Trajectory information-based road network generation method and device, equipment and storage medium

Abstract

The invention relates to a road network generation method and device based on trajectory information, equipment and a storage medium. The method comprises the steps that a vehicle track data set in an existing road network is collected, and the vehicle track data set is divided into a fixed path track set and a free path track set according to vehicle types; performing feature extraction on the fixed path track set and the free path track set to obtain a point cloud-based regional association feature, a frequent point sequence-based regional connectivity feature and a global distribution density feature with the highest GPS distribution density based on a road center line; and carrying out dimensionality reduction quantization on the regional association features, the regional connectivity features and the global distribution density features, inputting the regional association features, the regional connectivity features and the global distribution density features into an SVM (Support Vector Machine) model to carry out parallel road identification, inputting the identified trajectory data of the parallel roads into a map generation model, and generating an optimized road network. According to the embodiment of the invention, the generation effect of roads with complicated structures, such as main roads, auxiliary roads and elevated roads, is greatly improved.

Description

Technical Field

[0001] This application belongs to the field of electronic map generation technology, and specifically relates to a road network generation method, apparatus, device, and storage medium based on trajectory information. Background Technology

[0002] Obtaining real-time and accurate electronic maps is a prerequisite for location-based services such as navigation and autonomous driving. In reality, due to the extension and damage of urban roads, real-time generation or updating of electronic maps is necessary. Currently, some surveying companies utilize fleets equipped with positioning devices, laser sensors, or cameras to manually collect location data to generate high-precision electronic maps. However, this manual method is costly, time-consuming, and labor-intensive, making it difficult to meet the real-time requirements of vehicle sharing and navigation services. In recent years, with the widespread use of positioning devices in vehicles and mobile terminals, a large amount of trajectory data reflecting road connectivity and topology can be collected. Therefore, automatic electronic road network generation methods based on trajectory information have become an important research direction. Existing automatic electronic road network generation methods based on trajectory information mainly include the following:

[0003] 1) Trajectory clustering methods represented by Kmeans: This method extracts cluster centers by clustering GPS observation points based on spatial proximity, which are used as vertices of the road network, and adds edges between vertices according to the order relationship of trajectory points to form the entire road network.

[0004] 2) KDE (Kernel Density Estimation) technology based on image processing; this technology converts trajectory data into two-dimensional raster images based on geographical location and uses a thinning method to extract the road skeleton.

[0005] 3) Incremental merging method: This method finds common segments in multiple trajectories based on spatial proximity, and then adds the trajectories to the empty map in sequence.

[0006] 4) Intersection Connection Method: This method utilizes the fact that the trajectories of intersections have different headings or speeds and change significantly in a short period of time to detect road intersections and effectively connect them using the trajectories.

[0007] In summary, existing methods for automatically generating electronic road networks based on trajectory information mainly utilize supervised learning methods based on deep learning frameworks to extract road networks. These methods are suitable for simple scenarios with high GPS trajectory sampling rates and small positioning errors. However, in urban environments, especially in road scenarios with dense buildings and roads, heterogeneous structures (such as overpasses and underground tunnels), and complex geographical structures, the collected GPS data contains significant errors due to the influence of obstructions such as tall buildings, overpasses, and trees. Furthermore, the trajectory point sequence under sparse sampling rates makes it difficult to reflect the accessibility of neighboring areas, thus hindering the accurate differentiation of adjacent parallel roads and generating a large number of redundant road segments. This often leads to the misidentification of adjacent roads as a single road, resulting in errors in the generation of the electronic road network. Summary of the Invention

[0008] This application provides a road network generation method, apparatus, device, and storage medium based on trajectory information, aiming to at least partially solve one of the aforementioned technical problems in the prior art.

[0009] To address the above problems, this application provides the following technical solution:

[0010] A road network generation method based on trajectory information includes:

[0011] A vehicle trajectory dataset is collected from an existing road network, and the dataset is divided into a fixed-path trajectory set and a free-path trajectory set according to vehicle type. The fixed-path trajectory set includes vehicle trajectory data with fixed driving paths, and the free-path trajectory set includes vehicle trajectory data without fixed driving paths.

[0012] Feature extraction is performed on the fixed path trajectory set and the free path trajectory set to obtain regional association features based on point cloud, regional connectivity features based on frequent point sequences, and global distribution density features based on the fact that the road centerline has the highest GPS distribution density.

[0013] The regional association features, regional connectivity features, and global distribution density features are dimensionality reduced and quantized before being input into the SVM model for parallel road identification.

[0014] The trajectory data of the identified parallel roads are input into the map generation model to generate an optimized road network.

[0015] The technical solution adopted in this application embodiment further includes: after dividing the vehicle trajectory dataset into a fixed path trajectory set and a free path trajectory set according to the vehicle type, it further includes:

[0016] The road centerline and road width are extracted from the fixed path trajectory set and the free path trajectory set. Based on the road centerline and road width, target areas that may have parallel roads are selected from the existing road network, and the target areas are divided into cells of a set size.

[0017] The technical solution adopted in this application embodiment further includes: extracting road centerlines and road widths from the fixed path trajectory set and the free path trajectory set; selecting target areas where parallel roads may exist from the existing road network based on the road centerlines and road widths; and dividing the target areas into cells of a set size, specifically:

[0018] Road centerlines can be extracted using trajectory clustering and incremental trajectory merging methods or the KDE method.

[0019] Based on the road centerline, the adjacent area is extracted and the adjacent area is divided into cells c of a set size;

[0020] The distribution density of trajectory points in each cell is statistically analyzed. Thresholds are dynamically set based on the distribution of trajectory points in different regions. A density histogram is formed using the distribution density of trajectory points in all cells.

[0021] The density histogram is convolved with a Gaussian distribution function, and the gradient of the density histogram is calculated by applying the Sobel operator using a dynamic adjustment strategy of cell density gradient to generate a vertical gradient map. Then, the target region R is extracted by non-maximum suppression algorithm based on the maximum value of the width of the road edge cell and the road length information and length threshold in the existing road network.

[0022] The technical solution adopted in this application embodiment also includes: assuming that all bus routes in the target area R are represented by a set L. b This means that for each bus route l∈L b And the set of trajectories (or trajectory segments) collected by the vehicle during operation as it passes through the target area R. The process of extracting features from the fixed path trajectory set and the free path trajectory set yields regional association features based on point clouds, regional connectivity features based on frequent point sequences, and global distribution density features based on the fact that the road centerline has the highest GPS distribution density. Specifically:

[0023] The method for extracting the regional association features is as follows: statistically analyze the trajectory distribution in each cell of the target region, and convert each trajectory tr∈TR into a path grid cell sequence. The number of trajectories passing through each cell is counted, and cells with fewer trajectories and support less than a set threshold are filtered out to obtain a set of cells C0 that meets the conditions. All cells in C0 pass through a certain number of bus route l trajectory points. The cell set C0 is aggregated according to spatial proximity to obtain one or more connected regions {C1, C2, ...} with proximity relationships. The connected regions {C1, C2, ...} are then grouped and merged based on direction information to obtain aggregated connected regions. The number of cells in each aggregated connected region is counted. If the number of cells is greater than a set threshold, the regional association feature of each cell in the connected region corresponding to bus route l is set to 1; otherwise, it is set to 0.

[0024] The method for extracting regional connectivity features is as follows: for each bus route l∈L passing through the target region R, bus and all trajectories The regional connectivity features of each bus route at the online level are extracted through two steps: cross-cell sequence distribution statistics and maximum connected subgraph extraction.

[0025] The global distribution density feature extraction method is as follows: a constrained Gaussian mixture model (CGMM) is used to fit the trajectory point distribution of each road, and the peak value of each Gaussian component is the position of the road centerline; the probability density function of the CGMM is:

[0026]

[0027] Where K g ω is the number of Gaussian components in CGMM. i and σ i These are the weights and standard deviations of each Gaussian component, μ. i Δμ represents the location of the road centerline, and Δμ represents the road width.

[0028] The entropy evaluation function is used to evaluate the CGMM modeling results with different numbers of Gaussian components. The optimal model is selected when the entropy evaluation function value is the smallest, indicating that the CGMM is optimal. This model is chosen as the modeling result for the floating car trajectory. The number of Gaussian components K in the CGMM represents the number of parallel roads. Based on this, the global distribution density feature F of c in each cell is generated. g , defined as the Gaussian component to which the cell belongs, if K g =1, global distribution density feature F of all cells g The same, if K g =2, based on its Gaussian component, its F g The value can be 0 or 1.

[0029] The technical solution adopted in this application embodiment further includes: the step of inputting the region association features, region connectivity features, and global distribution density features into the SVM model after dimensionality reduction and quantization for parallel road identification, specifically:

[0030] Similarity is measured for the regional association features, regional connectivity features, and global distribution density features of all cells; the similarity measurement algorithm includes: for each cell c in the target region R, calculating its similarity with each horizontally neighboring cell c′, the calculation formula is:

[0031]

[0032] Where δ(a) i ,b i ) is an indicator function, when a i and b i The value is 1 if the values ​​are the same, otherwise it is 0.

[0033] Then calculate the mean E of all similarities. c σ c and skewness s c , respectively, are used to represent the average similarity, the dispersion and asymmetry of similarity, and are represented by a low-dimensional vector; the low-dimensional vector is input into an SVM classification model suitable for small samples to identify and classify whether the target region contains parallel roads.

[0034] The technical solution adopted in this application embodiment further includes: after the dimensionality reduction and quantization of the regional association features, regional connectivity features, and global distribution density features are input into the SVM model for parallel road identification, it further includes:

[0035] For a target area with parallel roads, the trajectory data along the route is clustered; the clustering algorithm includes: for each trajectory data tr of the target area R and its route cell sequence. The similarity between two trajectory data is calculated using a similarity metric algorithm. Each trajectory data is treated as a separate cluster, and a global similarity matrix is ​​constructed to represent the similarity between any pair of trajectory data. Hierarchical clustering is used to merge clusters from bottom to top, and the average similarity of trajectory data within two clusters is calculated. This process is then iterated until only two trajectory clusters, TR1 and TR2, remain.

[0036] The technical solution adopted in this application embodiment further includes: inputting the trajectory data of the identified parallel roads into the map generation model to generate an optimized road network, specifically:

[0037] The map generation model includes a K-Means-based map generation model, a KDE-based map generation model, and an intersection-connection-based map generation model. The road network generation algorithm of the K-Means-based map generation model includes: for two trajectory clusters TR1 and TR2, running Kharita on both types of trajectory clusters within the sub-map range of the target region R, and denoting the other trajectory sets in the sub-map of R (excluding TR1) as... Let the set of trajectories other than TR2 in the submap to which R belongs be denoted as exist and The Kharita algorithm is run on each road to generate a vector map representing the road structure associated with one of the parallel roads. The two vector maps are overlaid, and the nearest centroids in the same direction are connected to output the optimized road network.

[0038] The road network generation algorithm of the KDE-based map generation model includes: compressing the vehicle trajectory dataset into a single two-dimensional density estimate, then iteratively performing a binary skeletonization operation on the two-dimensional density estimate to eliminate the edges between the two sets of trajectory clusters TR1 and TR2, connecting the target region R with the existing road network, smoothing the region edges, and outputting the optimized road network.

[0039] The road network generation algorithm based on the intersection connection map generation model includes: firstly, scanning all GPS trajectories and recording the output and input trajectory information of each intersection node to create an initial road network, connecting nodes with trajectory parts; then, introducing the parallel road determination results and the generated two sets of trajectory clusters TR1 and TR2, deleting the connections of intersection nodes connected by the trajectories in the parallel road determination and the two sets of trajectory clusters TR1 and TR2, and then processing false roads on the same road; for trajectory clusters S1 and S2 on different parallel roads in the target area R, in the sub-map trajectory data Tr={tr1,tr1,...,tr n The complement of} and The map generation algorithm is applied to each map, generating two vector maps respectively. Finally, the two vector maps are overlaid and the optimized road network is output.

[0040] Another technical solution adopted in this application embodiment is: a road network generation device based on trajectory information, comprising:

[0041] The trajectory classification module is used to collect vehicle trajectory datasets from existing road networks and divide the vehicle trajectory datasets into fixed-path trajectory sets and free-path trajectory sets according to vehicle type. The fixed-path trajectory set includes vehicle trajectory data with fixed driving paths, and the free-path trajectory set includes vehicle trajectory data without fixed driving paths.

[0042] Feature extraction module: used to extract features from the fixed path trajectory set and the free path trajectory set to obtain regional association features based on point cloud, regional connectivity features based on frequent point sequences, and global distribution density features based on the fact that the road centerline has the highest GPS distribution density.

[0043] Road recognition module: used to reduce and quantize the regional association features, regional connectivity features and global distribution density features, and then input them into the SVM model for parallel road recognition;

[0044] Road optimization module: This module is used to input the trajectory data of the identified parallel roads into the map generation model to generate an optimized road network.

[0045] Another technical solution adopted in this application embodiment is: a device, the device including a processor and a memory coupled to the processor, wherein,

[0046] The memory stores program instructions for implementing the road network generation method based on trajectory information;

[0047] The processor is used to execute the program instructions stored in the memory to control the road network generation method based on trajectory information.

[0048] Another technical solution adopted in this application embodiment is: a storage medium storing processor-executable program instructions, the program instructions being used to execute the road network generation method based on trajectory information.

[0049] Compared to existing technologies, the beneficial effects of the embodiments of this application are as follows: The road network generation method, apparatus, device, and storage medium based on trajectory information of the embodiments of this application address the problem that existing road generation methods have difficulty in segmenting adjacent parallel roads. By classifying trajectory data using vehicle type and combining trajectory categories, a set of regional accessibility association feature extraction methods for low sampling rate and high error scenarios is proposed. Different types of trajectory data are extracted at the point level, the line level, and the surface level, respectively, to extract regional association features, regional connectivity features, and global distribution density features. All features are then dimensionality-reduced and quantized before being input into the SVM model. The SVM model uses the distribution differences of different types of trajectory data on roads to identify parallel roads. The parallel road identification results are used as a plug-in to existing road generation methods to generate optimized road networks, which greatly improves the generation effect of complex roads such as main roads, auxiliary roads, and elevated roads. Attached Figure Description

[0050] Figure 1 This is a flowchart of a road network generation method based on trajectory information according to an embodiment of this application;

[0051] Figure 2 This is a schematic diagram of the GPS point cloud distribution of the trajectory type introduced in the embodiments of this application;

[0052] Figure 3 A schematic diagram illustrating an example of a Gaussian distribution of GPS trajectory points;

[0053] Figure 4 A schematic diagram of two connected regions formed by a real bus route A in a certain city;

[0054] Figure 5 A schematic diagram showing the merging of the connecting areas of the two bus routes;

[0055] Figure 6 Let K be the number of roads in a city, including single roads and parallel roads. g =1 and K g Diagram showing two cases where = 2;

[0056] Figure 7 This is a schematic diagram of the road network generation device based on trajectory information according to an embodiment of this application;

[0057] Figure 8 This is a schematic diagram of the device structure according to an embodiment of this application;

[0058] Figure 9 This is a schematic diagram of the structure of the storage medium according to an embodiment of this application. Detailed Implementation

[0059] The technical solutions of the embodiments of this application 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 this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0060] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0061] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0062] Specifically, please refer to Figure 1 This is a flowchart of a road network generation method based on trajectory information according to an embodiment of this application. The road network generation method based on trajectory information according to an embodiment of this application includes the following steps:

[0063] S10: Collect vehicle trajectory datasets from existing road networks and divide the vehicle trajectory datasets into fixed path trajectory sets and free path trajectory sets according to vehicle type;

[0064] In this step, the existing road network is generated using existing map generation models, including but not limited to K-Means-based models, KDE-based models, and intersection-connection-based models. The road network is represented by a directed graph, denoted as G =<V,E> In this context, vertex v (v∈V) represents an intersection or road endpoint (lat, lon), and each edge e (e∈E) consists of a sequence of start, end, and path points, labeled (u, v, L), where u∈V, v∈V, and L={p1, p2, …, p |L| A vehicle trajectory consists of a series of spatial points represented by a polyline. A single vehicle trajectory is defined as a sequence of trajectory points, denoted as tr = ... <p1,p2,...,p |Tr| > The trajectory point sequence is arranged in chronological order, i.e., p i .t <p i+ 1.t. In this embodiment, the vehicle trajectory dataset is divided into fixed-path trajectory sets based on vehicle type. and free path trajectory set There are two trajectory types: the fixed-path trajectory set includes trajectory data of vehicles with fixed routes, such as buses; the free-path trajectory set includes trajectory data of vehicles without fixed routes, such as taxis or private cars. Each free-path trajectory (tr...) t Each is represented by a quadruple (id, t, lon, lat), where id is the unique identifier of the vehicle, t is the timestamp, lon is the latitude, and lat is the longitude. Each fixed path trajectory (tr) b Each of these can be represented by a quintuple (id,lid,t,lon,lat). Similarly, id is the unique identifier of the vehicle, lid represents the bus route that distinguishes directions, t is the timestamp, lon represents the latitude, and lat represents the longitude.

[0065] Classifying vehicle trajectory data by trajectory type, the distribution differences of different trajectory types help to accurately identify adjacent parallel roads in the target area. In this embodiment, parallel roads refer to two adjacent roads whose distance is greater than a set threshold. Specifically, given two adjacent directional roads L = {x1, x2, ..., x...} composed of trajectory point sequences... N} and L′={x1′,x2′,…,x′ N}, where each x includes longitude lon and latitude lat information, if the distance between the centerlines of these two roads is less than a first set threshold τ d That is, dist(x) n ,x n ′)<τ d And the length is greater than the second set threshold τ l(In this embodiment, the distance is set to 1km, but the actual distance can be set according to the actual application scenario.) These two roads are then called parallel roads. Parallel roads typically include two situations: first, the two parallel roads are on the same horizontal plane; second, the two parallel roads are distributed vertically and may overlap in the vertical direction, such as viaducts or underpasses. Specifically, as shown... Figure 2 As shown in the figure, this is a schematic diagram of the GPS point cloud distribution of trajectory types introduced in this application embodiment. (a) shows the point cloud map of two adjacent roads without distinguishing trajectory types. After introducing trajectory types, fixed path trajectories and free path trajectories are represented by different colors, as shown in (b). It can be seen from the figure that there is a difference in the distribution of the two trajectory types on the two parallel roads, and this difference in distribution has regional continuity, which helps to separate different roads. (c) further introduces the fixed path trajectory of bus routes. Different colors represent different bus routes, which can clearly show the difference in the distribution of bus routes on different roads, providing important auxiliary information for the determination of parallel roads and the generation of road networks.

[0066] S20: Extract road centerlines and road widths from the fixed path trajectory set and the free path trajectory set. Based on the extracted road centerlines and road widths, select target areas from the existing road network that may contain parallel roads, and divide the target areas into cells of a set size.

[0067] In this step, the selection of the target region includes the extraction of the road centerline and width. The road centerline extraction methods include trajectory clustering and incremental trajectory merging, as well as the KDE method. The trajectory clustering and incremental trajectory merging method uses edges formed by connecting point sequences as the road centerline, outputting a directed graph containing both point and edge information. The KDE method first extracts a binarized pixel image and then combines it with a skeletonization algorithm to extract the road centerline.

[0068] The method for extracting road width is as follows: Based on the road centerline, the adjacent area is extracted and divided into cells c of a set size (e.g., 1 meter × 1 meter, the specific size can be set according to the actual application scenario) as the smallest processing unit; then, the trajectory point distribution density of each cell is calculated, and cells with a trajectory point distribution density greater than a set density threshold are selected and merged to form the target area. However, due to the differences in trajectory point distribution density in different areas, setting a fixed threshold is difficult to adapt to scenarios with uneven trajectory point distribution. For example, for roads with few vehicles, setting a uniform fixed threshold may mistakenly eliminate some road width information, resulting in incomplete extraction of the target area. Therefore, this embodiment dynamically sets the threshold according to the trajectory point distribution in different areas, uses the trajectory point distribution density of all cells within the target area to form a density histogram, and to reduce the impact of large GPS errors on the boundary, the formed density histogram is convolved with a Gaussian distribution function. Then, using a dynamic adjustment strategy for cell density gradient, the Sobe l operator is applied to calculate the gradient of the density histogram, generating a gradient map, i.e.:

[0069]

[0070] By analyzing the density distribution trend of trajectory points along the road cross-section (i.e., the direction of movement perpendicular to the trajectory on the road), a vertical gradient map of the target region is generated by convolving the density histogram with a vertical convolution kernel. Then, the target region, denoted as R, is extracted using a non-maximum suppression algorithm based on the maximum upper and lower ranges of the width of the road edge cells, as well as the road length information and length threshold in the existing road network.

[0071] S30: Extract complementary features of points, lines and surfaces from the target area to obtain regional association features based on point cloud, regional connectivity features based on frequent point sequences, and global distribution density features based on the fact that the road centerline has the highest GPS distribution density.

[0072] In this step, the regional association features F at the point level are used. p This refers to the regional association features based on the differences in the trajectory category distribution of adjacent cells in discrete point cloud data, and the regional connectivity features F based on the line level. r This refers to the regional connectivity features of the maximum generating graph based on frequent point sequences across cells, and the global distribution density features F based on the surface level. gThis is based on the fact that the centerline of a road has the highest global distribution density feature in GPS. Due to large positioning errors, there are numerous location points belonging to both parallel roads in the non-road area between them. Furthermore, due to the low trajectory sampling rate, it is difficult to obtain road connectivity in adjacent areas through trajectory point sequences, making it difficult to distinguish between the two parallel roads. If parallel roads exist in the target area, the trajectory points of the same bus route will differ in the areas where the two parallel roads are located; this difference helps to distinguish between the two parallel roads.

[0073] In this embodiment, the method for extracting regional association features is as follows: First, the trajectory category distribution features of each cell in the target region are extracted. Then, neighboring cells with similar features are grouped together, and the same cell neighboring features are set for this group. Specifically, in this embodiment, the cells above, below, to the left, right, and four diagonal cells of the target cell are defined as neighboring units, meaning each target cell has at most 8 neighboring cells and at least 3 neighboring cells. For example... Figure 3 The diagram shows an example of a Gaussian distribution of GPS trajectory points, where cells c1 and c2 are neighboring, while c3, c4, and c5 are non-neighboring. All bus routes in the target area R are represented by the set L. b This means that for each bus route l∈L b And the set of trajectories (or trajectory segments) collected by the vehicle during operation as it passes through the target area R. Extract region association features using the following steps:

[0074] (1) Trajectory point distribution statistics; First, the trajectory distribution in each cell is statistically analyzed, and each trajectory tr∈TR is converted into a path grid cell sequence. And count the number of trajectories passing through each cell; for cell c in region R, use N c This represents the number of bus trajectories passing through cell c in the trajectory set TR. Considering the uneven distribution of trajectories, support is used to standardize them. This represents the ratio of trajectory l passing through cell c to all trajectories of this route passing through the target region R. Then, to reduce noise, cells with too few trajectories and too little support are filtered out, i.e., N. c ≥N min And Sup c ≥Sup min This yields a set of cells C0 that meet the conditions, where each cell in C0 passes through a certain number of bus route l trajectory points. Note that GPS density varies across different types of roads in the city, N... min and Sup min It can be dynamically set according to the cell density in the target region R, based on the 95% confidence interval.

[0075] (2) Neighboring Cell Aggregation and Grouping: Aggregate the cell set C0 according to spatial proximity to obtain one or more connected regions {C1, C2, ...} with proximity relationships. The aggregation method is as follows: First, starting from any cell c ∈ C0, remove c from C0 and put it into a new set C1; second, find the neighboring cells of the cells in C1 from C0, remove them from C0 and put them into C1 together, and continue to expand until no neighboring cells are added to the cells in C1. Then, select the next cell from C0, put it into a new set C2, and repeat the above steps until C0 is empty, thus obtaining one or more connected regions {C1, C2, ...} with proximity relationships. (Details follow...) Figure 4 The diagram shows two connected areas formed by a real bus route A in a certain city. It can be seen that the two connected areas belong to the same road, and GPS is lost due to reasons such as signal loss.

[0076] (3) Grouping and merging connected regions {C1, C2, ...} based on direction information; due to uneven data spatial distribution, coefficients, etc., a large number of connected regions may be formed. These connected regions are similar and adjacent in direction, and are likely to belong to the same road. In order to find larger connected regions, for any two connected regions, if the angle between the average directions of the trajectory points does not exceed a set angle threshold (e.g., 10 degrees) and the distance is less than a set distance threshold, they are merged to obtain an aggregated connected region. Each connected region includes a certain number of cells. Specifically, as follows... Figure 5 The diagram shows a merged connected region diagram of two bus routes. Then, the number of cells in each merged connected region is counted. If the number of cells exceeds a set threshold, the regional association feature of each cell corresponding to bus route l in that connected region is set to 1; otherwise, it is set to 0. Therefore, assuming L... b K in b For each cell in the target region R, the point-level regional association features of a bus route are represented by a K. b A Boolean vector of dimension.

[0077] The method for extracting line-level regional connectivity features is as follows: Given a bus route with a fixed path, all its trajectory points follow the same roads. The trajectory point sequence reflects not only the connectivity of neighboring cells along the route but also the connectivity of non-neighboring cells. Concatenating multiple non-neighboring cell sequences helps to obtain the correlations at the line level of neighboring regions. Specifically, for each bus route l∈L passing through the target region R... bus and all trajectories The regional connectivity features of each bus route at the online level are extracted through two steps: cross-cell sequence distribution statistics and maximum connected subgraph extraction. The cross-cell sequence distribution statistics algorithm includes: firstly, converting each trajectory tr∈TR into a sequence of grid cells along its path. Then, count the number of every two consecutive cell pairs {c,c′} or the number of trajectories that pass consecutively through cells c and c′ in the cell sequence of all trajectories, denoted as Num. c,c′ This is used to reflect the probability that cell pairs belong to the same road. Considering the uneven distribution of trajectories in different areas, this application embodiment uses support to standardize it. This yields the quadruple (c, c′, Num) c,c′ ,Conf c,c′ The more trajectories a cell pair has, the greater the likelihood that the cell pair belongs to the same road. Due to noise and the differences in trajectory density collected on different types of roads in the city, a threshold Cn is set. min and Conf min Filter out cell pairs that have passed through a number of trajectories or whose confidence level is less than a set threshold.

[0078] The feature extraction algorithm based on the maximum connected subgraph is as follows: Based on the filtered cell pairs, all cells in the target region R are treated as nodes, and edges are added between the node pairs corresponding to the filtered cell pairs to obtain a directed graph G.<C,E> To extract a maximum connected graph from graph M that reflects the largest connected regions belonging to the same road in the target region R, a depth-first search method is used to extract the largest connected subgraph G′ in graph G. For each cell in G′, its regional connectivity feature is assigned a value of 1; otherwise, it is assigned a value of 0. Therefore, for the set of all bus routes L... b Each cell is extracted into a maximum connected graph, and Boolean features are obtained based on whether each cell belongs to this connected graph. Assume L... b The number of medium-sized bus routes is K. b Then, for the target region G, extract a k b The Boolean characteristic of dimensionality, denoted as F r .

[0079] The specific algorithm for extracting global distribution density features at the surface level is as follows: Figure 2 As shown, due to the influence of errors, GPS density typically decreases with increasing distance from the road centerline, consistent with a Gaussian distribution. Considering the presence of parallel roads in the area, a constrained Gaussian mixture model (CGMM) is used to fit the trajectory point distribution of each road. The peak value of each Gaussian component represents the location of the road centerline. The probability density function of the CGMM is:

[0080]

[0081] Where K g ω is the number of Gaussian components in CGMM, which reflects the potential number of parallel paths. i and σ i These are the weights and standard deviations of each Gaussian component, μ. i Δμ represents the location of the road centerline, and Δμ represents the road width.

[0082] Modeling all trajectories (buses and taxis) using the CGMM essentially involves solving for the parameters of the CGMM's probability density function. When the number of roads K... g When the parameters of a CGMM are known, the Expectation-Maximization (EM) algorithm can be used to predict them. However, in actual calculations, the existence and number of parallel roads are unknown. To calculate the optimal fit K... g The entropy evaluation function is used to evaluate the CGMM modeling results with different numbers of Gaussian components, and the optimal model is selected. When the entropy evaluation function value is the smallest, it indicates that the CGMM is optimal, and this model is selected as the modeling result for the floating car trajectory. The number of Gaussian components K in the CGMM is the number of parallel roads. Based on this, the global distribution density feature F of c for each cell is generated. g , is defined as the Gaussian component to which the cell belongs. If K g =1, global distribution density feature F of all cells g The same, if K g =2, based on its Gaussian component, its F g The value can be either 0 or 1. Specifically, as shown below... Figure 6 As shown, this represents the number of roads K in a city, including single roads and parallel roads. g =1 and K g Diagram showing two cases where the sum is 2.

[0083] S40: After dimensionality reduction and quantization of regional association features, regional connectivity features, and global distribution density features, the data are input into an SVM (Support Vector Machine) model suitable for small samples. The SVM model is then used to identify parallel roads in the target region.

[0084] In this step, due to the large number of cells and high feature dimensionality, but the limited number of parallel road samples, directly inputting the region association features, region connectivity features, and global distribution density features of all cells into the classification model for parallel road determination would lead to the curse of dimensionality. Therefore, this embodiment first performs similarity measurement on the region association features, region connectivity features, and global distribution density features of all cells to obtain statistically significant features, which are then represented by a low-dimensional vector (e.g., a distribution histogram) to achieve dimensionality reduction. Then, the generated low-dimensional vector is input into an SVM model suitable for small samples to determine whether parallel roads exist.

[0085] Specifically, the similarity measurement algorithm includes the following: if parallel roads exist in the target area, each cell should have more similar features to its horizontally adjacent cells, while its features should differ significantly from those of its vertically adjacent cells. Taking horizontally adjacent cells as an example, we define k as the number of cells whose directions are similar to its trajectory points. v The similarity of adjacent cells is such that each cell has at most 2×k similarities. v There are at least k horizontally adjacent cells. v There are several horizontal neighboring cells. Therefore, for each cell c in the target region R, its similarity to each horizontal neighboring cell c′ is first calculated. Since the feature value is of discrete Boolean type, the simple matching coefficient shown in formula (3) is used to calculate the similarity between each horizontal neighboring cell c′ and c, where δ(a i ,b i ) is an indicator function, when a i and b i The value is 1 if the values ​​are the same, otherwise it is 0.

[0086]

[0087] If there are 2×k v If there are 2×k horizontal neighboring cells, then 2×k will be generated. v Each similarity is calculated. Then, the mean E of all similarities is calculated separately. c σ c and skewness s c Three statistical features are used to represent average similarity, the dispersion of similarity, and asymmetry, respectively, and are represented by a low-dimensional vector (e.g., a distribution histogram) to achieve dimensionality reduction. Then, the low-dimensional vector is input into an SVM classification model suitable for small samples to classify whether the target region contains parallel roads. The SVM classification model uses a Gaussian radial basis function (RBF) as the kernel function.

[0088] S50: For target areas with parallel roads, cluster the trajectory data along the route into two sets of trajectory clusters, and input the two sets of clustered trajectory clusters into the existing map generation model to generate an optimized road network.

[0089] In this step, for each trajectory data point (including buses and taxis) in the target area R, the trajectory is described using the comprehensive distribution of trajectory point categories in the pathway cells. This is given by each trajectory data point tr and its pathway cell sequence. Each cell has a Boolean feature vector F, which quantifies the features of the trajectory using a Boolean feature vector of the same length. The corresponding feature value in the vector is the median of this feature in all cell sequences, similar to a majority vote. The similarity between two trajectory data is calculated using the matching coefficient shown in formula (3). Each trajectory data is regarded as a separate cluster, and a global similarity matrix is ​​constructed to represent the similarity between any pair of trajectory data. Then, a hierarchical clustering method is used to merge clusters from bottom to top. First, the average similarity of trajectory data within two clusters is calculated, and then the process is iterated until only two trajectory clusters, TR1 and TR2, remain. Since trajectory data belonging to the same bus route can only belong to the same category, for the classification of trajectory data for the same route, this embodiment of the application determines the category of all trajectory data belonging to the same route by post-processing the clustering results and sampling the majority vote method.

[0090] Furthermore, taking existing map generation models based on K-Means, KDE, and intersection connections as examples, and based on the parallel road identification results of the embodiments of this application, the road network optimization methods of the above three map generation models will be described in detail below.

[0091] The K-means-based map generation model: For two trajectory clusters TR1 and TR2 obtained in the target region R, Kharita is run on the two types of trajectory clusters within the sub-map range to which R belongs. The set of other trajectories in the sub-map to which R belongs, excluding TR1, is denoted as... Similarly, the set of other trajectories in the submap to which R belongs, excluding TR2, is denoted as... exist and The standard Kharita algorithm is run on each of the two maps to generate the road structure associated with one of the parallel roads. Since the map generated by the Kharita algorithm is a vector map, the two vector maps can be overlaid, and the two nearest centroids in the same direction can be connected. This reduces erroneous roads and road breaks caused when the target area R is connected to the edge of the map generated by the original algorithm, and outputs an optimized map road network.

[0092] The KDE-based map generation model first compresses the complete vehicle trajectory dataset into a single two-dimensional density estimate, then iteratively performs binary skeletonization on each integer density level. Parallel roads and edges between two trajectory clusters TR1 and TR2 are considered false roads, requiring the removal of edges between different trajectory clusters to reduce pseudo-segments between parallel roads. Wherein, S... iTo determine parallel roads and the set of trajectory clusters on different parallel roads, edges between different trajectory clusters are deleted, i.e., redundant road segments between parallel roads are removed. Then, the regenerated target region R is connected to the roads generated by the original algorithm, and the edges of the region are smoothed to handle the edges of the road region R.

[0093] Map Generation Model Based on Intersection Connections: This embodiment improves upon the urban road generation algorithm proposed by Karagiorgou. Karagiorgou identifies vehicle turning positions in GPS trajectories and aggregates turning points, assuming all turning point clusters represent road intersections, forming the vertices of the road network. After obtaining isolated intersection nodes, they are connected. First, all GPS trajectories are scanned, and the output and input trajectory information of each intersection node is recorded, thereby creating an initial road network and connecting nodes (intersections) with trajectory portions. At this stage, this embodiment introduces parallel road determination and two sets of trajectory clusters TR1 and TR2. The output and input trajectory information of each intersection node has been recorded, and the connections between intersection nodes connected by the trajectories in the parallel road determination and the two sets of trajectory clusters TR1 and TR2 are deleted to remove redundant roads between parallel roads. Then, false roads on the same road are processed. Specifically, by identifying trajectory portions close to existing connections, their geometry is merged into the geometry of existing connections. At this time, no new connections are introduced; only the geometry of existing connections is adjusted. Then, for trajectory clusters S1 and S2 on different parallel roads in the target area R, in the sub-map trajectory data Tr={tr1,tr2,...,tr3} n The complement of} and The map generation algorithm is applied to each map, generating two vector maps respectively. Finally, the two vector maps are overlaid and the optimized road network is output.

[0094] Based on the above, the road network generation method based on trajectory information in this application addresses the problem that existing road generation methods struggle to segment adjacent parallel roads. It classifies trajectory data by vehicle type and proposes a method for extracting regional accessibility association features in scenarios with low sampling rates and high errors, combining these trajectory categories. This method extracts regional association features at the point level, regional connectivity features at the line level, and global distribution density features at the surface level for different categories of trajectory data. All features are then dimensionality-reduced and quantized before being input into an SVM model. The SVM model uses the distribution differences of different types of trajectory data on roads to identify parallel roads. The parallel road identification results are used as a plugin for existing road generation methods to generate optimized road networks, significantly improving the generation effect of complex roads such as main roads, auxiliary roads, and elevated roads.

[0095] Please see Figure 7This is a schematic diagram of the road network generation device based on trajectory information according to an embodiment of this application. The road network generation method device 40 based on trajectory information according to an embodiment of this application includes:

[0096] Trajectory classification module 41: used to collect vehicle trajectory datasets from existing road networks and divide the vehicle trajectory datasets into fixed-path trajectory sets and free-path trajectory sets according to vehicle type; wherein, the fixed-path trajectory set includes vehicle trajectory data with fixed driving paths, and the free-path trajectory set includes vehicle trajectory data without fixed driving paths.

[0097] Feature extraction module 42: used to extract features from the fixed path trajectory set and the free path trajectory set to obtain regional association features based on point cloud, regional connectivity features based on frequent point sequences, and global distribution density features based on the fact that the road centerline has the highest GPS distribution density.

[0098] Road recognition module 43: used to input the dimensionality reduction and quantization of the regional association features, regional connectivity features and global distribution density features into the SVM model for parallel road recognition;

[0099] Road optimization module 44: This module is used to input the trajectory data of the identified parallel roads into the map generation model to generate an optimized road network.

[0100] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0101] The apparatus provided in this application can be applied to the foregoing method embodiments. For details, please refer to the description of the above method embodiments, which will not be repeated here.

[0102] Please see Figure 8 This is a schematic diagram of the device structure according to an embodiment of this application. The device 50 includes:

[0103] Memory 51 storing executable program instructions;

[0104] Processor 52 connected to memory 51;

[0105] The processor 52 is used to call the executable program instructions stored in the memory 51 and perform the following steps: Collect vehicle trajectory datasets from an existing road network, and divide the vehicle trajectory datasets into fixed-path trajectory sets and free-path trajectory sets according to vehicle type; wherein, the fixed-path trajectory set includes vehicle trajectory data with fixed driving paths, and the free-path trajectory set includes vehicle trajectory data without fixed driving paths; extract features from the fixed-path trajectory set and the free-path trajectory set to obtain regional association features based on point clouds, regional connectivity features based on frequent point sequences, and global distribution density features based on the fact that the road centerline has the highest GPS distribution density; after dimensionality reduction and quantization, the regional association features, regional connectivity features, and global distribution density features are input into an SVM model for parallel road identification; input the trajectory data of the identified parallel roads into a map generation model to generate an optimized road network.

[0106] The processor 52 can also be referred to as a CPU (Central Processing Unit). The processor 52 may be an integrated circuit chip with signal processing capabilities. The processor 52 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor.

[0107] Please see Figure 9This is a schematic diagram of the storage medium in an embodiment of this application. The storage medium in this embodiment stores program instructions 61 capable of implementing the following steps: collecting vehicle trajectory datasets from an existing road network, and dividing the vehicle trajectory datasets into a fixed-path trajectory set and a free-path trajectory set according to vehicle type; wherein the fixed-path trajectory set includes vehicle trajectory data with fixed driving paths, and the free-path trajectory set includes vehicle trajectory data without fixed driving paths; extracting features from the fixed-path trajectory set and the free-path trajectory set to obtain regional association features based on point clouds, regional connectivity features based on frequent point sequences, and global distribution density features based on the road centerline being the area with the highest GPS distribution density; quantizing and reducing the dimensionality of the regional association features, regional connectivity features, and global distribution density features, and then inputting them into an SVM model for parallel road identification; inputting the trajectory data traversed by the identified parallel roads into a map generation model to generate an optimized road network. This program instruction 61 can be stored in the aforementioned storage medium in the form of a software product, including several instructions to cause a device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods in various embodiments of this application. The aforementioned storage media include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program instructions, or terminal devices such as computers, servers, mobile phones, and tablets. Servers can be independent servers or cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, apparatuses, or units, and may be electrical, mechanical, or other forms.

[0109] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A road network generation method based on trajectory information, characterized in that, include: A vehicle trajectory dataset is collected from an existing road network, and the dataset is divided into a fixed-path trajectory set and a free-path trajectory set according to vehicle type. The fixed-path trajectory set includes vehicle trajectory data with fixed driving paths, and the free-path trajectory set includes vehicle trajectory data without fixed driving paths. Feature extraction is performed on the fixed path trajectory set and the free path trajectory set to obtain regional association features based on point cloud, regional connectivity features based on frequent point sequences, and global distribution density features based on the fact that the road centerline has the highest GPS distribution density. The regional association features, regional connectivity features, and global distribution density features are dimensionality reduced and quantized before being input into the SVM model for parallel road identification. The trajectory data of the identified parallel roads are input into the map generation model to generate an optimized road network.

2. The road network generation method based on trajectory information according to claim 1, characterized in that, After dividing the vehicle trajectory dataset into a fixed-path trajectory set and a free-path trajectory set according to vehicle type, the method further includes: The road centerline and road width are extracted from the fixed path trajectory set and the free path trajectory set. Based on the road centerline and road width, target areas that may have parallel roads are selected from the existing road network, and the target areas are divided into cells of a set size.

3. The road network generation method based on trajectory information according to claim 2, characterized in that, The process involves extracting road centerlines and road widths from the fixed path trajectory set and the free path trajectory set. Based on the road centerlines and road widths, target areas where parallel roads may exist are selected from the existing road network, and these target areas are divided into cells of a predetermined size. Specifically: Road centerlines can be extracted using trajectory clustering and incremental trajectory merging methods or the KDE method. Based on the road centerline, the adjacent area is extracted and the adjacent area is divided into cells c of a set size; The distribution density of trajectory points in each cell is statistically analyzed. Thresholds are dynamically set based on the distribution of trajectory points in different regions. A density histogram is formed using the distribution density of trajectory points in all cells. The density histogram is convolved with a Gaussian distribution function, and the gradient of the density histogram is calculated by applying the Sobel operator using a dynamic adjustment strategy of cell density gradient to generate a vertical gradient map. Then, the target region R is extracted by non-maximum suppression algorithm based on the maximum value of the width of the road edge cell and the road length information and length threshold in the existing road network.

4. The road network generation method based on trajectory information according to claim 3, characterized in that, Assume that all bus routes in the target region R are represented by the set L. b This means that for each bus route l∈L b And the set of trajectories (or trajectory segments) collected by the vehicle during operation as it passes through the target area R. The process of extracting features from the fixed path trajectory set and the free path trajectory set yields regional association features based on point clouds, regional connectivity features based on frequent point sequences, and global distribution density features based on the fact that the road centerline has the highest GPS distribution density. Specifically: The method for extracting the regional association features is as follows: statistically analyze the trajectory distribution in each cell of the target region, and convert each trajectory tr∈TR into a path grid cell sequence. The number of trajectories passing through each cell is counted, and cells with a number of trajectories and a support score less than a set threshold are filtered out to obtain a set of cells C0 that meets the conditions. All cells in C0 have passed through a certain number of bus trajectory points of route l. The cell set C0 is aggregated according to spatial proximity to obtain one or more connected regions {C1,C2,...} with proximity. The connected regions {C1,C2,...} are then grouped and merged based on direction information to obtain the aggregated connected regions. Count the number of cells in each aggregated connected region. If the number of cells is greater than a set threshold, set the regional association feature of bus line l corresponding to each cell in the connected region to 1; otherwise, set it to 0. The method for extracting regional connectivity features is as follows: for each bus route l∈L passing through the target region R, bus and all trajectories The regional connectivity features of each bus route at the online level are extracted through two steps: cross-cell sequence distribution statistics and maximum connected subgraph extraction. The global distribution density feature extraction method is as follows: a constrained Gaussian mixture model (CGMM) is used to fit the trajectory point distribution of each road, and the peak value of each Gaussian component is the position of the road centerline; the probability density function of the CGMM is: where K g is the number of Gaussian components in CGMM, ω i and σ i are the weight and standard deviation of each Gaussian component, respectively, μ i is the position of the road centerline, and Δμ is the road width; The CGMM modeling results of different Gaussian component numbers are evaluated using an entropy evaluation function, and the optimal model is selected, when the entropy evaluation function value is the smallest, it indicates that the CGMM is optimal, the model is selected as the modeling result of the probe vehicle trajectory, and the Gaussian component number K of the CGMM is the number of parallel roads; on this basis, the global distribution density feature F of each cell c is generated g , which is defined as the Gaussian component to which the cell belongs, if K g = 1, the global distribution density features F g of all cells are the same, if K g = 2, according to the Gaussian component to which it belongs, the F g value is 0 or 1.

5. The road network generation method based on trajectory information according to claim 4, characterized in that, The process of dimensionality reduction and quantization of the regional association features, regional connectivity features, and global distribution density features, followed by inputting them into the SVM model for parallel road identification, is as follows: Perform similarity measurements on the regional association features, regional connectivity features, and global distribution density features of all cells; The similarity measurement algorithm includes: for each cell c in the target region R, calculating its similarity with each horizontally neighboring cell c′, using the following formula: Where δ(a) i ,b i The function ) is an indicator function; it returns 1 when ai and bi have the same value, and 0 otherwise. Then calculate the mean E of all similarities. c σ c and skewness s c , respectively, are used to represent the average similarity, the dispersion and asymmetry of similarity, and are represented by a low-dimensional vector; the low-dimensional vector is input into an SVM classification model suitable for small samples to identify and classify whether the target region contains parallel roads.

6. The road network generation method based on trajectory information according to any one of claims 1 to 5, characterized in that, After dimensionality reduction and quantization of the regional association features, regional connectivity features, and global distribution density features, and then inputting them into the SVM model for parallel road identification, the method further includes: For a target area with parallel roads, the trajectory data along the route is clustered; the clustering algorithm includes: for each trajectory data tr of the target area R and its route cell sequence. The similarity between two trajectory data is calculated using a similarity metric algorithm. Each trajectory data is treated as a separate cluster, and a global similarity matrix is ​​constructed to represent the similarity between any pair of trajectory data. Hierarchical clustering is used to merge clusters from bottom to top, and the average similarity of trajectory data within two clusters is calculated. This process is then iterated until only two trajectory clusters, TR1 and TR2, remain.

7. The road network generation method based on trajectory information according to claim 6, characterized in that, The process of inputting the trajectory data of the identified parallel roads into the map generation model to generate an optimized road network specifically involves: The map generation model includes a K-Means-based map generation model, a KDE-based map generation model, and an intersection-connection-based map generation model. The road network generation algorithm of the K-Means-based map generation model includes: for two trajectory clusters TR1 and TR2, running Kharita on both types of trajectory clusters within the sub-map range of the target region R, and denoting the other trajectory sets in the sub-map of R (excluding TR1) as... Let the set of trajectories other than TR2 in the submap to which R belongs be denoted as exist and The Kharita algorithm is run on each road to generate a vector map representing the road structure associated with one of the parallel roads. The two vector maps are overlaid, and the nearest centroids in the same direction are connected to output the optimized road network. The road network generation algorithm of the KDE-based map generation model includes: compressing the vehicle trajectory dataset into a single two-dimensional density estimate, then iteratively performing a binary skeletonization operation on the two-dimensional density estimate to eliminate the edges between the two sets of trajectory clusters TR1 and TR2, connecting the target region R with the existing road network, smoothing the region edges, and outputting the optimized road network. The road network generation algorithm based on the intersection connection map generation model includes: firstly, scanning all GPS trajectories and recording the output and input trajectory information of each intersection node to create an initial road network, connecting nodes with trajectory parts; then, introducing the parallel road determination results and the generated two sets of trajectory clusters TR1 and TR2, deleting the connections of intersection nodes connected by the trajectories in the parallel road determination and the two sets of trajectory clusters TR1 and TR2, and then processing false roads on the same road; for trajectory clusters S1 and S2 on different parallel roads in the target area R, in the sub-map trajectory data Tr={tr1,tr1,...,tr n The complement of} and The map generation algorithm is applied to each map, generating two vector maps respectively. Finally, the two vector maps are overlaid and the optimized road network is output.

8. A road network generation device based on trajectory information, characterized in that, include: The trajectory classification module is used to collect vehicle trajectory datasets from existing road networks and divide the vehicle trajectory datasets into fixed-path trajectory sets and free-path trajectory sets according to vehicle type. The fixed-path trajectory set includes vehicle trajectory data with fixed driving paths, and the free-path trajectory set includes vehicle trajectory data without fixed driving paths. Feature extraction module: used to extract features from the fixed path trajectory set and the free path trajectory set to obtain regional association features based on point cloud, regional connectivity features based on frequent point sequences, and global distribution density features based on the fact that the road centerline has the highest GPS distribution density. Road recognition module: used to reduce and quantize the regional association features, regional connectivity features and global distribution density features, and then input them into the SVM model for parallel road recognition; Road optimization module: This module is used to input the trajectory data of the identified parallel roads into the map generation model to generate an optimized road network.

9. A device, characterized in that, The device includes a processor and a memory coupled to the processor, wherein, The memory stores program instructions for implementing the road network generation method based on trajectory information as described in any one of claims 1-7; The processor is used to execute the program instructions stored in the memory to control the road network generation method based on trajectory information.

10. A storage medium, characterized in that, The system stores processor-executable program instructions for performing the road network generation method based on trajectory information as described in any one of claims 1 to 7.

Images