adjacent time stamps on the time axis, including the following formula: ,in, is the height difference of the roadbed; is the roadbed height at the latest timestamp on the timeline, It is the roadbed height of the timestamp before the latest timestamp on the timeline; The loose thickness and thickness after rolling of the filler are obtained by the following steps: The three-dimensional point cloud data after the filler is laid on the roadbed is extracted according to the timestamp, and the three-dimensional point cloud data after the filler is compacted is compared between the two sets of data to calculate the loose laying thickness and the thickness after rolling of the filler. Among them, the loose laying thickness is , calculated, where is the loose thickness of the filler, N is the total number of points in the scanned point cloud data, is the height difference of the i-th point in the point cloud data; After rolling, the thickness Calculated, among which, is the thickness after rolling; To satisfy a point that is less than a threshold value used to distinguish whether the paved fill is compacted; The compaction density of each layer is obtained by the following steps: Compare the 3D point cloud data of two adjacent timestamps and calculate the elevation difference of each point cloud point , ,in is the point cloud point at the next timestamp on the time axis, is the point cloud point of the previous timestamp on the time axis; The elevation differences of the calculated point cloud points are gridded in 1m×1m grids, and the average elevation difference Δz of each grid is calculated. i , where i represents the i-th grid, so that the volume difference V in each grid i It is calculated by multiplying the elevation difference by the grid area. , where V i is the volume difference of the i-th grid; Δz i is the average elevation difference of the i-th grid; Sum the volume differences of all grids to get the volume difference of the entire filling layer. , where V 总 is the total volume difference of the entire filling layer, and n is the total number of grids; The filling mass is calculated according to the density of the filling material, and finally the compaction density of each layer is obtained according to the filling mass.

5. The method for monitoring roadbed compaction based on airborne laser radar according to claim 1, characterized in that: The step S500 of extracting and analyzing the changes in roadbed compaction degree in different time periods from the classified feature data using the RBF neural network includes the following sub-steps: S510, preprocessing the relevant parameter data in the relevant parameter database and the three-dimensional point cloud data in the historical database to ensure data quality and consistency; S520, sorting the parameters related to the roadbed compaction degree according to the time axis and collecting them into a characteristic data set of roadbed compaction degree data, where the data set covers the changes of the parameters related to the roadbed compaction degree in different time periods; S530, determining the RBF neural network structure, setting the initial number of nodes to twice the number of features, and selecting the number of nodes with the best performance through experiments with different numbers of nodes, and using the K-means clustering algorithm to determine the number of center points, which will serve as hidden layer nodes; The selection rules of the hidden layer nodes include: According to the formula, To determine the initial number of hidden layer nodes of the RBF neural network, where α is the 3D point cloud data adjustment parameter; l is the number of RBF network input nodes; r is the number of hidden layer nodes; k is the number of RBF network output layer nodes; is a constant coefficient, ranging from 10 to 20; S540, performing cluster analysis on the training set data, determining the center position of the hidden layer node, calculating the radial basis function parameters of the RBF neural network, and constructing an interpolation matrix of the RBF neural network, which is calculated based on the distance between the input data and the center position of the hidden layer; S550, using the least squares method, training the weights of the RBF neural network so that the network outputs a true compaction value.

6. A computer system, characterized in that: The computer system comprises: processor; A memory storing a computer program, which, when executed by a processor, implements the roadbed compaction monitoring method based on airborne laser radar as described in any one of claims 1 to 5.

Images