Road marking extraction method based on laser point cloud

Abstract

The invention relates to the field of intelligent capital construction, and provides a road marking extraction method based on laser point cloud, which comprises the following steps: acquiring vehicle-mounted laser point cloud data of an existing highway road and carrying out noise filtering preprocessing, carrying out pseudo-color mapping through an intensity value to generate a visual intensity distribution diagram, establishing an initial feature point set according to marking type features, and extracting a marking line according to the initial feature point set. A target point cloud point set is iteratively calculated by adopting a dynamic sampling radius and a self-adaptive error threshold, a point set topological relation model is established, marked line extension extraction is realized through a vector space propagation algorithm, finally, a minimum outsourcing cube is constructed for different marked line types, three-dimensional space parameters are calculated, and marked line geometric feature data are generated. The method ensures the accuracy and reliability of marking line extraction, is good in performance in road scenes with different complexities, and has relatively high practicability and popularization value.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent infrastructure, in particular to a road marking extraction method based on laser point cloud. BACKGROUND

[0002] With the continuous expansion of highway scale, the traditional extensive management mode has been difficult to meet the demand of modern highway operation and maintenance. Highway digital management is an important foundation to realize intelligent transportation, and the primary task of digital construction is to complete the high-precision three-dimensional modeling of highway infrastructure. Among them, road marking is a key element to guide vehicle driving and ensure traffic safety, and the precision and efficiency of its digital modeling directly affect the overall construction effect. However, in the face of the huge highway mileage scale, the traditional manual measurement and marking method is not only inefficient and costly, but also difficult to ensure the accuracy and consistency of the data. It is estimated that it will take several years to complete the digital modeling of all highway markings by manual method, and it is difficult to deal with dynamic changes such as marking wear and update. SUMMARY

[0003] In order to solve the problems existing in the prior art, the purpose of the present application is to provide a road marking extraction method based on laser point cloud, which collects high-precision point cloud data by vehicle-mounted laser radar, uses noise filtering, color distribution, intensity mapping and other technical means, and compared with the traditional measurement method, can well deal with the situation that the marking wear or identification error is difficult to avoid, increase the data accuracy, combine the vector space propagation algorithm to quickly mark the corresponding point cloud data, realize the automatic extraction and modeling of the marking, can quickly and accurately complete the digital reconstruction of the road marking, the automatic processing method can well deal with the dynamic update problem of the marking, at the same time, improve the work efficiency of obtaining the marking information, significantly reduce the digital construction cost, and provide efficient technical support for highway digital construction.

[0004] In order to achieve the above purpose, the present application adopts the following technical measures: a road marking extraction method based on laser point cloud, comprising the following steps:

[0005] S1, obtaining vehicle-mounted laser point cloud data of existing highway road and carrying out noise filtering pretreatment;

[0006] S2, performing pseudo-color mapping processing on the point cloud by intensity value to generate a visual intensity distribution map;

[0007] S3, establishing an initial feature point set according to the type characteristics of the marking by using single-point or double-point sampling strategy;

[0008] S4, iteratively calculating the target point cloud point set based on dynamic sampling radius and adaptive error threshold;

[0009] S5, a point set topology relationship model is established, and the line extension extraction is realized through a vector space propagation algorithm;

[0010] S6, a minimum outer package cube is constructed for the entity connection type and the non-continuous type of the line respectively, and the three-dimensional space parameters are calculated to generate the line geometric feature data.

[0011] Optionally, the vehicle-mounted laser point cloud data of the existing highway road is acquired, and noise filtering preprocessing is performed, which comprises:

[0012] (1) Point cloud data loading and analysis:

[0013] The LAS1.4 format point cloud data collected by the vehicle-mounted laser radar is read through the laspy library, and the three-dimensional space coordinates (x, y, z), the reflection intensity, and the echo number field are extracted. When the original point cloud data is a relative coordinate system, a coordinate conversion model is established based on the synchronously acquired GNSS positioning data and inertial navigation information, and conversion to the WGS84 geodetic coordinate system is realized to obtain three-dimensional point cloud data with global geographical reference;

[0014] (2) Construction of spatial index:

[0015] The octree spatial index is created by using the open3d library, and the specific parameters are:

[0016] Maximum tree depth: 12 levels;

[0017] Voxel resolution: 0.1m x 0.1m x 0.05m;

[0018] Node splitting condition: when a single node contains more than 50 points, perform octant division;

[0019] Index optimization: apply a weight coefficient ω_z=1.5 to the z-axis direction to enhance the road point cloud retrieval efficiency;

[0020] (3) Noise filtering processing:

[0021] An improved statistical outlier filtering algorithm is applied to each point A spherical search domain is established with a radius r=0.3m, and the number of neighborhood points K=50 is counted to calculate the neighborhood point distance distribution:

[0022]

[0023] Wherein, is the distance between point and point , is the average distance between point and other points in the neighborhood, is the distance between point With the standard deviation of the distance between other points in the neighborhood, if any of the following conditions is met, it is marked as a noise point:

[0024]

[0025] wherein, is the mean of the intensity of all points in the neighborhood, is the intensity value of the point , and is the standard deviation of the intensity of all points in the neighborhood;

[0026] According to the determination result, a three-level filtering strategy is performed: first, primary filtering σ=1.2 is performed to remove significant outliers, then secondary filtering σ=1.0 is performed to process boundary fuzzy points, and finally the region with a density change rate ≤15% is reserved.

[0027] Optionally, the pseudo-color mapping processing of the point cloud by the intensity value to generate a visual intensity distribution map comprises: first, performing global statistical analysis on the original point cloud reflection intensity value, calculating the maximum value and the minimum value of the intensity value in the point cloud data set, and mapping the original intensity value to the interval [0, 255] through a linear normalization formula:

[0028]

[0029] Subsequently, a pseudo-color mapping model based on the HSV color space is established, the hue component H=240°×(1-I') is set to construct a gradual color spectrum from dark blue I'=0 to light blue I'=255, while the saturation component S=100% and the brightness component V=100% are fixed, and a high-contrast visual intensity distribution map is generated; finally, the HSV color space is converted to the RGB standard color space, and the color rendering of the three-dimensional point cloud data is realized through the point cloud coloring interface of the open3d library, wherein the area with an intensity value higher than the road base threshold T_s=200 is automatically marked as a red warning color, forming a road marking enhancement display effect that can be directly distinguished by the human eye.

[0030] Optionally, the initial feature point set is established according to the single-point or double-point sampling strategy according to the marking type characteristics, which comprises: first, according to the geometric continuity characteristics of the road marking, it is divided into two categories: solid connection type marking and non-continuous type marking, wherein the solid connection type marking includes lane edge solid line and center no-turn yellow line, and a single-point sampling strategy is adopted, that is, a feature point is selected as an initial sampling point at any position of the marking, which needs to meet:

[0031]

[0032] wherein, is the intensity value of the feature point, The maximum value of the intensity of the local area of the marking line;

[0033] The discontinuous marking line includes pedestrian crossing lines and dashed and solid combined lines, a double-point sampling strategy is adopted, two sampling points are required to be located at adjacent marking line segments, the line direction of the two sampling points should be consistent with the design direction of the marking line, and the direction angle deviation Δθ is controlled within ±5°.

[0034] Optionally, the iterative calculation of the target point cloud point set based on the dynamic sampling radius and the adaptive error threshold value comprises: calculating the associated point cloud point set of the target point P based on the dynamic sampling radius R and the adaptive error threshold value T through a directional region growing algorithm, and the specific process is as follows:

[0035] (1) Directional neighborhood search:

[0036] Six candidate points P1~P6 are generated by offsetting the initial point P(x0, y0, z0) by a distance R along the positive and negative axes of the three-dimensional Cartesian coordinate system, and the coordinates are calculated as:

[0037]

[0038] (2) Intensity similarity screening:

[0039] Calculate the intensity difference of each candidate point , and retain the points that meet to form a first associated point set {P i}, and introduce an intensity gradient constraint :

[0040]

[0041] Wherein, T is the adaptive error threshold value, R is the dynamic sampling radius, and the intensity gradient in the x or y direction cannot exceed 0.2 times T / R, so as to exclude interference points in the intensity mutation area;

[0042] (3) Iterative expansion strategy:

[0043] Recursively execute steps (1)-(2) in step S4 for each point in {P i}, generate a second associated point set {P ij}, and expand layer by layer until any of the following termination conditions is met:

[0044] 1) Intensity convergence condition: the proportion of newly added points in two consecutive iterations is less than 5%;

[0045] 2) Geometric constraint condition: the standard deviation of the distance between points ; ​​​

[0046] The final output target point cloud point set .

[0047] Optionally, the step of establishing a point set topology model and extracting the datum line extension using a vector space propagation algorithm includes: based on the two initial target point cloud point sets obtained in step S4. and First, calculate its three-dimensional centroid. and The centroid coordinates are determined using a weighted average method:

[0048]

[0049] in, For point set The number of points, This is the intensity weighting coefficient. Point cloud intensity value;

[0050] Then calculate the spatial vector V = between the two centroids. - and Euclidean distance L= And expand along the +V and −V directions respectively, to form a new candidate point set. The centroid O′ must satisfy:

[0051]

[0052] in, For the previous centroid, For the new vector;

[0053] If the average intensity of the new point set Deviation from historical mean ; or the reduction rate of the point set coverage area in three consecutive iterations. If the end of the marking is determined, the extension will terminate.

[0054] Optionally, the step of constructing minimum bounding cubes and calculating three-dimensional spatial parameters for both solid-connected and discontinuous markings to generate marking geometric feature data includes:

[0055] (1) Extraction of geometric features of solid connection type markings:

[0056] The principal direction vector of the point set along the extension direction of the datum line is calculated using principal component analysis to determine the datum direction vector with a fixed step size. The point cloud of the entity-connected road markings is segmented, and the range of each segment is defined as follows: Where k is the segment number, the point cloud data is divided into continuous segments along the main direction, and the starting point of the k-th segment is... The endpoint is (k+1). Generate oriented bounding boxes for each segment along the main direction, and generate point clouds within each 5m segment. Calculate its axial center point :

[0057]

[0058] in,( , , () represents the coordinates of the i-th point within the k-th point cloud's bounding box;

[0059] Finally, output the center point sequence { , ,..., } and the corresponding set of normal vectors;

[0060] (2) Extraction of geometric features of discontinuous markings:

[0061] Point cloud of independent paving segments To address the potential arbitrary directionality of the marking lines, principal component analysis is used to determine the principal direction vector of the marking line point cloud. The point cloud is then rotated to align the principal direction with the coordinate axes. The coordinates of the four corner points, width, and length parameters of each marking line segment are extracted from the boundary range of the rotated point cloud.

[0062] Furthermore, the specific logic for extracting the coordinates of the four corner points, width, and length parameters of each line segment includes:

[0063] After aligning the point cloud along the principal direction, calculate its minimum and maximum values ​​along the principal X-axis and the vertical Y-axis to obtain the boundary range [Xmin, Xmax] × [Ymin, Ymax]. Based on this, construct an outer rectangle with its four corner points as follows:

[0064] P1 = (Xmin, Ymin), P2 = (Xmax, Ymin), P3 = (Xmax, Ymax), P4 = (Xmin,Ymax);

[0065] The length L is Xmax - Xmin;

[0066] The width W is Ymax - Ymin;

[0067] The length is along the main direction of the marking, i.e., the X-axis; the width is along the vertical direction, i.e., the Y-axis.

[0068] Compared with existing technologies, this invention has the following main technical advantages: The method of this invention ensures the accuracy and reliability of road marking extraction through multi-step optimization. First, noise filtering preprocessing improves data quality, laying the foundation for subsequent processing. Pseudo-color mapping enhances the visualization of road marking features, facilitating identification. The introduction of dynamic sampling radius and adaptive error threshold optimizes the efficiency and accuracy of point cloud processing. The establishment of a topological relationship model and the application of a propagation algorithm ensure the continuity and integrity of the road markings. Finally, the construction of the minimum bounding cube and the calculation of three-dimensional parameters provide geometric feature data of the road markings, providing a reliable foundation for subsequent applications. The method of this invention performs well in road scenarios of varying complexity, demonstrating high practicality and promotional value. Detailed Implementation

[0069] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0070] A method for extracting road markings based on laser point clouds includes the following steps:

[0071] S1. Acquire vehicle-mounted laser point cloud data of existing highways and perform noise filtering preprocessing; the specific implementation method of step S1 includes:

[0072] (1) Point cloud data loading and parsing:

[0073] The laspy library is used to read LAS 1.4 format point cloud data collected by vehicle-mounted lidar and extract fields such as three-dimensional spatial coordinates (x, y, z), reflection intensity, and return number. When the original point cloud data is in a relative coordinate system, a coordinate transformation model is established based on synchronously acquired GNSS positioning data and inertial navigation information to achieve the transformation to the WGS84 geodetic coordinate system, so as to obtain three-dimensional point cloud data with global geographic reference.

[0074] (2) Spatial index construction:

[0075] An octree spatial index is created using the open3d library, with the following parameters:

[0076] Maximum tree depth: 12 levels;

[0077] Voxelization resolution: 0.1m × 0.1m × 0.05m;

[0078] Node splitting condition: When a single node contains more than 50 points, it is divided into eight equal parts.

[0079] Index optimization: Apply a weighting coefficient ω_z=1.5 to the z-axis direction to enhance the efficiency of road point cloud retrieval.

[0080] (3) Noise filtering:

[0081] An improved statistical outlier filtering algorithm is applied to each point. Establish a spherical search domain with radius r = 0.3m, count the number of neighboring points K = 50, and calculate the distance distribution of the neighboring points:

[0082]

[0083] in, For point With point spacing, For point The average distance between the points and other points in the neighborhood. For point A point is marked as noise if it meets any of the following conditions, based on the standard deviation of its distance from other points in the neighborhood:

[0084]

[0085] in, This is the average intensity of all points within the neighborhood. For point The intensity value, This represents the standard deviation of the intensity at all points within this neighborhood.

[0086] Based on the judgment results, a three-level filtering strategy is executed: first, primary filtering (σ=1.2) is performed to remove significant outliers, then secondary filtering (σ=1.0) is performed to process blurred boundary points, and finally, regions with a point cloud density change rate ≤15% are retained.

[0087] S2. Perform pseudo-color mapping on the point cloud using intensity values ​​to generate a visualized intensity distribution map;

[0088] The specific implementation method of step S2 includes: firstly, performing a global statistical analysis on the original point cloud reflection intensity values, and calculating the maximum value of the intensity values ​​in the point cloud dataset. and minimum value By using the linear normalization formula:

[0089]

[0090] The original intensity values ​​are mapped to the [0,255] interval to eliminate the inconsistency in intensity dimensions caused by sensor differences. Here, I' is the intensity value of I after normalization to the (0,255) interval. Then, a pseudo-color mapping model based on the HSV color space is established. The hue component H is set to 240°×(1-I') to construct a gradient color spectrum from dark blue (I'=0) to light blue (I'=255). At the same time, the saturation component S=100% and the lightness component V=100% are fixed to generate a high-contrast visual intensity distribution map. Finally, the HSV color space is converted to the RGB standard color space, and the color rendering of the three-dimensional point cloud data is realized through the point cloud shading interface of the Open3D library. The area with an intensity value higher than the road base threshold T_s=200 is automatically marked as a red warning color, forming a road marking display effect that can be intuitively identified by the human eye.

[0091] S3. Based on the characteristics of the marking type, establish an initial feature point set using a single-point or double-point sampling strategy.

[0092] The specific implementation method of step S3 includes: firstly, classifying road markings into two categories based on their geometric continuity characteristics: solid connection markings and discontinuous markings. Solid connection markings include marking types with continuous geometric characteristics, such as solid lane edge lines and yellow lines prohibiting lane changes in the center. A single-point sampling strategy is adopted, that is, a feature point is selected at any position of the marking as the initial sampling point, and this point must satisfy the following conditions:

[0093]

[0094] in, The intensity value of the feature point. This represents the maximum intensity of the marking in a local area.

[0095] Discontinuous road markings include pedestrian crossings, dashed and solid combination lines, and other types of road markings with discontinuous characteristics. A two-point sampling strategy is adopted, requiring two sampling points to be located in adjacent road marking segments. At the same time, the direction of the line connecting the two sampling points should be consistent with the design direction of the road marking, and the directional angle deviation Δθ should be controlled within ±5°.

[0096] S4. Based on the dynamic sampling radius R and the adaptive error threshold T, iteratively calculate the target point cloud set;

[0097] The specific implementation method of step S4 includes: calculating the associated point cloud set of the target point P using a directional region growing algorithm based on the dynamic sampling radius R and the adaptive error threshold T. The specific process is as follows:

[0098] (1) Directional neighborhood search:

[0099] Centered on the initial point P(x0,y0,z0), offset by a distance R along the positive and negative axes (front, back, left, right, up, down) of the three-dimensional Cartesian coordinate system, generating 6 candidate points P1~P6, whose coordinates are calculated as follows:

[0100]

[0101] (2) Intensity similarity screening:

[0102] Calculation points With each candidate point Strength difference , retain satisfaction The points constitute a first-level related point set {P} i Introducing intensity gradient constraints. :

[0103]

[0104] in, The adaptive error threshold is R, which is based on the dynamic sampling radius. The intensity gradient in the x or y direction cannot exceed 0.2 times T / R. Within the sampling length R, the intensity change cannot be too fast, otherwise there may be discontinuous abrupt changes in the region, thus eliminating interference points in the intensity abrupt change region.

[0105] (3) Iterative expansion strategy:

[0106] For {P i For each point in}, recursively execute steps (1)-(2) in S4 to generate a second-level associated point set {P}. ij}, and expand layer by layer until any of the following termination conditions are met:

[0107] 1) Strength convergence condition: the percentage of newly added points in two consecutive iterations 5%;

[0108] 2) Geometric constraints: The spatial distribution of the point set is too discrete, i.e., the standard deviation of the point spacing is too large. The final output is the target point cloud set. .

[0109] S5. Establish a topological relationship model of point sets and extract the extended markings using a vector space propagation algorithm;

[0110] In step S5, the target point cloud point sets for two points are calculated, and then the point clouds at other locations are deduced based on the relationship between the two target point cloud point sets. Specifically, this method includes: based on the two initial target point cloud point sets obtained in step S4... and First, calculate its three-dimensional centroid. and The centroid coordinates are determined using a weighted average method:

[0111]

[0112] in, For point set The number of points, This is the intensity weighting coefficient. This represents the point cloud intensity value.

[0113] Then calculate the spatial vector V = between the two centroids. - and Euclidean distance L= And expand along the +V and −V directions respectively, to form a new candidate point set. The centroid O′ must satisfy:

[0114]

[0115] in, For the previous centroid, This is the new vector.

[0116] If the average intensity of the new point set Deviation from historical mean Or, in three consecutive iterations, the reduction rate of the point set coverage area If the endpoint of the datum line is determined at a certain time, the extension is terminated. Here, the average intensity of the new point set is the average intensity of the point cloud corresponding to the Sp' point set, the historical average is the average intensity of the point cloud corresponding to the Sp point set, and T is the adaptive error threshold.

[0117] S6. Construct minimum bounding cubes for both solid-connected and discontinuous markings and calculate three-dimensional spatial parameters to generate marking geometric feature data.

[0118] In step S6, a minimum enclosing cube is constructed and geometric feature parameters are calculated for both connected and discontinuous markings, as specifically implemented as follows:

[0119] (1) Extraction of geometric features of solid-connected markings:

[0120] Principal component analysis (PCA) is used to calculate and determine the principal direction vector of the point set along the datum extension direction, with a fixed step size. The point cloud of the entity-connected road markings is segmented, and the range of each segment is defined as follows: Where k is the segment number, the point cloud data is divided into continuous segments along the main direction, and the starting point of the k-th segment is... The endpoint is (k+1). The actual length of each segment is always 1. For example: if =0.5 meters, then the range of the second segment (k=2) is [1.0, 1.5) meters. Generate an oriented bounding box (OBB) for each segment along the main direction, and generate the point cloud within each 5m segment. Calculate its axial center point :

[0121]

[0122] in,( , , () represents the coordinates of the i-th point in the bounding box of the k-th point cloud.

[0123] Finally, output the center point sequence { , ,..., } and the corresponding set of normal vectors.

[0124] (2) Extraction of geometric features of discontinuous markings:

[0125] Point cloud of independent line segments (such as pedestrian crossing lines) To address the potential arbitrary orientation of the marking lines, Principal Component Analysis (PCA) is used to determine the principal direction vector of the marking line point cloud. The point cloud is then rotated to align the principal direction with the coordinate axes. For the boundary range of the rotated point cloud, parameters such as the coordinates of the four corner points, width, and length of each marking line segment are extracted. The specific logic is as follows:

[0126] After aligning the point cloud along the principal direction, calculate its minimum and maximum values ​​in the principal direction (X-axis) and the vertical direction (Y-axis) to obtain the boundary range [Xmin, Xmax] × [Ymin, Ymax]. Based on this, construct an outer rectangle with its four corner points as follows:

[0127] P1 = (Xmin, Ymin), P2 = (Xmax, Ymin), P3 = (Xmax, Ymax), P4 = (Xmin,Ymax);

[0128] The length L is Xmax - Xmin;

[0129] The width W is Ymax - Ymin;

[0130] The length is along the main direction of the marking (X-axis), and the width is along the vertical direction (Y-axis).

[0131] The above description is merely a specific embodiment of the present invention, and the scope of protection of the present invention is not limited thereto. Any transformations or substitutions that can be conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for extracting road marking based on laser point cloud, characterized in that, The method comprises the following steps: S1, acquiring vehicle-mounted laser point cloud data of an existing highway road and performing noise filtering pretreatment; S2, performing pseudo-color mapping processing on the point cloud through intensity value to generate a visual intensity distribution map; S3, establishing an initial feature point set according to the type characteristics of the marking line by using a single-point or double-point sampling strategy; S4, iteratively calculating a target point cloud point set based on a dynamic sampling radius and an adaptive error threshold; S5, establishing a point set topological relationship model, and realizing marking line extension extraction through a vector space propagation algorithm; S6, constructing a minimum outer box cube for the entity connection type and the non-continuous type marking line respectively and calculating three-dimensional space parameters to generate marking line geometric feature data. 2.The method of claim 1, wherein, The acquisition of vehicle-mounted laser point cloud data of an existing highway road and the noise filtering pretreatment comprise: (1) Point cloud data loading and analysis: LAS1.4 format point cloud data collected by a vehicle-mounted laser radar is read through a laspy library, and three-dimensional space coordinates (x, y, z), reflection intensity, and echo number fields are extracted. When the original point cloud data is in a relative coordinate system, a coordinate conversion model is established based on synchronously acquired GNSS positioning data and inertial navigation information to realize conversion to a WGS84 geodetic coordinate system to obtain three-dimensional point cloud data with global geographical reference; (2) Spatial index construction: An open3d library is used to create an octree spatial index, and the specific parameters are as follows: Maximum tree depth: 12 levels; Voxel resolution: 0.1m x 0.1m x 0.05m; Node splitting condition: when a single node contains more than 50 points, perform octant division; Index optimization: apply a weight coefficient ω_z=1.5 to the z-axis direction to enhance the retrieval efficiency of road surface point cloud; (3) Noise filtering processing: The improved statistical outlier filtering algorithm is applied to each point A spherical search domain is established with a radius r = 0.3 m and a neighborhood point number K = 50, and the neighborhood point distance distribution is calculated: ； where, is a point is a point is the average distance to other points in the neighborhood, is a point is the average distance to other points in the neighborhood, is a point is the standard deviation of the distance to other points in the neighborhood, is marked as a noise point if any of the following conditions are met: ； wherein, is the mean intensity of all points in the neighborhood, is the intensity value of the point , and is the standard deviation of intensity of all points in the neighborhood; According to the determination result, a three-level filtering strategy is performed: first, primary filtering σ=1.2 is performed to remove obvious outliers, then secondary filtering σ=1.0 is performed to process boundary fuzzy points, and finally the area with a point cloud density change rate of ≤15% is retained. 3.The method of claim 1, wherein, The pseudo-color mapping processing of the point cloud by the intensity value includes: firstly, performing global statistical analysis on the original point cloud reflection intensity value, calculating the maximum value and minimum value of the intensity value in the point cloud data set And the minimum value , through the linear normalization formula: ； The original intensity value is mapped to the [0, 255] interval; then a pseudo-color mapping model based on the HSV color space is established, the hue component H=240°×(1-I') is set to construct a gradual color spectrum from deep blue I'=0 to light blue I'=255, and the saturation component S=100% and the brightness component V=100% are fixed to generate a high-contrast visual intensity distribution map; finally, the HSV color space is converted to the RGB standard color space, and the color rendering of three-dimensional point cloud data is realized through the point cloud coloring interface of the open3d library, wherein the area with an intensity value higher than the road base threshold T_s=200 is automatically marked as a red warning color, forming a road marking enhancement display effect that can be directly distinguished by the human eye. 4.The method of claim 1, wherein, The establishment of an initial feature point set according to the type characteristics of the marking line by using a single-point or double-point sampling strategy comprises the following steps: first, the road marking is divided into two categories of entity connection type marking line and non-continuous type marking line according to the geometric continuity characteristics of the road marking, wherein the entity connection type marking line includes lane edge solid line and center no-turn yellow line, and a single-point sampling strategy is adopted, that is, a feature point is selected as an initial sampling point at any position of the marking line, and the point needs to satisfy ； wherein, is a feature point intensity value, is a local area intensity maximum of the reticle. The discontinuous marking line includes a pedestrian crossing line and a virtual-actual combined line, a double-point sampling strategy is adopted, two sampling points are required to be located at adjacent marking line segments, and the direction of the line connecting the two sampling points should be consistent with the design direction of the marking line, and the direction angle deviation Δθ is controlled within ±5°. 5.The laser point cloud based road marking extraction method of claim 1, wherein, The iterative calculation of the target point cloud point set based on the dynamic sampling radius and the adaptive error threshold value includes: based on the dynamic sampling radius R and the adaptive error threshold value T, the associated point cloud point set of the target point P is calculated through the directional region growing algorithm, and the specific process is: (1) Directional neighborhood search: Taking the initial point P(x0, y0, z0) as the center, offsetting the positive and negative axes of the three-dimensional Cartesian coordinate system by a distance R, six candidate points P1~P6 are generated, and the coordinates are calculated as: ； (2) Intensity similarity screening: Computing the intensity difference of each candidate point , retaining the points satisfying to form a primary associated point set {P i}, and introducing an intensity gradient constraint : ； wherein, is an adaptive error threshold, R is based on a dynamic sampling radius, and the intensity gradient in the x or y direction cannot exceed 0.2 times T / R to exclude points of interference from regions of intensity abrupt change. (3) Iterative expansion strategy: {P i} recursively for each point in {P ij} and expand layer by layer until any of the following termination conditions is met: 1) Strength convergence condition: the percentage of newly added points in two consecutive iterations 5%; 2) Geometric constraints: point distance standard deviation ; Final output target point cloud point set . 6.The method of claim 1, wherein, The point set topological relation model is established by a vector space propagation algorithm to realize the extraction of the line extension, including: based on the two initial target point cloud point sets obtained in step S4 and , first, the three-dimensional space center of mass thereof is calculated and wherein the center of mass coordinates are determined by a weighted average method: ； wherein, is the number of points of the point set is the number of points of the point set is the intensity weight coefficient, is the point cloud intensity value; The spatial vector V of the two centroids is then calculated - and the Euclidean distance L is and expanded in the +V and −V directions respectively, the new candidate point set whose centroid O′ must satisfy: ； wherein, is the previous centroid, is the new vector; If the new point set average intensity is within a predetermined deviation from the historical average , or if the point set coverage area reduction rate is less than a predetermined threshold for three consecutive iterations, the extension is terminated.

7. The method of claim 1, wherein, The minimum outer package cube is constructed for the entity connection type and the discontinuous type marking line respectively, and the three-dimensional space parameters are calculated to generate the marking line geometric feature data, including: (1) Entity connection type marking line geometric feature extraction: The principal direction vector of the point set along the extension direction of the reticle is determined by a principal component analysis method, and the principal direction vector is determined by a principal component analysis method The reticle point cloud of the entity connection type is segmented, and the range of each segment point cloud is defined as , wherein k is the segment number, the point cloud data is divided into continuous segments along the principal direction, the starting point of the kth segment is , and the end point is (k+1) , and a directional bounding box is generated for each segment along the principal direction, and the axial center point of each 5m segment point cloud , is calculated : ； in,( , , () represents the coordinates of the i-th point within the k-th point cloud's bounding box; Finally, output the sequence of center points { , ,..., } and the corresponding set of normal vectors; (2) Discontinuous type marking line geometric feature extraction: Point cloud of independent marking segment For any directionality that the marking line may have, a principal direction vector of the marking line point cloud is determined by a principal component analysis method, the point cloud is rotated so that the principal direction is aligned with the coordinate axis, the boundary range of the rotated point cloud is calculated, and the four corner point coordinates, width and length parameters of each marking segment are extracted.

8. The method of claim 7, wherein, The four corner point coordinates, width and length parameters of each marking line segment are extracted, specifically including: After the point cloud is aligned in the main direction, the minimum and maximum values of the point cloud in the main direction X axis and the vertical direction Y axis are calculated, and the boundary range [Xmin, Xmax] × [Ymin, Ymax] is obtained, and the outer rectangle is constructed according to the boundary range, and the four corner points are: P1 = (Xmin, Ymin), P2 = (Xmax, Ymin), P3 = (Xmax, Ymax), P4 = (Xmin,Ymax); The length L is Xmax - Xmin; The width W is Ymax - Ymin; Wherein, the length is along the main direction of the marking line, that is, the X axis; the width is along the vertical direction, that is, the Y axis.

Images