Roadbed compaction laser radar point cloud evaluation system and method

Abstract

This invention discloses a system and method for assessing roadbed compaction using a LiDAR point cloud. These methods belong to the field of intelligent road construction technology. The system includes a LiDAR point cloud acquisition module, an adaptive multi-scale octree spatial segmentation module, a differential geometry feature extraction module, a compaction field construction module, a compaction visualization module, and a vibration control module. The system first collects point cloud data from the roadbed surface, then applies adaptive multi-scale octree spatial segmentation to extract differential geometry features. These features are then integrated to construct a compaction field, generating a color-coded contour image. Finally, the system plans a vibration path and controls roller vibration parameters. Through LiDAR point cloud analysis, the system achieves accurate assessment and intelligent control of roadbed compaction, overcoming the shortcomings of traditional compaction methods, such as reliance on manual experience, inability to perform real-time assessments, and uneven compaction. This significantly improves roadbed compaction quality and construction efficiency, possessing significant engineering application value.

Description

Technical Field

[0001] The present invention relates to the field of road engineering construction, and in particular to a system and method for accurately evaluating and intelligently controlling roadbed compaction using laser radar point cloud technology. Background Art

[0002] Roadbed compaction is a critical step in road construction, and its quality directly impacts the lifespan and safety of roads. Traditional methods for testing roadbed compaction include sand filling, knife ring, and nuclear density meters. These methods suffer from limitations such as limited sampling points, insufficient representativeness, and low efficiency, making them inadequate for the precise control of compaction quality required in modern road construction.

[0003] In recent years, with the development of LiDAR technology, three-dimensional spatial analysis based on point cloud data has provided a new technical means for assessing roadbed compaction. However, existing LiDAR point cloud-based compaction assessment systems still have many shortcomings: first, spatial segmentation uses fixed thresholds, which cannot adapt to complex roadbed environments; second, feature extraction is mostly based on simple geometric parameters, lacking comprehensive multi-dimensional and multi-scale analysis; and third, there is a lack of effective connection between compaction assessment results and compaction control, making it difficult to form a closed-loop intelligent control system.

[0004] Therefore, there is an urgent need for a system and method that can accurately evaluate the roadbed compaction degree and realize intelligent vibration control to improve the roadbed compaction quality and construction efficiency. Summary of the Invention

[0005] The purpose of the present invention is to provide a roadbed compaction lidar point cloud evaluation system and method thereof, which realizes accurate evaluation and intelligent control of roadbed compaction by organically combining lidar point cloud technology, adaptive space segmentation, differential geometry feature extraction and multi-scale feature fusion.

[0006] The present invention proposes a roadbed compaction laser radar point cloud evaluation system, comprising:

[0007] The LiDAR point cloud acquisition module is installed in the roller of the road roller to collect point cloud data of the roadbed surface;

[0008] An adaptive multi-scale octree space segmentation module is connected to the lidar point cloud acquisition module, and is used to receive the point cloud data, dynamically adjust the segmentation threshold based on the local point cloud density and geometric structure characteristics, maintain the topological characteristics of the roadbed structure, and construct a multi-level spatial representation;

[0009] A differential geometry feature extraction module, connected to the adaptive multi-scale octree space segmentation module, is used to calculate the curvature field tensor of the roadbed surface, analyze the particle distribution characteristics, and extract surface morphological features;

[0010] a compaction field construction module, connected to the differential geometry feature extraction module, for receiving the curvature field tensor, particle distribution features, and surface morphology features, fusing multi-scale features, and establishing a mapping from discrete point clouds to continuous compaction fields;

[0011] a compaction visualization module, connected to the compaction field construction module, for converting the compaction field into a color-coded contour image;

[0012] A vibration control module is connected to the compaction visualization module and is used to plan a vibration path based on the color-coded contour image and control vibration parameters of the roller.

[0013] Preferably, the adaptive multi-scale octree space segmentation module includes:

[0014] Local density calculation unit, used to calculate the point cloud density of any area in the point cloud space;

[0015] Structured point cloud clustering unit, used for preliminary region segmentation based on point cloud geometric distribution and spatial continuity;

[0016] Material property reference unit, used to assist in identifying regional material types based on point cloud reflection intensity and spatial distribution characteristics, and provide reference coefficients;

[0017] an adaptive threshold determination unit, configured to dynamically determine an optimal segmentation threshold based on the region density, geometric characteristics, and material reference coefficients;

[0018] Topological characteristic analysis unit, used to identify the connected area set in the point cloud and calculate the topological feature vector of the connected area to ensure that the topological characteristics are not destroyed before and after segmentation;

[0019] Multi-scale link construction unit, used to establish link relationships between octree nodes of different scales to achieve smooth transition between different resolution levels;

[0020] A multi-source data cross-validation unit is used to integrate roller vibration feedback data and GPS location information to improve the reliability of regional division.

[0021] Preferably, the differential geometry feature extraction module includes:

[0022] Normal vector estimation unit, used to estimate the normal vector of each point based on its K nearest neighbor point set;

[0023] Curvature tensor calculation unit, used to construct the curvature tensor of each point, including the principal curvature and principal direction;

[0024] The curvature field construction unit is used to construct a continuous curvature field function based on the curvature tensor of discrete points;

[0025] Particle detection and segmentation unit, used to identify and segment independent particle units from point clouds;

[0026] A particle morphology analysis unit, used to calculate the morphological feature vector of each particle;

[0027] Particle population statistics unit, used to perform regional statistical analysis on particles smaller than the identification limit;

[0028] Surface morphology feature extraction unit, used to analyze surface roughness characteristics at multiple scales and construct roughness distribution function;

[0029] The multi-temporal data fusion unit is used to fuse point cloud data collected by multiple rolling operations to increase the effective point cloud density.

[0030] Preferably, the compaction field construction module includes:

[0031] Feature normalization unit, used to normalize different features and eliminate dimensional differences;

[0032] Feature weighting unit, used to assign dynamic weights to different features based on roadbed type and engineering requirements;

[0033] Feature dimension reduction and selection unit, used to achieve feature dimension reduction and optimal selection through wavelet transform and principal component analysis;

[0034] A nonlinear mapping unit, used to establish a nonlinear mapping function from feature space to compaction value;

[0035] A local adaptive adjustment unit, used to dynamically adjust mapping parameters according to local characteristics;

[0036] The compaction field evolution prediction unit is used to predict the changing trend of roadbed compaction under given vibration parameters.

[0037] Preferably, the compaction visualization module includes:

[0038] A spatial discretization unit is used to discretize the continuous compaction field into a grid structure;

[0039] A color mapping unit, used to establish a mapping function from compaction value to RGB color space;

[0040] A contour extraction unit, used to extract a set of contour lines based on a specific compaction threshold;

[0041] A 3D rendering unit for displaying color coding and contour lines in a 3D view;

[0042] Real-time update unit, used to dynamically update the visualization results to reflect the compaction progress.

[0043] Preferably, the vibration control module includes:

[0044] The compaction grade judgment unit is used to classify the roadbed area into three categories: uncompacted, preliminarily compacted, and compacted;

[0045] A compaction requirement graph generating unit, configured to generate a weight graph representing compaction requirements based on the compaction degree field;

[0046] Path optimization unit, which plans the optimal path by comprehensively considering compaction requirements, energy consumption, and time;

[0047] A vibration parameter mapping unit, used to establish a mapping relationship between compaction requirements and vibration parameters;

[0048] State feedback control unit, used to dynamically adjust vibration parameters based on real-time compaction results;

[0049] The automatic execution unit is used to control the vibrator to perform compaction operations along the planned path and automatically stop vibration when the target compaction degree is reached.

[0050] Preferably, in the topological characteristic analysis unit, the connectivity determination distance is 10 cm, the topological characteristic threshold is that the connectivity change in each dimension does not exceed 20%, the structural complexity change tolerance is ±15%, and the minimum connected area size is 0.3 m³.

[0051] Preferably, in the normal vector estimation unit, the number of neighboring points K is 30 points, and the normal vector smoothing radius is 15 cm; in the curvature tensor calculation unit, the principal curvature threshold is ±0.5 / m; in the particle detection and segmentation unit, the particle recognition threshold is that the point spacing change rate is greater than 200%, which is considered a particle boundary, and the minimum particle size is 5 cm³; in the surface morphology feature extraction unit, the roughness calculation window is 25 cm×25 cm, and the roughness grading standard is 5 levels.

[0052] Preferably, in the color mapping unit, the color mapping scheme uses blue to represent low compaction areas, green to represent medium compaction areas, and red to represent high compaction areas; in the contour line extraction unit, the contour line interval is 10% compaction.

[0053] The roadbed compaction degree lidar point cloud evaluation method includes the following steps:

[0054] Collect lidar point cloud data of the roadbed surface;

[0055] Applying adaptive multi-scale octree spatial segmentation to the point cloud data, dynamically adjusting the segmentation threshold based on local point cloud density and geometric structure characteristics, supplemented by material property reference, maintaining the topological characteristics of the roadbed structure, and constructing a multi-level spatial representation;

[0056] Extract differential geometric features of roadbed point clouds, calculate the curvature field tensor of the roadbed surface, analyze particle distribution characteristics and their population statistics, extract surface morphological features, and improve analysis accuracy through multi-temporal data fusion;

[0057] Construct a compaction field, normalize and weight the features, optimize the feature space through feature dimensionality reduction and selection, and establish a mapping from discrete point clouds to a continuous compaction field;

[0058] generating a compaction visualization result, converting the compaction field into a color-coded contour image;

[0059] According to the compaction visualization result, a vibration path is planned, vibration parameters of the roller are controlled, and the roadbed compaction operation is performed.

[0060] The beneficial effects of the present invention include:

[0061] 1. Improved the accuracy and comprehensiveness of compaction assessment. Using adaptive multi-scale octree spatial segmentation technology, the system dynamically adjusts segmentation parameters based on subgrade material properties and point cloud density, preserving the topological characteristics of the subgrade structure and improving the accuracy of spatial representation. Furthermore, through differential geometry feature extraction and multi-scale feature fusion, the system comprehensively analyzes subgrade compaction from multiple dimensions, including macrostructure, mesoscopic particles, and microscopic surfaces.

[0062] 2. Intelligent control of the compaction process is achieved. Based on compaction field construction and visualization technology, the system can generate intuitive compaction cloud maps, providing a precise basis for vibration path planning and parameter control. Through adaptive vibration control technology, the system can automatically adjust vibration parameters according to the compaction status of different roadbed areas, achieving precise compaction.

[0063] 3. Significantly improved construction efficiency and quality. Practical applications have shown that the system and method of the present invention have improved roadbed compaction uniformity by over 30%, reduced substandard compaction areas by over 85%, shortened construction time by over 35%, reduced fuel consumption by over 20%, reduced roadbed settlement by over 40%, and extended service life by 25% to 30%. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 Schematic diagram of the overall architecture of the roadbed compaction laser radar point cloud evaluation system of the present invention;

[0065] Figure 2 Schematic diagram of the structure of the adaptive multi-scale octree space segmentation module of the present invention;

[0066] Figure 3 Schematic diagram of the structure of the differential geometry feature extraction module of the present invention;

[0067] Figure 4 This is a schematic diagram of the structure of the compaction field building module of the present invention;

[0068] Figure 5 This is a schematic structural diagram of the compaction visualization module of the present invention;

[0069] Figure 6 Schematic diagram of the structure of the vibration control module of the present invention;

[0070] Figure 7 This is a flow chart of the roadbed compaction laser radar point cloud evaluation method of the present invention. DETAILED DESCRIPTION

[0071] Please refer to Figure 1 - Figure 7 The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0072] like Figure 1 As shown, the roadbed compaction laser radar point cloud evaluation system of the present invention includes a laser radar point cloud acquisition module 1, an adaptive multi-scale octree space segmentation module 2, a differential geometry feature extraction module 3, a compaction field construction module 4, a compaction visualization module 5 and a vibration control module 6.

[0073] The LiDAR point cloud acquisition module 1 is installed inside the roller drum and is used to collect point cloud data from the roadbed surface. In a preferred embodiment of the present invention, the LiDAR point cloud acquisition module 1 uses a 128-line vehicle-mounted LiDAR with a scanning frequency of 40Hz, achieving a point cloud density of up to 400 points / m². This module, combined with a high-precision inertial navigation system, ensures precise positioning and registration of the point cloud data.

[0074] The adaptive multi-scale octree spatial segmentation module 2 is connected to the lidar point cloud acquisition module 1. It receives point cloud data and dynamically adjusts the segmentation threshold based on the local point cloud density and material properties, maintaining the topological characteristics of the roadbed structure and constructing a multi-level spatial representation. This module's innovation lies in its ability to adaptively adjust segmentation parameters based on the different properties of the roadbed material and the local density of the point cloud, ensuring the rationality and effectiveness of spatial segmentation.

[0075] The differential geometry feature extraction module 3 is connected to the adaptive multi-scale octree spatial segmentation module 2 to calculate the curvature field tensor of the roadbed surface, analyze particle distribution characteristics, and extract surface morphological features. This module uses differential geometry theory and point cloud analysis technology to deeply analyze the geometric characteristics of the roadbed surface from a microscopic perspective, providing an important basis for compaction assessment.

[0076] The compaction field construction module 4 is connected to the differential geometry feature extraction module 3. It receives the curvature field tensor, particle distribution characteristics, and surface morphology features, fuses multi-scale features, and establishes a mapping from the discrete point cloud to the continuous compaction field. This module is the core of the system. Through feature fusion and nonlinear mapping, it achieves accurate compaction assessment from point cloud data.

[0077] The compaction visualization module 5 is connected to the compaction field construction module 4 and is used to convert the compaction field into a color-coded contour image. This module makes the compaction assessment results easier to understand and apply through intuitive visualization.

[0078] Vibration Control Module 6, connected to Compactness Visualization Module 5, plans the vibration path based on the color-coded contour image and controls the roller's vibration parameters. This module organically combines compaction assessment results with vibration control, creating a closed-loop intelligent system from assessment to control.

[0079] like Figure 2 As shown, the adaptive multi-scale octree space segmentation module 2 includes a local density calculation unit 21, a structured point cloud clustering unit 22, a material property reference unit 23, an adaptive threshold determination unit 24, a topological property analysis unit 25, a multi-scale link construction unit 26 and a multi-source data cross-validation unit 27.

[0080] The local density calculation unit 21 is used to calculate the point cloud density of any region in the point cloud space. In a preferred embodiment of the present invention, for any region R in the point cloud space, its point cloud density is The calculation formula is:

[0081] .

[0082] in: is the point cloud density of region R, in points / m³; is the number of point clouds in region R, in points; is the volume of region R, in m³. Point cloud density is an important reference for subsequent spatial segmentation. The higher the regional density, the higher the segmentation threshold should be to ensure reasonable segmentation. Typically, for roadbed projects, point cloud density ranges from 100 to 1000 points / m³. Areas with a density below 50 points / m³ are generally considered noisy areas.

[0083] The structured point cloud clustering unit 22 is used to perform preliminary region division based on the geometric distribution and spatial continuity of the point cloud. This unit uses the DBSCAN density clustering algorithm to achieve region division, which mainly relies on the spatial position relationship of the points rather than the reflection intensity. The clustering formula is:

[0084] .

[0085] in: Indicates a point and The connection relationship, is the Euclidean distance between two points, is the distance threshold, in meters. After the initial clustering, the system optimizes the region division through merging and splitting operations to ensure the geometric consistency within the region.

[0086] The material property reference unit 23 is used to assist in identifying the regional material type based on the point cloud reflection intensity and spatial distribution characteristics, and provides a reference coefficient. Considering that the reflection intensity is easily affected by environmental factors, this unit adopts a comprehensive strategy of multi-feature fusion and environmental factor correction:

[0087] .

[0088] in: is the corrected reflection intensity, is the original reflection intensity, The angle between the laser beam and the surface normal (the angle of incidence) is θ. The system also combines point cloud geometric distribution characteristics, spatial frequency characteristics, and texture characteristics to improve the reliability of material property identification. However, throughout the segmentation process, material properties are only used as auxiliary reference factors to reduce the impact of environmental interference.

[0089] The adaptive threshold determination unit 24 is used to dynamically determine the optimal segmentation threshold based on the regional density, geometric characteristics and material reference coefficient. Different from the traditional fixed threshold segmentation, the present invention adopts an improved adaptive threshold function , and its calculation formula is:

[0090] .

[0091] in: is the adaptive segmentation threshold, which indicates the minimum number of points in the octree node, in points; It is the basic threshold coefficient, dimensionless, and usually ranges from 0.05 to 0.2; is the regional point cloud density, in points / m³; is the density influence factor, dimensionless, usually ranging from 0.5 to 0.7; is the material type reference coefficient, dimensionless; It is an engineering adjustment factor, dimensionless, and is fine-tuned according to engineering requirements, with a value range of 0.8-1.2. Note that in this formula, the material type factor only affects 20% of the weight. Even if there is an error in material identification, the impact on the final segmentation result is limited. This adaptive threshold mechanism enables the system to flexibly adjust the segmentation parameters according to different roadbed characteristics and point cloud characteristics, thereby improving the adaptability and accuracy of spatial segmentation. The topological characteristic analysis unit 25 is used to identify the connected area set in the point cloud, calculate the topological feature vector of the connected area, and ensure that the topological characteristics are not destroyed before and after segmentation. In one embodiment of the present invention, the topological feature vector The calculation of is based on homology group theory and is expressed as:

[0092] .

[0093] in: is the topological eigenvector, dimensionless; is the zero ViBetti number, which represents the number of connected components and is dimensionless; is the one-dimensional Betti number, representing the number of loops, dimensionless; is the two-dimensional Betti number, which represents the number of cavities and is dimensionless; is the structural complexity, dimensionless, defined as:

[0094] .

[0095] in: is the structural complexity, dimensionless; 、 and The meaning is the same as above. To ensure the preservation of topological characteristics, the system will compare the topological feature vectors before and after segmentation. When the topological feature changes exceed the preset threshold, the topology repair algorithm is triggered. Specifically, the connectivity judgment distance is 10cm, the topological feature threshold is that the connectivity change of each dimension does not exceed 20%, and the structural complexity change tolerance is , the minimum connected area size is 0.3m 3 .

[0096] The multi-scale link construction unit 26 is used to establish the link relationship between the octree nodes of different scales to achieve a smooth transition between different resolution levels. In the present invention, an L-layer scale space is constructed. , each layer represents the spatial representation at different resolutions. Adjacent scale layers are mapped up and down through the mapping function and To achieve feature information transfer. In practical applications, a five-layer scale space is usually set, with a resolution ratio of 2:1 between adjacent layers. Feature transfer uses an inverse distance weighting method with a minimum scale resolution of 5 cm to ensure that the particle size of interest can be captured.

[0097] The multi-source data cross-validation unit 27 integrates roller vibration feedback data and GPS location information to improve the reliability of segmentation. This unit correlates point cloud data with vibration feedback and location information to achieve multi-source data cross-validation. When abnormal fluctuations in reflection intensity data are detected, the system automatically reduces their weight, increasing the reference value of geometric features and ensuring the stability and reliability of segmentation results under various environmental conditions.

[0098] like Figure 3 As shown, the differential geometry feature extraction module 3 includes a normal vector estimation unit 31, a curvature tensor calculation unit 32, a curvature field construction unit 33, a particle detection and segmentation unit 34, a particle morphology analysis unit 35, a particle population statistics unit 36, a surface morphology feature extraction unit 37 and a multi-phase data fusion unit 38.

[0099] The normal vector estimation unit 31 is used to estimate the normal vector of each point based on its K nearest neighbor point set. , whose normal vector The estimation of is based on the principal component analysis (PCA) method, and the specific steps are as follows:

[0100] First, calculate the point The K nearest neighbor point set The covariance matrix of :

[0101] .

[0102] in: is the covariance matrix with a dimension of 3×3; is the number of neighbor points, dimensionless; For the The coordinates of the nearest neighbor points, the dimension is 3×1, the unit is m; is the centroid coordinate of the K nearest neighbor point set, with a dimension of 3×1 and a unit of m; Represents matrix transpose. Indicates the The displacement vectors of the neighboring points relative to the center of mass, Represents the outer product of the displacement vector, resulting in a 3×3 matrix. Sum operation Indicates the accumulation of calculation results of all K neighboring points.

[0103] Then, the covariance matrix Perform eigenvalue decomposition to obtain eigenvalues and the corresponding eigenvector . Minimum eigenvalue The corresponding eigenvector That is the point Normal vector .

[0104] In practical applications, the number of neighboring points K is usually set to 30 points and the normal vector smoothing radius is 15 cm to ensure the stability and accuracy of the normal vector estimation.

[0105] The curvature tensor calculation unit 32 is used to construct the curvature tensor of each point, including the principal curvature and principal direction. , whose curvature tensor It can be expressed as:

[0106] .

[0107] in: is the curvature tensor with dimension 2×2; and is the principal curvature, in units of 1 / m; and is the main direction, a unit vector, dimensionless. The calculation of the principal curvature is based on the rate of change of the normal vector field, and the specific formula is:

[0108] .

[0109] in: is the jth principal curvature, in 1 / m; is the covariance matrix The j-th eigenvalue of , in m²; is the sum of the characteristic values, in m². In the present invention, the principal curvature threshold is set to ±0.5 / m to distinguish between flat areas and undulating areas. Generally, a well-compacted roadbed surface shows a small principal curvature change ( ), while the undercompacted areas show a larger change in principal curvature ( ).

[0110] The curvature field construction unit 33 is used to construct a continuous curvature field function based on the curvature tensor of the discrete points. In the present invention, the radial basis function (RBF) interpolation method is used to construct the continuous curvature field. :

[0111] .

[0112] in: For spatial points The curvature field value at , in units of 1 / m; is the number of points involved in the interpolation, dimensionless; is the weight coefficient of the i-th point, dimensionless; is the radial basis function, usually a Gaussian function ,in is the distance in m, is the shape parameter, the unit is 1 / m, which controls the smoothness of the function; Represents a spatial point To the i-th point The Euclidean distance is in meters. In practical applications, the curvature field resolution is usually set to 10 cm to balance the computational efficiency and accuracy requirements.

[0113] The particle detection and segmentation unit 34 is used to identify and segment independent particle units from the point cloud. During the roadbed compaction process, the particle arrangement directly reflects the degree of compaction. The present invention uses a region growing algorithm for particle segmentation. The specific steps are as follows:

[0114] First, calculate the rate of change of the distance between each point in the point cloud and its neighboring points :

[0115] .

[0116] in: for point and The rate of change of the distance between them is dimensionless; for point and The Euclidean distance between them is in meters; is the local average point spacing, in meters.

[0117] When the distance change rate When the difference is greater than the particle identification threshold (set to 200%), the two points are considered to belong to different particles. Based on this criterion, the region growing algorithm is used to segment independent particle units. ,in In practice, the minimum particle size is set to 5 cm³. Clusters smaller than this value are considered unreliable segmentation results and are processed by the particle population statistics unit.

[0118] The particle morphology analysis unit 35 is used to calculate the morphological feature vector of each particle. , its morphological feature vector Defined as:

[0119] .

[0120] in: is the morphological characteristic vector, dimensionless; Indicates particle size (volume), unit is cm3; Represents shape characteristics (such as roundness, major axis ratio, etc.), dimensionless; Represents directional characteristics, expressed as a unit vector, dimensionless; represents the distribution density, dimensionless, and is defined as the number of particles per unit volume. These characteristic parameters directly reflect the compaction state. For example, particles in well-compacted areas usually have a higher distribution density ( and more consistent directionality ( The standard deviation of the two groups is <0.3).

[0121] The particle population statistics unit 36 ​​is used to perform regional statistical analysis on particles smaller than the identification limit. For particles smaller than 5 cm³, individual identification is no longer performed, but regional statistical analysis is used to evaluate the particle population distribution characteristics:

[0122] .

[0123] in: is the aggregation degree of small particles in region R, in units of pieces / m, is the estimated number of small particles in the area, in pieces; is the area volume, in m 3 In addition, the directional distribution and group arrangement pattern of small particles are calculated to provide supplementary information for compaction assessment.

[0124] The surface morphology feature extraction unit 37 is used to analyze the surface roughness characteristics at multiple scales and construct a roughness distribution function. In the present invention, a local surface mesh model is first reconstructed based on the point cloud data, and then morphological operations are applied to extract surface structural features. The calculation is based on the standard deviation of the height variation:

[0125] .

[0126] in: For location The surface roughness at , in cm; is the number of points in the calculation window, dimensionless; is the height value of the i-th point, in cm; The average height of the points in the window, in cm. This indicates that the calculation results for all N points within the window are accumulated. In practice, the roughness calculation window is typically set to 20 cm × 20 cm, and the roughness grading scale is 5 levels (from very smooth to very rough). A well-compacted roadbed surface typically exhibits a low roughness value (R < 0.5 cm).

[0127] The multi-temporal data fusion unit 38 is used to fuse point cloud data collected from multiple rolling operations to increase the effective point cloud density. Since the roller rolls over the same area multiple times during the construction process, the system takes advantage of this feature to increase the effective point cloud density through multi-temporal point cloud data fusion:

[0128] .

[0129] in: is the fused point cloud dataset, is the point cloud dataset collected at time t, Transform To point cloud Transform to reference time Coordinate system operations, This method can accumulate multiple rolling data, significantly improve the point cloud density, and make detailed feature analysis more reliable.

[0130] like Figure 4 As shown, the compaction degree field construction module 4 includes a feature normalization unit 41, a feature weighting unit 42, a multi-scale feature fusion unit 43, a nonlinear mapping unit 44, a local adaptive adjustment unit 45 and a compaction degree field evolution prediction unit 46.

[0131] The feature normalization unit 41 is used to normalize different features to eliminate dimensional differences. , and its normalization process uses the minimum-maximum normalization method:

[0132] .

[0133] in: is the normalized eigenvalue, dimensionless, ranging from [0,1]; is the raw eigenvalue, the unit depends on the feature type; and Characteristics The minimum and maximum values ​​of Through normalization, all eigenvalues ​​are mapped to the interval [0, 1], which facilitates subsequent feature fusion and processing.

[0134] The feature weighting unit 42 is used to assign dynamic weights to different features according to the roadbed type and engineering requirements. In the preferred embodiment of the present invention, the feature weight vector Defined as:

[0135] .

[0136] in: is the feature weight vector, dimensionless; is the geometric feature weight, dimensionless, ranging from [0,1]; is the particle feature weight, dimensionless, ranging from [0,1]; is the morphological feature weight, dimensionless, ranging from [0,1]; is the topological feature weight, dimensionless, ranging from [0,1]; and satisfies In practical applications, the weights of each feature are adjusted accordingly depending on the type of roadbed. For example, for coarse-grained soil roadbed, the weight of the particle feature is higher ( ), while for fine-grained soil roadbed, the geometric feature weight is higher The weight adjustment range is usually between 0.1-0.9 to ensure the adaptability of the system to different roadbed materials.

[0137] In this invention, wavelet transform is not used to directly increase feature dimensions, but is used as an effective feature analysis and dimensionality reduction tool. The specific implementation is as follows:

[0138] The multi-scale feature fusion unit 43 actually adopts a combination strategy of wavelet transform and feature selection. After performing wavelet decomposition on the original feature vector F, the final feature dimension is controlled through importance evaluation and selection operations:

[0139] .

[0140] in: is the feature representation after wavelet transform, is the original eigenvector, is the maximum number of layers of wavelet decomposition (set to 4), is the current scale level, dimensionless, ranging from 1 to J; is the position index at the current scale level, dimensionless; is the wavelet coefficient, dimensionless; is the wavelet function, dimensionless; is the approximate coefficient at the coarsest scale, dimensionless; is a scaling function and is dimensionless. Indicates accumulation of all scale levels and all positions k; Indicates the accumulation of all positions k at the coarsest scale.

[0141] However, after the wavelet transform, the system does not retain all coefficients, but uses sparsity constraints and principal component analysis to reduce the dimension:

[0142] .

[0143] in: is the eigenvector after PCA dimensionality reduction, is the cumulative variance contribution rate threshold, set to 95%. In addition, the system also calculates the correlation coefficient matrix between features and merges features with correlations higher than the threshold (set to 0.8) to further reduce feature redundancy. This combined strategy of decomposition-selection-dimensionality reduction ensures that the number of samples ( and feature dimensions The reasonable proportion of the system is strictly controlled in actual application , effectively preventing the overfitting problem.

[0144] Specific dimensionality reduction strategies include:

[0145] 1. Sparsity constraint: Utilizing the sparse expression characteristics of wavelet transform, only wavelet coefficients with amplitudes exceeding a specified threshold are retained. This can usually retain 90% of the information while reducing the coefficients by more than 70%.

[0146] 2. Principal Component Analysis (PCA) Dimensionality Reduction: Apply PCA to the features after wavelet transformation and select the number of principal components based on the cumulative variance contribution rate (set to 95%):

[0147] .

[0148] in: is the eigenvector after PCA dimensionality reduction, is the cumulative variance contribution rate threshold.

[0149] 3. Feature correlation analysis: Calculate the correlation coefficient matrix between features and merge features with correlations above a threshold (set to 0.8), significantly reducing feature redundancy.

[0150] 4. This combined strategy of decomposition-selection-dimensionality reduction ensures the number of samples ( and feature dimensions The reasonable proportion of the system is strictly controlled in actual application , effectively preventing overfitting problems. At the same time, regularization and cross-validation techniques are also used in the model training process to further improve generalization capabilities.

[0151] The nonlinear mapping unit 44 is used to establish a nonlinear mapping function from the feature space to the compaction value. In the present invention, the kernel method is used to implement the nonlinear mapping. :

[0152] .

[0153] in: For location The compaction value at , dimensionless, ranges from [0,1]; is the number of training samples, dimensionless; is the weight coefficient of the i-th training sample, dimensionless; is the kernel function, and the eigenvector is calculated and The similarity between them is dimensionless; is the bias term, dimensionless. Commonly used kernel functions include radial basis function (RBF) ,in is the kernel parameter, which controls the smoothness of the function, and its unit is 1 / square of the eigenvector norm; polynomial kernel ,in is a constant term, dimensionless, is the degree of the polynomial, dimensionless. and Obtained through training data optimization. In practical applications, the complexity of the mapping model is designed to express more than 50 nonlinear relationships to meet the complex mapping requirements of different roadbed materials and compaction states.

[0154] The local adaptive adjustment unit 45 is used to dynamically adjust the mapping parameters according to the local characteristics. In the present invention, considering the spatial continuity constraint, the initial compaction value is locally smoothed:

[0155] .

[0156] in: is the adjusted compaction value, dimensionless, ranging from [0,1]; for point The neighborhood of point A set of adjacent points; Neighborhood A point in is the spatial weight function, dimensionless, representing the point Point the extent of the impact; for point The initial compaction value at , dimensionless, ranges from [0,1]. The spatial weight function is usually defined as a decreasing function of distance ,in is the smoothing parameter, the unit is 1 / m ,control the degree of smoothness; for point and The Euclidean distance between points in the neighborhood is expressed in meters. The numerator represents the sum of the weighted compactness values ​​of all points in the neighborhood, and the denominator represents the sum of the weights, which is used for normalization. In practice, the local adjustment window size is typically set to 1m×1m to balance local detail preservation and global consistency.

[0157] The compaction field evolution prediction unit 46 is used to predict the changing trend of the roadbed compaction under given vibration parameters. In the present invention, a dynamic evolution model of the compaction state is constructed:

[0158] .

[0159] in: For the moment The compaction state of the roadbed is represented as a state vector, which includes multiple parameters such as compaction degree and particle arrangement. is the time step, in s; For the moment The compaction state of the roadbed; is the vibration parameter vector, including frequency (in Hz), amplitude (in mm) and duration (in s); is the roadbed material characteristic vector, including parameters such as material type and moisture content; is the state evolution function, which predicts the state at the next moment based on the current state, vibration parameters and material properties. Based on physical principles and historical data, the system can predict compaction trends under different vibration parameters. In practice, the state space dimension is set to 8, including key parameters such as compaction and particle characteristics, with a time step of 0.5 seconds and a prediction accuracy requirement of less than 5%.

[0160] like Figure 5 As shown, the compaction visualization module 5 includes a space discretization unit 51 , a color mapping unit 52 , a contour extraction unit 53 , a three-dimensional rendering unit 54 and a real-time updating unit 55 .

[0161] The space discretization unit 51 is used to discretize the continuous compaction field into a grid structure. In the present invention, the three-dimensional space is discretized into a regular grid. , each grid point corresponds to a compaction value In practical applications, the grid resolution is usually set to 10 cm × 10 cm to balance visualization accuracy and computational efficiency.

[0162] The color mapping unit 52 is used to establish a mapping function from the compaction value to the RGB color space. In a preferred embodiment of the present invention, a blue-green-red color mapping scheme is used, corresponding to the change from low to high compaction:

[0163] .

[0164] in: is the compaction value The corresponding RGB color, using triples Indicates that the range of each component is is the compaction value, dimensionless, ranging from [0,1]; and is the color conversion threshold, dimensionless, usually set to , The first condition Corresponding to the low compaction area, the color transitions from pure blue (0,0,1) to light blue; the second condition Corresponding to the medium compaction area, the color transitions from light blue to pure green (0,1,0); the third condition ( ) corresponds to high compaction areas, and the color transitions from pure green to pure red (1,0,0). This color mapping scheme intuitively expresses the distribution of compaction, with blue representing low compaction areas ( 0.6), green represents the medium compaction area ( ), red indicates high compaction area ( ).

[0165] The contour extraction unit 53 is used to extract a set of contour lines based on a specific compaction threshold. In the present invention, the marchingcubes algorithm is used to extract contour lines from the three-dimensional compaction field:

[0166] .

[0167] in: is a set of contour lines; is the i-th contour line, indicating that the compaction value is equal to The set of points; For spatial points The compaction value at , dimensionless, ranges from [0,1]; is the ith preset compaction contour threshold, dimensionless, ranging from [0,1]; is the number of contour lines, dimensionless. In practical applications, the contour line interval is usually set to 10% compaction degree, that is, , to clearly show the boundaries of areas with different compaction degrees.

[0168] The 3D rendering unit 54 is used to overlay and display color coding and contour lines in the 3D view. In the present invention, OpenGL is used to implement 3D rendering, which supports interactive operations such as rotation, zooming and translation, allowing users to observe the compaction distribution from different angles.

[0169] The real-time update unit 55 is used to dynamically update the visualization results to reflect the compaction progress. In a preferred embodiment of the present invention, the visualization update frequency is 2 Hz, ensuring real-time display of changes in the compaction state and providing timely visual feedback to the operator.

[0170] like Figure 6As shown, the vibration control module 6 includes a compaction level judgment unit 61 , a compaction requirement map generation unit 62 , a path optimization unit 63 , a vibration parameter mapping unit 64 , a state feedback control unit 65 and an automatic execution unit 66 .

[0171] The compaction level judgment unit 61 is used to divide the roadbed area into three categories: uncompacted, preliminarily compacted, and compacted. In the present invention, the roadbed area is divided into the following three levels according to the compaction level value C:

[0172] Uncompacted: C<0.7;

[0173] Initial compaction: ;

[0174] Compacted: ;

[0175] This grading method provides a decision-making basis for subsequent vibration control, and different compaction levels correspond to different vibration strategies.

[0176] The compaction requirement graph generation unit 62 is used to generate a weight graph representing the compaction requirement based on the compaction degree field. Defined as:

[0177] .

[0178] in: For location The compaction demand weight at , dimensionless, range [0,1]; For location The compaction degree value at is dimensionless and ranges from [0,1]. The compaction demand weight directly reflects the compaction priority of different areas. The higher the weight, the more urgent the compaction demand. In practical applications, the path planning unit size is usually set to ,Compaction priority is divided into 5 levels, from highest priority to negligible.

[0179] The path optimization unit 63 is used to plan the optimal path by comprehensively considering the compaction requirements, energy consumption and time. In the present invention, the path optimization problem can be expressed as:

[0180] .

[0181] in: is a sequence of path points, each Represents a two-dimensional coordinate point is the dimensionless number of waypoints; From point arrive Energy consumption, in joules (J); is the time consumed, in seconds. Indicates the accumulation of the calculation results of all adjacent point pairs in the path. Minimizing the objective function means finding a path with low energy consumption, short time and passing through the high compaction requirement area. In practical applications, the path optimization goal is usually set to compaction quality first, that is, , to ensure that areas with high compaction needs are prioritized.

[0182] The vibration parameter mapping unit 64 is used to establish a mapping relationship between compaction requirements and vibration parameters. In the present invention, the vibration parameter vector The mapping relationship between (frequency, amplitude and duration) and compaction level is:

[0183] Uncompacted: , ;

[0184] Initial compaction: ;

[0185] Compacted: , , ;

[0186] This mapping ensures that vibration parameters match compaction requirements, avoiding problems such as over-compaction or under-compaction.

[0187] The state feedback control unit 65 is used to dynamically adjust the vibration parameters based on the real-time compaction effect. In the present invention, a closed-loop feedback control strategy is adopted to adjust the vibration parameters according to the compaction gain. Adjust vibration parameters:

[0188] .

[0189] in: For the moment The vibration parameter vector includes frequency (in Hz), amplitude (in mm) and duration (in s); For the moment The vibration parameter vector of For the moment The actual compaction gain is the increase in compaction degree per unit time and is dimensionless; is the target compaction gain, dimensionless; K is the control gain matrix, 3 × 1, used to convert compaction gain errors into adjustments to the vibration parameters. In practice, the control response time is typically set to ≤ 0.2 s to ensure that the system can respond promptly to changes in the compaction state.

[0190] Automatic execution unit 66 controls the vibrator to perform compaction operations along the planned path and automatically stops vibration when the target compaction level is reached. In a preferred embodiment of the present invention, when the compaction level of an area reaches the target value (typically 90%), the system automatically stops vibrating in that area and moves to the next target area, achieving fully automatic compaction control.

[0191] like Figure 7 As shown, the roadbed compaction laser radar point cloud evaluation method of the present invention includes the following steps:

[0192] Step 1: Collect lidar point cloud data of the roadbed surface.

[0193] In a preferred embodiment of the present invention, a 128-line laser radar installed in the roller of a road roller is used to collect point cloud data of the roadbed surface at a scanning frequency of 40 Hz. The point cloud density reaches 400 points / m², ensuring the accuracy and comprehensiveness of the data.

[0194] Step 2: Apply adaptive multi-scale octree spatial segmentation to the point cloud data, dynamically adjust the segmentation threshold based on the local point cloud density and geometric structure characteristics, supplemented by material properties reference, maintain the topological characteristics of the roadbed structure, and construct a multi-level spatial representation.

[0195] The core of this step lies in the adaptive segmentation mechanism. The system dynamically adjusts the segmentation threshold based on point cloud density (typically between 100 and 1,000 points / m³) and geometric structural features. Material properties serve as a secondary reference factor, influencing only 20% of the weight. Preliminary region segmentation is performed using the DBSCAN density clustering algorithm, relying primarily on the spatial relationship of points rather than reflection intensity. Simultaneously, the system integrates roller vibration feedback data and GPS location information to achieve multi-source data cross-validation and improve the reliability of region segmentation. The integrity of the roadbed structure is maintained through topological analysis (connectivity is determined at a distance of 10 cm, and the topological feature threshold is set at a connectivity change of no more than 20% in each dimension). Finally, a five-scale space is constructed, with a minimum resolution of 5 cm, achieving multi-scale spatial representation.

[0196] Step 3: Extract the differential geometric features of the roadbed point cloud, calculate the roadbed surface curvature field tensor, analyze the particle distribution characteristics and their population statistical properties, extract the surface morphological features, and improve the analysis accuracy through multi-temporal data fusion.

[0197] This step first estimates the normal vector of each point using principal component analysis (K is 30 nearest neighbors, and the normal vector smoothing radius is 15 cm). The curvature tensor is then calculated (with a principal curvature threshold of ±0.5 / m) and a continuous curvature field is constructed (with a resolution of 10 cm). A region growing algorithm is used to identify and segment individual particles (with a particle spacing change threshold of greater than 200%, and a minimum particle size adjusted to 5 cm³), and their morphological characteristics are analyzed. For particles smaller than 5 cm³, individual identification is no longer performed, and instead, regional statistical analysis is used to assess the distribution characteristics of the particle population. Furthermore, surface roughness is calculated based on the deviation of height variation from the local fitting plane. The calculation window is adjusted to 25 cm × 25 cm to better match the point cloud resolution, and a grading scale of 5 is used. By fusing point cloud data collected through multiple rolling operations, the effective point cloud density is significantly increased, making detailed feature analysis more reliable.

[0198] Step 4: Construct the compaction field, normalize and weight the features, optimize the feature space through feature dimensionality reduction and selection, and establish a mapping from discrete point cloud to continuous compaction field.

[0199] This step first normalizes the different features to eliminate dimensional differences, and then assigns dynamic weights based on the roadbed type (weight adjustment range is 0.1-0.9). Features are analyzed using wavelet transform, but rather than increasing feature dimensions, sparsity constraints and principal component analysis are combined to reduce and select features. In practical applications, the system strictly controls the ratio of sample number to feature dimension (N ≥ 15D), effectively preventing overfitting. The core step is to establish a nonlinear mapping function that maps the feature space to compaction values ​​(the complexity of the mapping model can express more than 50 nonlinear relationships). At the same time, considering spatial continuity constraints, the compaction values ​​are locally adaptively adjusted (adjusting the window size to 1m×1m). Finally, a dynamic evolution model of the compaction state is constructed to predict the impact of vibration parameters on compaction (the state space dimension is 8, the time step is 0.5s, and the prediction accuracy requirement is less than 5%).

[0200] Step 5: Generate compaction visualization results and convert the compaction field into a color-coded contour image.

[0201] This step first discretizes the continuous compaction field into a grid structure (with a resolution of 10 cm × 10 cm). A mapping of compaction values ​​to RGB color space is then established (blue indicates low compaction, green indicates medium compaction, and red indicates high compaction). Next, a set of contour lines (intervals of 10% compaction) is extracted and displayed overlaid with the color coding and contour lines in a 3D view. The visualization updates in real time (at a frequency of 2 Hz) to provide an intuitive overview of compaction progress.

[0202] Step 6: Based on the compaction visualization results, plan the vibration path, control the roller vibration parameters, and perform the roadbed compaction operation.

[0203] This step first divides the roadbed area into three compaction levels (uncompacted: <70%, preliminarily compacted: 70%-90%, and compacted: >90%). A weighted map representing compaction requirements is then generated, and an optimal vibration path is planned (optimization prioritizes compaction quality). Vibration parameters (frequency, amplitude, and duration) are mapped to each compaction level and dynamically adjusted based on real-time compaction performance (control response time ≤ 0.2s). Finally, the vibrator is controlled to perform compaction along the planned path, automatically stopping when the target compaction level is reached.

[0204] Through the above six steps, the present invention realizes the intelligence of the entire process from point cloud data acquisition to compaction control, greatly improving the roadbed compaction quality and construction efficiency.

[0205] In summary, the roadbed compaction lidar point cloud assessment system and method of the present invention, through innovative technologies such as adaptive multi-scale octree space segmentation, differential geometry feature extraction and multi-scale feature fusion, achieves accurate assessment and intelligent control of roadbed compaction, significantly improves roadbed compaction quality and construction efficiency, and has important engineering application value.

[0206] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. Roadbed compaction laser radar point cloud assessment system, characterized by: include: The LiDAR point cloud acquisition module is installed in the roller of the road roller to collect point cloud data of the roadbed surface; An adaptive multi-scale octree space segmentation module is connected to the lidar point cloud acquisition module, and is used to receive the point cloud data, dynamically adjust the segmentation threshold based on the point cloud density, material type reference coefficient and engineering adjustment factor of any area in the point cloud space, and construct a multi-level spatial representation; A differential geometry feature extraction module, connected to the adaptive multi-scale octree space segmentation module, is used to calculate the curvature tensor of the roadbed surface, analyze the particle distribution characteristics, and extract surface morphological features; a compaction field construction module, connected to the differential geometry feature extraction module, for receiving the curvature tensor, particle distribution features, and surface morphology features, fusing multi-scale features, and establishing a mapping from discrete point clouds to a continuous compaction field; a compaction visualization module, connected to the compaction field construction module, for converting the compaction field into a color-coded contour image; a vibration control module, connected to the compaction visualization module, for planning a vibration path based on the color-coded contour image and controlling vibration parameters of the roller; The adaptive multi-scale octree space segmentation module includes: The density calculation unit is used to calculate the point cloud density of any area in the point cloud space; for any area R in the point cloud space, the point cloud density The calculation formula is: , in: is the point cloud density of region R, in points / m³; is the number of point clouds in region R, in points; is the volume of region R, in m³; The structured point cloud clustering unit is used to perform preliminary region division based on the geometric distribution and spatial continuity of the point cloud. The structured point cloud clustering unit uses the DBSCAN density clustering algorithm to achieve region division. The clustering formula is: ， in: Indicates a point and The connection relationship, is the Euclidean distance between two points, is the distance threshold, in meters; The material property reference unit is used to assist in identifying the regional material type based on the point cloud reflection intensity and spatial distribution characteristics, and provide a reference coefficient. The material property reference unit adopts a comprehensive strategy of multi-feature fusion and environmental factor correction: ， in: is the corrected reflection intensity, is the original reflection intensity, is the angle between the laser beam and the surface normal vector; an adaptive threshold determination unit, configured to dynamically determine an optimal segmentation threshold based on the point cloud density, the material type reference coefficient, and the engineering adjustment factor of the arbitrary area; The adaptive threshold determination unit adopts an improved adaptive threshold function , and its calculation formula is: ， in: is the adaptive segmentation threshold, which indicates the minimum number of points in the octree node, in points; is the basic threshold coefficient, dimensionless, ranging from 0.05 to 0.2; is the regional point cloud density, in points / m³; is the density influence factor, dimensionless, with a value of 0.5-0.7; is the material type reference coefficient, dimensionless; is the engineering adjustment factor, dimensionless, ranging from 0.8 to 1.2; Topological characteristic analysis unit, used to identify the connected area set in the point cloud and calculate the topological feature vector of the connected area to ensure that the topological characteristics are not destroyed before and after segmentation; Multi-scale link construction unit, used to establish link relationships between octree nodes of different scales to achieve smooth transition between different resolution levels; A multi-source data cross-validation unit is used to integrate roller vibration feedback data and GPS location information to improve the reliability of regional division.

2. The roadbed compaction laser radar point cloud assessment system according to claim 1, characterized in that: The differential geometry feature extraction module includes: Normal vector estimation unit, used to estimate the normal vector of each point based on its K nearest neighbor point set; Curvature tensor calculation unit, used to construct the curvature tensor of each point, including the principal curvature and principal direction; The curvature field construction unit is used to construct a continuous curvature field function based on the curvature tensor of discrete points; Particle detection and segmentation unit, used to identify and segment independent particle units from point clouds; A particle morphology analysis unit, used to calculate the morphological feature vector of each particle; Particle population statistics unit, used to perform regional statistical analysis on particles smaller than the identification limit; Surface morphology feature extraction unit, used to analyze surface roughness characteristics at multiple scales and construct roughness distribution function; The multi-temporal data fusion unit is used to fuse point cloud data collected by multiple rolling operations to increase the effective point cloud density.

3. The roadbed compaction laser radar point cloud assessment system according to claim 1, characterized in that: The compaction field construction module includes: Feature normalization unit, used to normalize different features and eliminate dimensional differences; Feature weighting unit, used to assign dynamic weights to different features based on roadbed type and engineering requirements; Feature dimension reduction and selection unit, used to achieve feature dimension reduction and optimal selection through wavelet transform and principal component analysis; A nonlinear mapping unit, used to establish a nonlinear mapping function from feature space to compaction value; A local adaptive adjustment unit, used to dynamically adjust mapping parameters according to local characteristics; The compaction field evolution prediction unit is used to predict the changing trend of roadbed compaction under given vibration parameters.

4. The roadbed compaction laser radar point cloud assessment system according to claim 1, characterized in that: The compaction visualization module includes: A spatial discretization unit is used to discretize the continuous compaction field into a grid structure; A color mapping unit, used to establish a mapping function from compaction value to RGB color space; A contour extraction unit, used to extract a set of contour lines based on a specific compaction threshold; A 3D rendering unit for displaying color coding and contour lines in a 3D view; Real-time update unit, used to dynamically update the visualization results to reflect the compaction progress.

5. The roadbed compaction laser radar point cloud assessment system according to claim 1, characterized in that: The vibration control module includes: The compaction grade judgment unit is used to classify the roadbed area into three categories: uncompacted, preliminarily compacted, and compacted; A compaction requirement graph generating unit, configured to generate a weight graph representing compaction requirements based on the compaction degree field; Path optimization unit, which plans the optimal path by comprehensively considering compaction requirements, energy consumption, and time; A vibration parameter mapping unit, used to establish a mapping relationship between compaction requirements and vibration parameters; State feedback control unit, used to dynamically adjust vibration parameters based on real-time compaction results; The automatic execution unit is used to control the vibrator to perform compaction operations along the planned path and automatically stop vibration when the target compaction degree is reached.

6. The roadbed compaction laser radar point cloud assessment system according to claim 1, characterized in that: In the topological characteristic analysis unit, the connectivity judgment distance is 10 cm, the topological feature threshold is that the connectivity change in each dimension does not exceed 20%, the structural complexity change tolerance is ±15%, and the minimum connected area size is 0.3 m³.

7. The roadbed compaction laser radar point cloud assessment system according to claim 2, characterized in that: In the normal vector estimation unit, the number of neighboring points K is 30 points, and the normal vector smoothing radius is 15 cm; in the curvature tensor calculation unit, the principal curvature threshold is ±0.5 / m; in the particle detection and segmentation unit, the particle recognition threshold is that the point spacing change rate greater than 200% is considered a particle boundary, and the minimum particle size is 5 cm³; in the surface morphology feature extraction unit, the roughness calculation window is 25 cm×25 cm, and the roughness grading standard is 5 levels.

8. The roadbed compaction laser radar point cloud assessment system according to claim 4, characterized in that: In the color mapping unit, the color mapping scheme uses blue to represent low compaction areas, green to represent medium compaction areas, and red to represent high compaction areas; in the contour line extraction unit, the contour line interval is 10% compaction.

9. A method for evaluating roadbed compaction using a laser radar point cloud, comprising: The following steps are involved: Collect lidar point cloud data of the roadbed surface; Applying adaptive multi-scale octree spatial segmentation to the point cloud data, dynamically adjusting the segmentation threshold based on the point cloud density, material type reference coefficient, and engineering adjustment factor of any region in the point cloud space, maintaining the topological characteristics of the roadbed structure, and constructing a multi-level spatial representation; Extract differential geometric features of roadbed point clouds, calculate the curvature tensor of the roadbed surface, analyze particle distribution characteristics and their population statistics, extract surface morphological features, and improve analysis accuracy through multi-temporal data fusion; Construct a compaction field, normalize and weight the features, optimize the feature space through feature dimensionality reduction and selection, and establish a mapping from discrete point clouds to a continuous compaction field; generating a compaction visualization result, converting the compaction field into a color-coded contour image; According to the compaction visualization result, a vibration path is planned, vibration parameters of the roller are controlled, and the roadbed compaction operation is performed.

Images