Lane line geometry reconstruction method based on high-frequency navigation track, medium and equipment

Abstract

The present application relates to the technical field of lane line recognition, and in particular to a lane line geometry reconstruction method based on high-frequency navigation trajectories, a medium and equipment. The lane line geometry reconstruction method based on high-frequency navigation trajectories comprises: collecting high-frequency navigation trajectory data and performing data cleaning; identifying intersections and independent road sections of a target trunk road; introducing a variational Bayesian Gaussian mixture model, using a fine-grained lane number estimation method combining geometric projection and statistical modeling, and an HDBSCAN clustering algorithm to obtain a plurality of lane clusters, fitting the lane line geometry based on the longest trajectory in each lane cluster and road width constraints to obtain a fitted center line corresponding to each lane cluster; and offsetting the fitted center line by half a lane width on both sides according to road design specifications to complete lane line geometry reconstruction of the target trunk road. The present application realizes high-precision extraction of city trunk road lane-level geometry, has good feasibility and robustness, and has practical engineering application potential.

Description

Technical Field

[0001] This invention relates to the field of lane line recognition technology, and in particular to a method, medium, and device for lane line geometry reconstruction based on high-frequency navigation trajectories. Background Technology

[0002] With the rapid evolution of intelligent transportation systems and autonomous driving technologies, high-precision maps are no longer merely navigation tools, but have been redefined as an indispensable "digital infrastructure" for autonomous driving. At this critical juncture of transitioning from assisted driving to Level 3 and above advanced autonomous driving, vehicles' demand for geospatial data has shifted from traditional road-level topology navigation to a reliance on lane-level geometric accuracy and fine-grained semantic information. Lane-level information, as the skeleton of high-precision maps, carries the underlying logic for centimeter-level positioning, refined path planning, and safety decisions in complex scenarios. However, traditional methods relying on mobile mapping systems suffer from significant bottlenecks such as high data collection costs, delayed update cycles, and limited coverage, making it difficult to meet the real-time requirements of dynamically changing urban arterial road environments. Furthermore, compared to relatively regular road types such as highways, lane geometry reconstruction on urban arterial roads faces challenges such as stronger errors, frequent lane changes, and temporary stops and detours, making it difficult to maintain accuracy robustness under conditions of topological changes and strong noise.

[0003] Current research mainly employs image-based and trajectory-based lane-level road network extraction methods, providing diverse approaches for lane-level road network reconstruction. Image-based methods have advantages in spatial continuity and geometric accuracy, making them suitable for constructing large-scale, high-precision base maps in a single operation. However, they heavily rely on high-resolution, low-occlusion, and timely-updated image data, and have limited ability to characterize severe occlusion, worn road markings, and temporary traffic organization changes on urban arterial roads. Furthermore, the costs of large-scale resampling and reprocessing are also significant.

[0004] Trajectory-based data-driven solutions use trajectories recorded by floating car GPS devices as the primary data source. By mining the spatial and temporal distribution features of these trajectories, they achieve lane number inference and lane line extraction. Trajectory-based data-driven methods have advantages such as wide coverage, low acquisition cost, and insensitivity to illumination and occlusion. They can carry out low-cost and sustainable lane-level attribute updates and geometry construction on roads. However, the accuracy limit of such methods is strongly constrained by data quality and behavioral noise, making lane number inference, trajectory clustering, and lane line fitting prone to instability under different road types. This limits their generalization ability and geometric accuracy in complex scenarios.

[0005] Under the conditions of urban arterial roads, there are more prominent challenges in terms of structural complexity and dynamism: On the one hand, most existing urban lane line constructions use low-frequency, low-quality trajectory data, which is insufficient for mining full-sample trajectory data of road space. The low density of floating car trajectories makes it difficult to obtain their own features, thus limiting their generalization ability and usability in complex urban arterial road scenarios. On the other hand, whether the lane line data obtained in the current research meets the actual needs has not been practically verified. This is mainly manifested in the difficulty of effectively evaluating the lane line extraction accuracy and geometric accuracy, making it impossible to practically verify its actual effectiveness.

[0006] Therefore, it is necessary to provide a new method, medium, and device for lane line geometry reconstruction based on high-frequency navigation trajectories to solve the above-mentioned technical problems. Summary of the Invention

[0007] The main objective of this invention is to provide a method, medium, and device for lane line geometry reconstruction based on high-frequency navigation trajectories, aiming to solve the problems of high cost, long update cycle, and limited coverage of existing road information extraction methods.

[0008] To achieve the above objectives, the present invention proposes a lane line geometry reconstruction method based on high-frequency navigation trajectory, comprising the following steps: S1: Use a mobile device to acquire high-frequency navigation trajectory data, including the trajectories of several vehicles, at a set acquisition frequency; S2: Perform data cleaning on the high-frequency navigation trajectory data to obtain preprocessed trajectory data; S3: Identify intersections and independent road segments of the target main road based on the preprocessed trajectory data; S4: Estimate the number of lanes and extract lane-level geometry from independent road segments, specifically including: S4.1 Introducing a variational Bayesian-Gaussian mixture model, a fine-grained lane number estimation method that integrates geometric projection and statistical modeling is used to estimate the number of lanes and obtain the corresponding number of lanes. S4.2. Using the number of lanes as a priori constraint parameter, perform the HDBSCAN clustering algorithm to obtain several lane clusters. Assign a lane label to each trajectory within each lane cluster and assign it to all trajectory points in the trajectory. Among them, noisy trajectory points are marked as invalid labels. S4.3. Based on the longest trajectory within each lane cluster, the geometric shape of the lane lines is fitted by road width constraints to obtain the fitted center line corresponding to each lane cluster. S4.4. The fitted centerline of each lane is offset to both sides by half the lane width according to the road design specifications to generate the lane boundary line, thus completing the geometric reconstruction of the lane lines of the target main road.

[0009] Optionally, the trajectory points in the high-frequency navigation trajectory data are represented as: ,in: For the first A trajectory point, , , , as well as Representing the first The timestamp, longitude, latitude, speed, and heading angle of each trajectory point; S2 includes: S2.1, Remove duplicate trajectory points, abnormal heading angle points, jitter points, and abnormal speed points from the trajectory points, specifically: By calculating the distance and heading angle difference between the repeated trajectory point and its normal adjacent trajectory points before and after it, the repeated trajectory points can be identified and eliminated. Calculate the geometric heading angle between adjacent trajectory points ,like The corresponding trajectory point is a heading angle anomaly, and these heading angle anomalies are removed; where: I The preset angle threshold; Calculate the actual ground distance between adjacent points using the spherical distance formula. ,like The corresponding trajectory point is the jitter point, and the jitter point is removed; where: D The preset distance threshold; The average velocity is calculated based on the distance and time difference between adjacent trajectory points. ,like The corresponding trajectory point is a velocity anomaly, and these velocity anomalies are removed; where: V The preset speed threshold; S2.2. Use a recognition method based on the spatial envelope shape of the trajectory to filter out local circular trajectories formed by vehicle detouring or long-term parking; S2.3. Use a screening method based on spatial local density to filter out sparsely sampled trajectory points; S2.4. An adaptive filtering method based on Delaunay triangulation and median absolute deviation is used to filter out isolated discrete trajectory points and obtain preprocessed trajectory data.

[0010] Optionally, S2.2 includes: S2.2.1 Calculate the minimum bounding rectangle of each trajectory to obtain the aspect ratio of the trajectory. R With area S ; S2.2.2, Aspect Ratio Based on Trajectory R With area S Calculate the weighted comprehensive evaluation index The trajectory shapes are quantified, and the specific calculation formula is as follows: ; in: Adjustable weights; S2.2.3. Set a threshold based on the weighted comprehensive evaluation index quantiles of all trajectories. Screening score lower than The local circular trajectory.

[0011] Optionally, S2.3 includes: S2.3.1 Calculate the spherical distance between any two trajectory points and construct a Ball-Tree spatial index structure based on it; S2.3.2 Based on a preset radius, use the Ball-Tree spatial index structure to count the number of trajectory points in the neighborhood of each trajectory point; S2.3.3 Calculate the quantiles of the density of all trajectory points and set a threshold. Eliminate those with a density lower than The trajectory points.

[0012] Optionally, S2.4 includes: S2.4.1 Construct a Delaunay triangulation for all trajectory points, calculate the area threshold using the Median Absolute Deviation (MAD) method, and remove triangles in the Delaunay triangulation whose area exceeds the area threshold. The vertices of triangles exceeding the area threshold are the discrete points. S2.4.2 Calculate the heading angle between adjacent trajectory points on the trajectory after removing discrete points. And using chord length distance The heading angle of adjacent trajectory points Heading angle recorded by the trajectory The directional difference index is calculated using the following formula: ; S2.4.3 Calculate all trajectory points The median absolute deviation of the values ​​is calculated, and a deviation threshold is set to filter out low-precision trajectory points whose directions are inconsistent and exceed the deviation threshold.

[0013] Optionally, S3 includes: S3.1. Based on the preprocessed trajectory data, combined with changes in vehicle motion direction and spatial clustering patterns, identify the intersections of the target main roads and assign intersection markers to the corresponding trajectory points. Specifically, this includes: S3.1.1 Extract the trajectory turning direction and calculate the instantaneous angular velocity between trajectory points, and select points that exceed the threshold as candidate turning points; S3.1.2 Perform DBSCAN clustering on the index, turning radius, and minimum number of points of the candidate turning points, and cluster the spatially continuous candidate turning points into turning trajectories with the turning radius as the search range; S3.1.3 For each cluster, calculate the total turning angle of the two segments consisting of the starting point, the middle point and the ending point, and calculate the time taken for turning. S3.1.4, Based on the preset turning angle threshold and time threshold Eliminate invalid trajectories; S3.1.5. Use the MeanShift clustering algorithm to perform spatial clustering on the trajectory points corresponding to all the turning trajectories after removal. The cluster center is the intersection center point C. S3.1.6. Using C as the center, construct a KD-Tree index for all trajectory points, and perform adaptive region expansion to obtain the final intersection region. Specifically: S3.1.7 Gradually expand the search radius For the first m The second expansion adds indexes of trajectory points falling within the ring zone. The set is: ;in: ; S3.1.8 Calculate the average turning rate and average speed of each newly added trajectory point. If the average turning rate is below the threshold And the average speed is higher than Then, the expansion stops, the final intersection area is obtained, and all trajectory points within the intersection area are assigned intersection markers. Where: average turning rate is the percentage of turning points; S3.2. Based on intersection markings and time continuity, the trajectory is divided into several independent road segments.

[0014] Optionally, S3.2 includes: S3.2.1 Determine candidate breakpoints for independent road segments, including: The status of a trajectory point is determined based on the difference between the intersection markers of adjacent trajectory points within the intersection area. Specifically, a difference of 1 indicates that the trajectory has entered the intersection, a difference of -1 indicates that the trajectory has left the intersection, and a difference of 0 indicates that the trajectory is within the intersection area. Trajectory points with non-zero differences are then designated as candidate breakpoints. Calculate the time interval between adjacent points ,like Then the trajectory point is marked as a candidate breakpoint, where: The preset time threshold; Adjacent starting points of different trajectories are also marked as candidate breakpoints; S3.2.2. The trajectory is divided into multiple sub-trajectories using candidate breakpoints and labeled. The trajectory points in the intersection area are numbered with negative values. S3.2.3 Extract the latitude and longitude coordinates of the starting points of all non-intersection sub-trajectories. Latitude and longitude coordinates of the endpoint Construct the minimum distance matrix between the endpoints of the sub-trajectories. ,by As input, DBSCAN clustering is used to determine the main road distance radius. Using a threshold, sub-trajectories that are spatially close are grouped into the same independent road segment, with the minimum distance matrix. The specific formula is as follows: ; in: For sub-trajectories The starting point For sub-trajectories b The starting point For sub-trajectories The end point For sub-trajectories b The end point.

[0015] Optionally, S4.1 includes: S4.1.1 Within each independent road segment, select the trajectory that is closest to the median of all trajectories and whose lateral deviation does not exceed the set deviation threshold as the initial road centerline; S4.1.2. Using the extracted initial road centerline as a reference line, establish a Frenet coordinate system, and project all trajectory points within the independent road segment onto this coordinate system to obtain the arc length position of each trajectory point. s and lateral offset d ; S4.1.3 Calculate the lateral distance sequence of each trajectory relative to the initial road centerline in each independent road segment in the Frenet coordinate system, and take the median of the lateral distance sequence. A one-dimensional lane feature set is constructed as a robust descriptor representing the lateral position of the trajectory. Among them: the first n The median of the lateral offset of the trajectory is denoted as ; S4.1.4 Introduce a variational Bayesian-Gaussian mixture model, assuming that the lateral offset follows a coefficient... K A Gaussian mixture distribution consisting of Gaussian components is used to construct the median of each lateral distance sequence based on the model parameters of a variational Bayesian Gaussian mixture model. The probability density function; S4.1.5 Treat the model parameters of the variational Bayesian Gaussian mixture model as random variables and assign them a prior distribution. Infer the approximate posterior distribution through variational inference, and determine the optimal model parameters through the variational Bayesian expectation-maximization algorithm. This yields the trained variational Bayesian Gaussian mixture model. S4.1.6, One-dimensional lane feature set As input, the trained variational Bayesian-Gaussian mixture model outputs a soft-assigned probability vector for each trajectory. ,in: ; S4.1.7 Extract the maximum probability value from the soft-assignment probability vector of each trajectory. Set the filter confidence threshold ,like ≥ If so, the trajectory will be strictly assigned to the corresponding lane; if < If the noise is high, it is identified as a lane-changing vehicle or high-drift GPS noise and is removed. S4.1.8 Determine the number of lanes based on the trained variational Bayesian-Gaussian mixture model and the standard lane width range in the city.

[0016] Optionally, in S4.1.4, the specific formula for the probability density function is as follows: ; in, For the set of model parameters, k For Gaussian component index, k Take 1 to K natural numbers, For the first k The mixing weights of Gaussian components, For the first k The mean of the Gaussian components, For the first k Precision parameters of each Gaussian component K This represents the total number of Gaussian components.

[0017] Optionally, S4.1.5 includes: ① For mixed weights Assigning symmetric Dirichlet priors The mean of each Gaussian component and accuracy Using the Gamma distribution, the specific formula is as follows: ; in: For concentration parameters; ② Introduce latent variables Indicates lateral offset Is it by the first k Generates _ Gaussian components, where: ;when When, it indicates a sample Assigned to the k One Gaussian component; when When, it indicates a sample It does not belong to this Gaussian component; ③ Calculate the first based on the current variational distribution parameters t The posterior expectation of the latent variables in the next iteration, i.e., the degree of responsibility. The specific formula is as follows: ; in: For expectations; for The probability, Omission and Gaussian component index k After removing irrelevant constant terms, the normalized expression for the degree of responsibility is: ; ④ Update the concentration parameters of the Dirichlet distribution and the hyperparameters of the Gamma distribution; the specific formulas are as follows: ; ; ; ; ; in: For the iterative update of the 1st k Dirichlet posterior concentration parameters corresponding to each Gaussian component; For the first k Number of valid statistical samples corresponding to each Gaussian component ; The Gaussian mean prior initial accuracy hyperparameter; To update the posterior precision parameters of the mean distribution of the Gaussian components; The initial prior mean hyperparameter of the mean distribution; For the updated version k The posterior expected parameter of the mean of each Gaussian component; The first one is calculated based on the weighted sample responsibility value. k Mean of Gaussian component samples ; and These are the precision parameter Gamma, the shape parameter of the posterior distribution, and the rate parameter, respectively. and These are the prior hyperparameters corresponding to the shape parameter and the rate parameter, respectively; For the first k Responsibility-weighted dispersion of each Gaussian component; ⑤ Repeat steps ③ and ④ until the change in the lower bound of evidence between two consecutive iterations is less than the preset threshold. Or the number of iterations reaches the preset maximum number of iterations. The trained variational Bayesian Gaussian mixture model is obtained.

[0018] Optionally, S4.1.8 includes: Obtain the weights, mean, and standard deviation of each Gaussian component in the trained variational Bayesian Gaussian mixture model; The initial number of lanes is determined by filtering effective Gaussian components using component weight thresholds. Estimating physical lane width based on the 95% confidence interval formula of normal distribution; If the estimated width exceeds the standard lane width range in the city, mark the road segment as GPS interference or road topology anomaly; otherwise, retain the component. The number of valid components that conform to the standard lane width range in the city is the lane number. J .

[0019] Optionally, S4.3 includes: S4.3.1. Using the longest trajectory within each lane cluster as the initial lane centerline, and resampling it at equal intervals, an initial point sequence is obtained. Let the number of iterations be... h =0; S4.3.2, For each point on the current lane centerline Calculate its unit normal vector ; and with Centered on a point, construct a cross-sectional line segment extending along both sides of the normal direction, with a length equal to half the lane width. Find the intersection points of this cross-sectional line segment with all trajectories within the lane cluster, and project these intersection points onto the normal coordinate system to calculate the average value. ; S4.3.3, Based on unit normal vector and average Update the center point of the cross section along the normal direction, and generate the lane centerline for the current iteration based on the updated center point of the cross section. The specific update formula is as follows: ; in: For the first The trajectory point on the current lane centerline in the next iteration; S4.3.4 Calculate the average distance between the lane centerline of the current iteration and all trajectory points of the previous iteration. If the average distance is less than a preset threshold, the iteration terminates and the lane centerline of the current iteration is output as the final fitted centerline of the lane cluster; otherwise, return to step S4.3.2.

[0020] In addition, the present invention also provides a readable storage medium storing computer program instructions, which, when executed by a processor, implement the lane line geometry reconstruction method based on high-frequency navigation trajectory as described above.

[0021] The present invention also provides an electronic device, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, wherein the computer program instructions are executed by the processor to perform the lane line geometry reconstruction method based on high-frequency navigation trajectory as described above.

[0022] This invention reconstructs lane line geometry based on the vehicle's high-frequency navigation trajectory, enabling self-identification and fitting of lane numbers and centerlines without relying on other external information. Specific beneficial effects are as follows: This invention introduces the VB-GMM model and identifies lane numbers that match the actual number of lanes on a main road through a lane number identification method based on geometric projection and statistical models. Compared with traditional GMM models, this invention can autonomously select the optimal number of lanes for the corresponding main road without manual specification. Furthermore, the VB-GMM model of this invention avoids the pitfalls of traditional GMM models getting trapped in local optima. The lane number calculation of this invention matches the true value of the actual number of lanes and is more accurate in reflecting the actual situation compared to traditional GMM models that identify too many or too detailed lane numbers.

[0023] To evaluate the number of lanes, the actual lane range line is used as the true value. The trajectories of each cluster are mapped to the actual true value range, and the accuracy of each main road lane is statistically analyzed. The overall accuracy rate reaches 94%, which basically conforms to the actual lane range.

[0024] For lane geometry accuracy evaluation, the lane centerline of the actual road network is taken as the true value and discretized into an ordered set of points with fixed intervals. The lateral distance between each lane point in each arterial road and the true value point of the lane centerline is calculated and the mean value is obtained. This mean distance is used as the geometry accuracy error of the lane line. The average lateral distance of the lane lines of the entire arterial road is collected, and the records of each arterial road are statistically analyzed as the final geometry accuracy evaluation of the arterial road. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the lane line geometry reconstruction method based on high-frequency navigation trajectory in an embodiment of the present invention; Figure 2 This is a lane lateral distribution map based on the VB-GMM model in an embodiment of the present invention; Figure 3 This is a lateral lane distribution map based on the traditional GMM method.

[0027] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0029] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0030] Furthermore, in this invention, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0031] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0032] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0033] This invention proposes a lane line geometry reconstruction method, medium, and device based on high-frequency navigation trajectory, aiming to solve the problems of high cost, long update cycle, and limited coverage of existing road information extraction methods.

[0034] See Figure 1 This embodiment provides a lane line geometry reconstruction method based on high-frequency navigation trajectory. Based on field investigation and empirical calibration, the threshold parameters in this embodiment are set as follows: Heading angle difference threshold jitter trajectory distance threshold Speed ​​threshold ; Quantile threshold for detour trajectory Low density radius quantile threshold High angular velocity threshold Turning radius Minimum number of turning points Turning angle threshold Turning time threshold Intersection Expansion Step Size Maximum expansion radius Average turning rate threshold Intersection exit speed threshold The 85th percentile of the velocities of all trajectories was determined by statistical analysis, and the data in this dataset was selected from these quantiles. ; Mark segmentation time threshold Main road distance radius Standard lane width Component weight threshold ; Filter confidence threshold Lane centerline iterative convergence threshold ; Specifically, the following steps are included: S1: Acquire high-frequency navigation trajectory data, including several vehicle trajectories, using a mobile device at a set acquisition frequency; wherein: the acquisition time interval is 1s-3s, and the trajectory points in the high-frequency navigation trajectory data are represented as... ,in: For the first A trajectory point, , , , as well as Representing the first The timestamp, longitude, latitude, speed, and heading angle of each trajectory point; S2: Perform data cleaning on the high-frequency navigation trajectory data to obtain preprocessed trajectory data; S2 includes: S2.1, Remove duplicate trajectory points, abnormal heading angle points, jitter points, and abnormal speed points from the trajectory points, specifically: By calculating the distance and heading angle difference between the repeated trajectory point and its normal adjacent trajectory points before and after it, the repeated trajectory points can be identified and eliminated. Calculate the geometric heading angle between adjacent trajectory points ,like The corresponding trajectory point is a heading angle anomaly, and these heading angle anomalies are removed; where: I The preset angle threshold; Calculate the actual ground distance between adjacent points using the spherical distance formula. ,like The corresponding trajectory point is the jitter point, and the jitter point is removed; where: D The preset distance threshold; The average velocity is calculated based on the distance and time difference between adjacent trajectory points. ,like The corresponding trajectory point is a velocity anomaly, and these velocity anomalies are removed; where: V The preset speed threshold; S2.2, Employ a trajectory spatial envelope-based recognition method to filter out local circular trajectories formed by vehicle detours or prolonged stops; specifically including: S2.2.1 Calculate the minimum bounding rectangle of each trajectory to obtain the aspect ratio of the trajectory. R With area S ; S2.2.2, Aspect Ratio Based on Trajectory R With area S Calculate the weighted comprehensive evaluation index The trajectory shapes are quantified, and the specific calculation formula is as follows: ; in: The weight is adjustable; it is preferably set to 0.5 so that the aspect ratio and area contribute equally to the score.

[0035] S2.2.3. Set a threshold based on the weighted comprehensive evaluation index quantiles of all trajectories. Screening score lower than The local circular trajectory; in this embodiment, .

[0036] S2.3, Employ a spatial local density-based screening method to remove sparsely sampled trajectory points; specifically including: S2.3.1 Calculate the spherical distance between any two trajectory points and construct a Ball-Tree spatial index structure based on it; S2.3.2 Based on a preset radius, use the Ball-Tree spatial index structure to count the number of trajectory points in the neighborhood of each trajectory point; S2.3.3 Calculate the quantiles of the density of all trajectory points and set a threshold. Eliminate those with a density lower than The trajectory points; S2.4. An adaptive filtering method based on Delaunay triangulation and median absolute deviation is used to filter out isolated discrete trajectory points, resulting in preprocessed trajectory data; specifically including: S2.4.1. Construct a Delaunay triangulation for all trajectory points. Calculate the area threshold using the Median Absolute Deviation (MAD) method, and remove triangles from the Delaunay triangulation whose areas exceed the area threshold. The vertices of these triangles are then considered discrete points. Specifically: connect the trajectory points to form triangles, calculate the area of ​​each triangle, calculate the area threshold using the MAD method, and remove triangles with excessively large areas (i.e., those that are far apart or represent discrete trajectories). The vertices of these triangles are then considered discrete points.

[0037] S2.4.2 Calculate the heading angle between adjacent trajectory points on the trajectory after removing discrete points. And using chord length distance The heading angle between adjacent trajectory points Heading angle recorded by the trajectory The directional difference index is calculated using the following formula: ; when The smaller the value, the more consistent the two directions are. By using the median absolute deviation (MAD) of the chord length distance to filter out trajectory points with abnormal distances, low-precision trajectories with inconsistent directions can be effectively eliminated.

[0038] S2.4.3 Calculate all trajectory points The median absolute deviation of the values ​​is calculated, and a deviation threshold is set to filter out low-precision trajectory points whose directions are inconsistent and exceed the deviation threshold.

[0039] S3: Identify intersections and independent road segments of the target main road based on the preprocessed trajectory data; S3 includes: S3.1. Based on the preprocessed trajectory data, combined with changes in vehicle motion direction and spatial clustering patterns, identify the intersections of the target main roads and assign intersection markers to the corresponding trajectory points. Specifically, this includes: S3.1.1 Extract the trajectory turning direction and calculate the instantaneous angular velocity between trajectory points, and select points that exceed the threshold as candidate turning points; S3.1.2 Perform DBSCAN clustering on the index, turning radius, and minimum number of points of the candidate turning points, and cluster the spatially continuous candidate turning points into turning trajectories with the turning radius as the search range; In this embodiment, the turning radius refers to the index of candidate turning points. Using this radius as a range, other nearby turning points are searched, and this is used to determine whether the turning radius consists of consecutive candidate turning points. Specifically: First, the turning direction of the trajectory is extracted, and the heading angle and time of each trajectory are obtained to calculate the angular velocity. Excessively large angular velocities are selected as the candidate angular velocity set. The candidate high angular velocity points are clustered, aggregating discrete high angular velocity trajectory points into consecutive turning segments.

[0040] S3.1.3 For each cluster, calculate the total turning angle of the two segments consisting of the starting point, the middle point and the ending point, and calculate the time taken for turning. S3.1.4, Based on the preset turning angle threshold and time threshold Eliminate invalid trajectories; S3.1.5. Use the MeanShift clustering algorithm to spatially cluster the trajectory points corresponding to all the removed turning trajectories. The cluster centers are the intersection center points. C ; S3.1.6, with C Using a circle as the center, construct a KD-Tree to index all trajectory points, and perform adaptive region expansion to obtain the final intersection area. Specifically: Gradually expand the search radius For the first m The second expansion adds indexes of trajectory points falling within the ring zone. The set is: ;in: ; Calculate the average turning rate and average speed for each newly added trajectory point. If the average turning rate is below the threshold And the average speed is higher than Then, the expansion stops, the final intersection area is obtained, and all trajectory points within the intersection area are assigned intersection markers. Where: average turning rate is the percentage of turning points; S3.2. Based on intersection markings and time continuity, the trajectory is divided into several independent road segments, specifically including: S3.2.1 Determine candidate breakpoints for independent road segments, including: The status of a trajectory point is determined based on the difference between the intersection markers of adjacent trajectory points within the intersection area. Specifically, a difference of 1 indicates that the trajectory has entered the intersection, a difference of -1 indicates that the trajectory has left the intersection, and a difference of 0 indicates that the trajectory is within the intersection area. Trajectory points with non-zero differences are then designated as candidate breakpoints. Calculate the time interval between adjacent points ,like Then the trajectory point is marked as a candidate breakpoint, where: The preset time threshold; Adjacent starting points of different trajectories are also marked as candidate breakpoints. Each trajectory in this embodiment has a teammate's vehicle ID. Since the trajectories are stored in order, the last trajectory point (i.e., the end point) of the previous trajectory is directly connected to the starting trajectory point (i.e., the beginning point) of the current trajectory. Candidate breakpoints are set to prevent misjudgment as the same trajectory.

[0041] S3.2.2. The trajectory is divided into multiple sub-trajectories using candidate breakpoints and labeled. The trajectory points in the intersection area are numbered with negative values. S3.2.3 Extract the latitude and longitude coordinates of the starting points of all non-intersection sub-trajectories. Latitude and longitude coordinates of the endpoint Construct the minimum distance matrix between the endpoints of the sub-trajectories. ,by As input, DBSCAN clustering is used to determine the main road distance radius. Using a threshold, sub-trajectories that are spatially close are grouped into the same independent road segment, with the minimum distance matrix. The specific formula is as follows: ;

[0042] in: For sub-trajectories The starting point For sub-trajectories b The starting point For sub-trajectories The end point For sub-trajectories b The end point.

[0043] S4: Estimate the number of lanes and extract lane-level geometry from independent road segments, specifically including: S4.1 Introducing a variational Bayesian-Gaussian mixture model, a fine-grained lane number estimation method that integrates geometric projection and statistical modeling is used to estimate the number of lanes and obtain the corresponding number of lanes. S4.1 includes: S4.1.1 Within each independent road segment, select the trajectory that is closest to the median of all trajectories and whose lateral deviation does not exceed the set deviation threshold as the initial road centerline; S4.1.2. Using the extracted initial road centerline as a reference line, establish a Frenet coordinate system, and project all trajectory points within the independent road segment onto this coordinate system to obtain the arc length position of each trajectory point. s and lateral offset d ; S4.1.3 Calculate the lateral distance sequence of each trajectory relative to the initial road centerline in each independent road segment in the Frenet coordinate system, and take the median of the lateral distance sequence. A one-dimensional lane feature set is constructed as a robust descriptor representing the lateral position of the trajectory. Among them: the first n The median of the lateral offset of the trajectory is denoted as ; S4.1.4 Introduce the Variational Bayesian Gaussian Mixture Model (VB-GMM model), assuming that the lateral offset follows a formula... K A Gaussian mixture distribution consisting of Gaussian components is used to construct the median of each lateral distance sequence based on the model parameters of a variational Bayesian Gaussian mixture model. The probability density function; In S4.1.4, the specific formula for the probability density function is as follows: ; in, For the set of model parameters, k For Gaussian component index, k Take 1 to K natural numbers, For the first k The mixing weights of Gaussian components, For the first k The mean of the Gaussian components, For the first k Precision parameters of each Gaussian component K This represents the total number of Gaussian components.

[0044] This embodiment also includes: alongs Equidistant sampling in direction, calculate the maximum lateral distance between trajectory points on the cross section. minimum lateral distance Obtain the lateral width of the road Traversing all cross-sections along the reference line yields a width sequence. A rough lane count is then calculated using a formula, and this rough lane count is used as the upper bound of the VB-GMM model. ,in: Standard lane width; S4.1.5 Treat the model parameters of the variational Bayesian Gaussian mixture model as random variables and assign them a prior distribution. Infer the approximate posterior distribution through variational inference, and determine the optimal model parameters through the variational Bayesian expectation-maximization algorithm. This yields the trained variational Bayesian Gaussian mixture model. S4.1.5 includes: ① For mixed weights Assigning symmetric Dirichlet priors The mean of each Gaussian component and accuracy Using the Gamma distribution, the specific formula is as follows: ; in: For concentration parameters; ② Introduce latent variables Indicates lateral offset Is it by the first k Generates _ Gaussian components, where: when When, it indicates a sample Assigned to the kth One Gaussian component; when When, it indicates a sample It does not belong to this Gaussian component; ③ Calculate the first based on the current variational distribution parameters t The posterior expectation of the latent variables in the next iteration, i.e., the degree of responsibility. The specific formula is as follows: ; in: For expectations; for The probability, Omission and Gaussian component index k After removing irrelevant constant terms, the normalized expression for the degree of responsibility is: ; ④ Update the concentration parameters of the Dirichlet distribution and the hyperparameters of the Gamma distribution; the specific formulas are as follows: ; ; ; ; ; in: For the iterative update of the 1st k Dirichlet posterior concentration parameters corresponding to each Gaussian component. For the first k Number of valid statistical samples corresponding to each Gaussian component . The initial accuracy hyperparameter is the Gaussian mean prior. This is the posterior accuracy parameter for the updated Gaussian component mean distribution. The initial prior mean hyperparameter of the mean distribution. For the updated version k The posterior expected parameters of the mean of each Gaussian component. The first one is calculated based on the weighted sample responsibility value. k Mean of Gaussian component samples ; and These are the precision parameter Gamma, the shape parameter of the posterior distribution, and the rate parameter, respectively. and These are the prior hyperparameters corresponding to the shape parameter and the rate parameter, respectively; For the first k Responsibility-weighted dispersion of each Gaussian component; ⑤ Repeat steps ③ and ④ until the change in the lower bound of evidence between two consecutive iterations is less than the preset threshold. Or the number of iterations reaches the preset maximum number of iterations. The trained variational Bayesian Gaussian mixture model is obtained.

[0045] S4.1.6. Using the trained variational Bayesian Gaussian mixture model, output a soft-assigned probability vector for each trajectory. Wherein: the input dataset for the trained variational Bayesian-Gaussian mixture model is ; S4.1.7 Extract the maximum probability value from the soft-assignment probability vector of each trajectory. Set the filter confidence threshold ,like ≥ If so, the trajectory will be strictly assigned to the corresponding lane; if < If the noise is high, it is identified as a lane-changing vehicle or high-drift GPS noise and is removed. S4.1.8 Determine the number of lanes based on the trained variational Bayesian-Gaussian mixture model and the standard lane width range in the city.

[0046] S4.1.8 includes: Obtain the weights, mean, and standard deviation of each Gaussian component in the trained variational Bayesian Gaussian mixture model; The initial number of lanes is determined by filtering effective Gaussian components using component weight thresholds. Estimating physical lane width based on the 95% confidence interval formula of normal distribution; If the estimated width exceeds the standard lane width range in the city, mark the road segment as GPS interference or road topology anomaly; otherwise, retain the component. The number of valid components that conform to the standard lane width range in the city is the lane number. J .

[0047] S4.2. Using the number of lanes as a priori constraint parameter, the HDBSCAN clustering algorithm is executed to obtain several lane clusters. Each trajectory within each lane cluster is assigned a lane label, which is then assigned to all trajectory points within the trajectory. Noisy trajectory points are marked as invalid labels. In this embodiment, Euclidean distance is selected as the basic distance metric. The HDBSCAN algorithm will automatically calculate the core distance and mutual reachability distance based on this metric, thereby constructing a hierarchical structure reflecting the data density distribution and adapting to the density differences of trajectory data in lateral offset. The trajectory data in the one-dimensional lane feature set D is input into the HDBSCAN model, and the core parameter is set to min_cluster_size = N Remove all items smaller than 1000. N Unstable clusters are identified. Only trajectories that "persist" in the density hierarchy are classified into stable lane clusters; trajectories that do not remain in any stable cluster are marked as noise trajectories (label=-1). The cluster selection method uses Excess of Mass (EOM), prioritizing clusters with "a large number of trajectories and high inter-cluster separation." The core optimization objective is strictly followed. ,in For the first j The number of trajectories in each lane cluster, For the first j Minimum spatial distance between each cluster and other clusters; lane labeling for each trajectory. The points in the trajectory are assigned a label, while points in noisy trajectories are marked as invalid.

[0048] In this embodiment, a unique lane label is assigned to each trajectory cluster. The label is then reverse-mapped and assigned to all trajectory points within the cluster; trajectory points identified as noise by HDBSCAN are marked as invalid labels and removed in subsequent analysis.

[0049] S4.3. Based on the longest trajectory within each lane cluster, the geometric shape of the lane lines is fitted by road width constraints to obtain the fitted center line corresponding to each lane cluster. S4.3 includes: S4.3.1. Using the longest trajectory within each lane cluster as the initial lane centerline, and resampling it at equal intervals, an initial point sequence is obtained. Let the number of iterations be... h =0; S4.3.2, For each point on the current lane centerline Calculate its unit normal vector ; and with Centered on a point, construct a cross-sectional line segment extending along both sides of the normal direction, with a length equal to half the lane width. Find the intersection points of this cross-sectional line segment with all trajectories within the lane cluster, and project these intersection points onto the normal coordinate system to calculate the average value. ; S4.3.3, Based on unit normal vector and average Update the center point of the cross section along the normal direction, and generate the lane centerline for the current iteration based on the updated center point of the cross section. The specific update formula is as follows: ; in: For the first The trajectory point on the current lane centerline in the next iteration; S4.3.4 Calculate the average distance between the lane centerline of the current iteration and all trajectory points of the previous iteration. If the average distance is less than a preset threshold, the iteration terminates and the lane centerline of the current iteration is output as the final fitted centerline of the lane cluster; otherwise, return to step S4.3.2.

[0050] S4.4. The fitted centerline of each lane is offset to both sides by half the lane width according to the road design specifications to generate the lane boundary line, thus completing the geometric reconstruction of the lane lines of the target main road.

[0051] See Figure 2 and Figure 3 , Figure 2To demonstrate the number and lateral distribution of lanes in this embodiment, the VB-GMM model, with its adaptive advantage of having no prior lane count, automatically predicts the lane distribution based on the trajectory, accurately determines the actual number of lanes in each road segment, and simultaneously fits a lane centerline distribution that closely reflects real traffic flow patterns. This method eliminates the need for manually pre-setting the number of lanes, effectively removes abnormal trajectory noise interference, provides lane count recognition that better reflects actual road conditions, and achieves higher accuracy in fitting the lane centerline distribution.

[0052] Figure 3 This paper presents the lane number and lateral distribution prediction achieved by the traditional GMM method. It utilizes GMM to predict the lane number and distribution based on the lateral features of the trajectory, ultimately displaying the total number of lanes identified by this method and the spatial distribution of the centerlines of each lane. However, this method relies on a pre-defined number of clusters, is susceptible to interference from abnormal trajectories deviating from lanes, and suffers from issues such as lane misjudgment and unreasonable lane merging / segmentation.

[0053] The lane lines extracted in this embodiment are compared and verified with the high-precision map of the city collected in the same year. The intersection areas of two main roads (area 1 and area 2) are selected for verification using 9 main road segments.

[0054] Regarding lane number verification, the number of lanes in each direction calculated by VB-GMM matches the actual number of lanes. However, the number of lanes calculated by traditional GMM requires a preset number of components and is prone to getting trapped in local optima, resulting in a non-normal lane number distribution, as shown in Table 1.

[0055] Table 1. Diagram of the number of lane center lines identified in Zone 1 and Zone 2

[0056] Regarding lane geometric accuracy, this embodiment employs a quantitative analysis method to evaluate lane geometric accuracy. Using manually marked lane centerlines as the true value, these centerlines are discretized into ordered sets of points with fixed intervals. The lateral distance between each lane point in each main road and the true value point of the lane centerline is calculated, and the average value is obtained. This average distance is used as the geometric accuracy error of the lane line.

[0057] In addition, this embodiment calculates the distance between the extracted lane centerline and the lane centerline of the high-precision map to obtain the average lateral deviation, with an accuracy of about 1m, as shown in Table 2.

[0058] Table 2. Geometric accuracy evaluation of main roads in Region 1 and Region 2

[0059] Experimental results show that the method proposed in this embodiment can achieve high-precision extraction of lane-level geometry of urban arterial roads using only low-cost GPS trajectory data. It has good feasibility and robustness and has potential for practical engineering applications.

[0060] This embodiment also includes a readable storage medium storing computer program instructions, which, when executed by a processor, implement the lane line geometry reconstruction method based on high-frequency navigation trajectory as described above.

[0061] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0062] This embodiment also includes an electronic device, comprising: at least one processor, at least one memory, and computer program instructions stored in the memory, wherein the computer program instructions are executed by the processor to perform the lane line geometry reconstruction method based on high-frequency navigation trajectory as described above.

[0063] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0064] The electronic device can be a mobile phone, desktop computer, laptop, handheld computer, cloud server, or other computing device. The electronic device may include, but is not limited to, processors and memory. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0065] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0066] The memory can be used to store the computer program and / or modules. The processor implements the computer program by running or executing the computer program and / or modules stored in the memory, and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0067] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0068] The above description is only a preferred embodiment of the present invention and does not limit the scope of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the protection scope of the present invention.

Claims

1. A lane line geometry reconstruction method based on high-frequency navigation trajectory, characterized in that, Includes the following steps: S1: Use a mobile device to acquire high-frequency navigation trajectory data, including the trajectories of several vehicles, at a set acquisition frequency; S2: Perform data cleaning on the high-frequency navigation trajectory data to obtain preprocessed trajectory data; S3: Identify intersections and independent road segments of the target main road based on the preprocessed trajectory data; S4: Estimate the number of lanes and extract lane-level geometry from independent road segments, specifically including: S4.1 Introducing a variational Bayesian-Gaussian mixture model, a fine-grained lane number estimation method that integrates geometric projection and statistical modeling is used to estimate the number of lanes and obtain the corresponding number of lanes. S4.1 includes: S4.1.1 Within each independent road segment, select the trajectory that is closest to the median of all trajectories and whose lateral deviation does not exceed the set deviation threshold as the initial road centerline; S4.1.

2. Using the extracted initial road centerline as a reference line, establish a Frenet coordinate system, and project all trajectory points within the independent road segment onto this coordinate system to obtain the arc length position of each trajectory point. s and lateral offset d ; S4.1.3 Calculate the lateral distance sequence of each trajectory relative to the initial road centerline in each independent road segment in the Frenet coordinate system, and take the median of the lateral distance sequence. A one-dimensional lane feature set is constructed as a robust descriptor representing the lateral position of the trajectory. Among them: the first n The median of the lateral offset of the trajectory is denoted as ; S4.1.4 Introduce a variational Bayesian-Gaussian mixture model, assuming that the lateral offset follows a coefficient... K A Gaussian mixture distribution consisting of Gaussian components is used to construct the median of each lateral distance sequence based on the model parameters of a variational Bayesian Gaussian mixture model. The probability density function; S4.1.5 Treat the model parameters of the variational Bayesian Gaussian mixture model as random variables and assign them a prior distribution. Infer the approximate posterior distribution through variational inference, and determine the optimal model parameters through the variational Bayesian expectation-maximization algorithm. This yields the trained variational Bayesian Gaussian mixture model. S4.1.6, One-dimensional lane feature set As input, the trained variational Bayesian-Gaussian mixture model outputs a soft-assigned probability vector for each trajectory. ,in: ; S4.1.7 Extract the maximum probability value from the soft-assignment probability vector of each trajectory. Set the filter confidence threshold ,like ≥ If so, the trajectory will be strictly assigned to the corresponding lane; if < If the noise is high, it is identified as a lane-changing vehicle or high-drift GPS noise and is removed. S4.1.8 Determine the number of lanes based on the trained variational Bayesian-Gaussian mixture model and the standard lane width range in the city; S4.

2. Using the number of lanes as a priori constraint parameter, perform the HDBSCAN clustering algorithm to obtain several lane clusters. Assign a lane label to each trajectory within each lane cluster and assign it to all trajectory points in the trajectory. Among them, noisy trajectory points are marked as invalid labels. S4.

3. Based on the longest trajectory within each lane cluster, the geometric shape of the lane lines is fitted by road width constraints to obtain the fitted center line corresponding to each lane cluster. S4.

4. The fitted centerline of each lane is offset to both sides by half the lane width according to the road design specifications to generate the lane boundary line, thus completing the geometric reconstruction of the lane lines of the target main road.

2. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to claim 1, characterized in that, The trajectory points in high-frequency navigation trajectory data are represented as follows: ,in: For the first A trajectory point, , , , as well as Representing the first The timestamp, longitude, latitude, speed, and heading angle of each trajectory point; S2 includes: S2.1, Remove duplicate trajectory points, abnormal heading angle points, jitter points, and abnormal speed points from the trajectory points, specifically: By calculating the distance and heading angle difference between the repeated trajectory point and its normal adjacent trajectory points before and after it, the repeated trajectory points can be identified and eliminated. Calculate the geometric heading angle between adjacent trajectory points ,like The corresponding trajectory point is a heading angle anomaly, and these heading angle anomalies are removed; where: I The preset angle threshold; Calculate the actual ground distance between adjacent points using the spherical distance formula. ,like The corresponding trajectory point is the jitter point, and the jitter point is removed; where: D The preset distance threshold; The average velocity is calculated based on the distance and time difference between adjacent trajectory points. ,like The corresponding trajectory point is a velocity anomaly, and these velocity anomalies are removed; where: V The preset speed threshold; S2.

2. Use a recognition method based on the spatial envelope shape of the trajectory to filter out local circular trajectories formed by vehicle detouring or long-term parking; S2.

3. Use a screening method based on spatial local density to filter out sparsely sampled trajectory points; S2.

4. An adaptive filtering method based on Delaunay triangulation and median absolute deviation is used to filter out isolated discrete trajectory points and obtain preprocessed trajectory data.

3. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to claim 2, characterized in that, S2.2 includes: S2.2.1 Calculate the minimum bounding rectangle of each trajectory to obtain the aspect ratio of the trajectory. R With area S ; S2.2.2, Aspect Ratio Based on Trajectory R With area S Calculate the weighted comprehensive evaluation index The trajectory shapes are quantified, and the specific calculation formula is as follows: ； in: Adjustable weights; S2.2.

3. Set a threshold based on the weighted comprehensive evaluation index quantiles of all trajectories. Screening score lower than The local circular trajectory.

4. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to claim 2, characterized in that, S2.3 includes: S2.3.1 Calculate the spherical distance between any two trajectory points and construct a Ball-Tree spatial index structure based on it; S2.3.2 Based on a preset radius, use the Ball-Tree spatial index structure to count the number of trajectory points in the neighborhood of each trajectory point; S2.3.3 Calculate the quantiles of the density of all trajectory points and set a threshold. Eliminate those with a density lower than The trajectory points.

5. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to claim 2, characterized in that, S2.4 includes: S2.4.1 Construct a Delaunay triangulation for all trajectory points, calculate the area threshold using the Median Absolute Deviation (MAD) method, and remove triangles in the Delaunay triangulation whose area exceeds the area threshold. The vertices of triangles exceeding the area threshold are the discrete points. S2.4.2 Calculate the heading angle between adjacent trajectory points on the trajectory after removing discrete points. And using chord length distance The heading angle between adjacent trajectory points Heading angle recorded by the trajectory The directional difference index is calculated using the following formula: ； S2.4.3 Calculate all trajectory points The median absolute deviation of the values ​​is calculated, and a deviation threshold is set to filter out low-precision trajectory points whose directions are inconsistent and exceed the deviation threshold.

6. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to claim 1, characterized in that, S3 includes: S3.

1. Based on the preprocessed trajectory data, combined with changes in vehicle motion direction and spatial clustering patterns, identify the intersections of the target main roads and assign intersection markers to the corresponding trajectory points. Specifically, this includes: S3.1.1 Extract the trajectory turning direction and calculate the instantaneous angular velocity between trajectory points, and select points that exceed the threshold as candidate turning points; S3.1.2 Perform DBSCAN clustering on the index, turning radius, and minimum number of points of the candidate turning points, and cluster the spatially continuous candidate turning points into turning trajectories with the turning radius as the search range; S3.1.3 For each cluster, calculate the total turning angle of the two segments consisting of the starting point, the middle point and the ending point, and calculate the time taken for turning. S3.1.4, Based on the preset turning angle threshold and time threshold Eliminate invalid trajectories; S3.1.

5. Use the MeanShift clustering algorithm to spatially cluster the trajectory points corresponding to all the removed turning trajectories. The cluster centers are the intersection center points. C ; S3.1.6, with C Using a circle as the center, construct a KD-Tree to index all trajectory points, and perform adaptive region expansion to obtain the final intersection area. Specifically: Gradually expand the search radius For the first m The second expansion adds indexes of trajectory points falling within the ring zone. i The set is: ;in: ; Calculate the average turning rate and average speed for each newly added trajectory point. If the average turning rate is below the threshold And the average speed is higher than Then, the expansion stops, resulting in the final intersection area, and all trajectory points within the intersection area are assigned intersection markers. Where: average turning rate is the percentage of turning points; S3.

2. Based on intersection markings and time continuity, the trajectory is divided into several independent road segments.

7. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to claim 6, characterized in that, S3.2 includes: S3.2.1 Determine candidate breakpoints for independent road segments, including: The status of a trajectory point is determined based on the difference between the intersection markers of adjacent trajectory points within the intersection area. Specifically, a difference of 1 indicates that the trajectory has entered the intersection, a difference of -1 indicates that the trajectory has left the intersection, and a difference of 0 indicates that the trajectory is within the intersection area. Trajectory points with non-zero differences are then designated as candidate breakpoints. Calculate the time interval between adjacent points ,like Then, the trajectory point is marked as a candidate breakpoint, where: The preset time threshold; Adjacent starting points of different trajectories are also marked as candidate breakpoints; S3.2.

2. The trajectory is divided into multiple sub-trajectories using candidate breakpoints and labeled. The trajectory points in the intersection area are numbered with negative values. S3.2.3 Extract the latitude and longitude coordinates of the starting points of all non-intersection sub-trajectories. Latitude and longitude coordinates of the endpoint Construct the minimum distance matrix between the endpoints of the sub-trajectories. ,by As input, DBSCAN clustering is used to determine the main road distance radius. Using a threshold, sub-trajectories that are spatially close are grouped into the same independent road segment, with the minimum distance matrix. The specific formula is as follows: ； in: For sub-trajectories a The starting point For sub-trajectories b The starting point For sub-trajectories a The end point For sub-trajectories b The end point.

8. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to claim 1, characterized in that, In S4.1.4, the specific formula for the probability density function is as follows: ； in, For the set of model parameters, k For Gaussian component index, k Take 1 to K natural numbers, For the first k The mixing weights of Gaussian components, For the first k The mean of the Gaussian components, For the first k Accuracy parameters of each Gaussian component. K This represents the total number of Gaussian components.

9. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to claim 1, characterized in that, S4.1.5 includes: ① For mixed weights Assigning symmetric Dirichlet priors The mean of each Gaussian component and accuracy Using the Gamma distribution, the specific formula is as follows: ； in: For concentration parameters; ② Introduce latent variables Indicates lateral offset Is it by the first k Generates _ Gaussian components, where: ,when When, it indicates a sample Assigned to the k One Gaussian component; when When, it indicates a sample It does not belong to this Gaussian component; ③ Calculate the first based on the current variational distribution parameters t The posterior expectation of the latent variables in the next iteration, i.e., the degree of responsibility. The specific formula is as follows: ； in: For expectations; for The probability, Omission and Gaussian component index k After removing irrelevant constant terms, the normalized expression for the degree of responsibility is: ； ④ Update the concentration parameters of the Dirichlet distribution and the hyperparameters of the Gamma distribution; the specific formulas are as follows: ； ； ； ； ； in: For the iterative update of the 1st k Dirichlet posterior concentration parameters corresponding to each Gaussian component; For the first k Number of valid statistical samples corresponding to each Gaussian component ; The Gaussian mean prior initial accuracy hyperparameter; To update the posterior precision parameters of the mean distribution of the Gaussian components; The initial prior mean hyperparameter of the mean distribution; For the updated version k Posterior expected parameters of the mean of Gaussian components; The first one is calculated based on the weighted sample responsibility value. k Mean of Gaussian component samples ; and These are the precision parameter Gamma, the shape parameter of the posterior distribution, and the rate parameter, respectively. and These are the prior hyperparameters corresponding to the shape parameter and the rate parameter, respectively; For the first k Responsibility-weighted dispersion of each Gaussian component; ⑤ Repeat steps ③ and ④ until the change in the lower bound of evidence between two consecutive iterations is less than the preset threshold. Or the number of iterations reaches the preset maximum number of iterations. The trained variational Bayesian Gaussian mixture model is obtained.

10. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to claim 1, characterized in that, S4.1.8 includes: Obtain the weights, mean, and standard deviation of each Gaussian component in the trained variational Bayesian Gaussian mixture model; The initial number of lanes is determined by filtering effective Gaussian components using component weight thresholds. Estimating physical lane width based on the 95% confidence interval formula of normal distribution; If the estimated width exceeds the standard lane width range in the city, mark the road segment as GPS interference or road topology anomaly; otherwise, retain the component. The number of valid components that conform to the standard lane width range in the city is the lane number. J .

11. The lane line geometry reconstruction method based on high-frequency navigation trajectory according to any one of claims 1-10, characterized in that, S4.3 includes: S4.3.

1. Using the longest trajectory within each lane cluster as the initial lane centerline, and resampling it at equal intervals, an initial point sequence is obtained. Let the number of iterations be... h =0; S4.3.2, For each point on the current lane centerline Calculate its unit normal vector ; and with Centered on a point, construct a cross-sectional line segment extending along both sides of the normal direction, with a length equal to half the lane width. Find the intersection points of this cross-sectional line segment with all trajectories within the lane cluster, and project these intersection points onto the normal coordinate system to calculate the average value. ; S4.3.3, Based on unit normal vector and average Update the center point of the cross section along the normal direction, and generate the lane centerline for the current iteration based on the updated center point of the cross section. The specific update formula is as follows: ； in: For the first h+ The trajectory point on the current lane centerline in one iteration; S4.3.4 Calculate the average distance between the lane centerline of the current iteration and all trajectory points of the previous iteration. If the average distance is less than a preset threshold, the iteration terminates and the lane centerline of the current iteration is output as the final fitted centerline of the lane cluster; otherwise, return to step S4.3.

2.

12. A readable storage medium, characterized in that, It stores computer program instructions, which, when executed by a processor, implement the lane line geometry reconstruction method based on high-frequency navigation trajectory as described in claim 11.

13. An electronic device, characterized in that, include: The method for lane line geometry reconstruction based on high-frequency navigation trajectory as described in claim 11 includes at least one processor, at least one memory, and computer program instructions stored in the memory, which are executed by the processor when the computer program instructions are executed.

Images