A community refined road network extraction method, device and storage medium

Abstract

The application relates to the field of road extraction, and discloses a community fine road network extraction method, equipment and a storage medium. The method comprises the following steps: data acquisition: acquiring a crowdsourcing trajectory data source; initial trajectory flow set construction: constructing an initial trajectory flow set according to the crowdsourcing trajectory data source; trajectory flow screening: determining the confidence level of the trajectory flow set; establishing a speed confidence level hierarchical setting according to the speed value segmentation of the trajectory flow, and extracting a high-confidence trajectory flow set; identifying a trajectory flow set mode: classifying the trajectory flow set into a road mode and a free traffic mode by setting a pedestrian group correlation index threshold; and extracting community road network information: analyzing the shape of the trajectory flow set classification results in different modes, extracting community linear road elements and planar road area elements, and organizing the community road segments and free traffic area elements. The application has the beneficial effect that the structure and connection relationship of the road network can be more accurately depicted and are not affected by individual trajectory data.

Description

Technical Field

[0001] This invention relates to the field of road extraction, and more particularly to a method, device, and storage medium for extracting refined road networks in communities. Background Technology

[0002] Urban navigation is a fundamental cognitive task in people's daily lives, and its accuracy and completeness are crucial to the travel or navigation experience. During navigation, large- and medium-scale urban road network data facilities provide reliable navigation information support. However, at the micro-map scale, there are still significant gaps in navigation map data and services for the "last mile" of navigation activities—the urban community space—which has become a prominent and urgent weak link in seamless urban navigation.

[0003] Many mapping companies, such as Google, Here, TomTom, Baidu, and Gaode, have built detailed MVR networks, but the micro-scale road networks in communities used for navigation services have not yet been adequately considered. Currently, urban community roads in China are not mapped, and the existing maps do not accurately reflect the actual community navigation scenarios. This results in a severe deficiency in navigation applications and services within urban community spaces, as well as confusion regarding navigation in micro-space scenarios. Consequently, pedestrian navigation in community spaces is significantly more difficult than in spaces outside the community, hindering the in-depth development of community map applications.

[0004] Existing road extraction methods face the following challenges when acquiring community-level roads: (1) The environmental characteristics of the community space (such as dense high-rise buildings and trees) cause GPS trajectory interference, which makes it impossible for GPS trajectories to have noise or drift during sampling, exacerbating the difficulty of generating potential paths. These trajectory drifts and noise interferences result in a large number of unnecessary overestimations of the curves in the final generated roads.

[0005] (2) The micro-scale characteristics of community spaces (short road lengths, close road spacing, and dense arrangement) make existing methods unable to handle the sparsity and non-homogeneity of trajectory data caused by low-frequency sampling. For example, when dealing with low-frequency trajectory data sources at the community scale, the differences in the motion characteristics of pedestrian trajectories at road segments and intersections are greatly reduced due to low-frequency sampling, leading to errors in the identification of road segments and intersections. Furthermore, relying solely on individual trajectories is insufficient to characterize the intermediate clues of small-path features in the community, exacerbating the difficulty in capturing the motion characteristics of crowds under low-frequency sampling, all of which make the task of generating road networks at the community level even more challenging.

[0006] (3) The pedestrian freedom of movement characteristic of community spaces, i.e., there are usually a large number of non-road scenes within a community where pedestrians can freely pass and linger, increases the complexity of the community road extraction task. At the methodological level, existing methods fail to specifically distinguish between roads and other pedestrian-accessible areas in a community, thus easily generating roads that do not actually exist in non-road scenes. This undoubtedly increases the difficulty of generating an accurate community-level road network. Summary of the Invention

[0007] The purpose of this invention is to propose a method, device, and storage medium for refining community road network extraction, thereby solving the technical problems of complex extraction processes and inaccurate results in existing community road network extraction methods.

[0008] This invention provides a method for extracting refined road networks in communities, comprising the following steps: S1. Data Acquisition: Acquire crowdsourced trajectory data sources; S2. Initial trajectory stream set construction: Construct an initial trajectory stream set based on the crowdsourced trajectory data source; S3. Trajectory Flow Filtering: Determine the confidence level of the trajectory flow set; establish a speed confidence level setting based on the speed value of the trajectory flow segment, and extract the trajectory flow set with high confidence; S4. Identify trajectory flow patterns: By setting thresholds for pedestrian group correlation indicators, trajectory flow patterns are classified into road patterns and free-traffic patterns. S5. Extract community road network information: Analyze the morphology of trajectory flow set classification results under different modes, extract linear road elements and area road area elements in the community, and organize them into community road segments and free passage area elements. A storage medium storing instructions and data for implementing a method for refining community road network extraction.

[0009] A community-level refined road network extraction device includes: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement a community-level refined road network extraction method.

[0010] The beneficial effects provided by this invention are: 1. This invention establishes a multi-level confidence relaxation constraint strategy to flexibly handle the interference of trajectory noise. By constraining pedestrian trajectory data with high-confidence partitioning, the interference of low-confidence trajectory noise is suppressed, enabling trajectory flow clustering to highlight the set of potential road objects with significant linear characteristics. Low-confidence level data relaxation is performed based on the cluster range to more accurately quantify and calculate pattern classification features. By modulating the input trajectory data source, the trajectory data is layered to adapt to tasks at different semantic scales, making the distinction between two different types of pedestrian trajectories more accurate and achieving an optimal balance between accuracy and coverage. 2. This invention introduces pedestrian trajectory flow semantics for road network generation to overcome the semantic ambiguity problem of low-frequency trajectory data. By enhancing the representation of pedestrian motion features through trajectory flow, it can more accurately depict the structure and connectivity of the road network than traditional methods, and is not affected by individual trajectory data. Furthermore, it can better adapt to the complexity brought about by the random movement of pedestrians in special scenarios at the community scale. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of refined linear road extraction in a community according to an embodiment of the present invention; Figure 3 This is a schematic diagram of refined planar road extraction in a community according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware device of the present invention. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0013] Before formally describing the present invention, a general description of the solution of the present invention will be given first to facilitate understanding.

[0014] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating the method of the present invention; This invention provides a method for extracting refined road networks in communities, comprising the following steps: S1. Data Acquisition: Acquire crowdsourced trajectory data sources; It should be noted that the crowdsourced trajectory data source consists of pedestrian trajectory data, including: pedestrian ID and pedestrian position coordinate information (x, y) at different times. The pedestrian trajectory data is represented as: pti (ID, x, y, t), where t represents the time corresponding to the i-th record.

[0015] As one embodiment, the crowdsourced trajectory data source uses GPS trajectory data of different users collected via mobile phones.

[0016] S2, Construction of initial trajectory flow set: constructing an initial trajectory flow set according to the crowdsourced trajectory data source; It should be noted that in step S2, pedestrian trajectory data is converted into trajectory flow records, and trajectory flow records with abnormal time intervals are eliminated to form the initial trajectory flow set.

[0017] As one embodiment, step S2 is specifically: S21: extracting all pedestrian ID sets from pedestrian trajectories, and performing deduplication; S22: for each pedestrian ID, obtaining the time and position coordinate information of all trajectory points belonging to the ID, and organizing the information into the pedestrian trajectory tr = {pt1, pt2, …, ptn} of the ID, wherein pti(1 ≤ i<n) represents the spatio-temporal coordinate of the i-th trajectory point, and all trajectory points in tr are arranged in time sequence to obtain tr ’ = {pt1, pt2,…, ptn}, wherein when i<j, pti (t) ≤ ptj (t); S23: taking out tr ’ all adjacent trajectory point pairs {<pti, pti+1>| 1 ≤ i<n} therein, wherein the spatio-temporal coordinate of the former trajectory point is pti, and the spatio-temporal coordinate of the latter trajectory point is pti+1, and obtaining all trajectory flows {f (pti,pti+1) | 1 ≤ i<n} of the ID; S24: repeating the operations of S22 and S23 until all pedestrian IDs are processed, and the construction of the initial trajectory flow set is completed.

[0018] S3, Trajectory flow filtering: determining the confidence level hierarchy of the trajectory flow set; establishing speed confidence level hierarchical settings in sections according to the speed values of trajectory flows, and extracting a high-confidence trajectory flow set; As one embodiment, step S3 is specifically: S31: establishing a speed confidence calculation formula, and normalizing the confidence value to the range of 0 to 1; wherein, in step S31, the speed confidence calculation formula is as follows:

[0019] wherein v represents the speed of the trajectory flow, and the values of speed max are determined according to the concentrated interval of the moving speed of people in the community.

[0020] S32: Calculate the velocity value of the trajectory flow, substitute it into the velocity confidence formula to obtain the set of velocity confidence values ​​for all trajectory flows; S33: Extraction speed confidence value Greater than The trajectory flow, as a high-confidence trajectory flow, in which The set speed confidence threshold.

[0021] S4. Identify trajectory flow patterns: By setting thresholds for pedestrian group correlation indicators, trajectory flow patterns are classified into road patterns and free-traffic patterns. As one embodiment, step S4 specifically includes: S41: Calculate the spatial geometric similarity of high-confidence trajectory flows, and perform trajectory flow clustering and merging based on the similarity. Use the KNN clustering method to spatially cluster the trajectory flows, grouping sets of high-confidence trajectories that are geographically close and have similar geometric features into the same trajectory cluster; Specifically, the trajectory flow is abstracted into ordinary line segments, and the trajectory flow distance is the shortest Euclidean distance between the centers of the line segments; the K nearest neighbor (KNN) method is used to select the flow with adjacent distances, and the similarity is measured according to their parallelism and length. Transform any two flows belonging to the same neighborhood to the same origin O, where the target point Vj must be within a short distance of Vi. Identify similar flows whose target points lie within a boundary circle. To ensure symmetry, the circle is centered on the target point of the longer flow. Since using the same threshold r for flows of different lengths will lead to incorrect judgments, r needs to be dynamically adjusted according to the flow length. By setting the threshold r, we can obtain the dissimilarity sd between a group of trajectory flows f1 and f2 belonging to different trajectory clusters. 12 The dissimilarity determines whether the clusters of f1 and f2 can be merged; S42: Select a certain number of trajectory flows that are spatially clustered and group them into a group, ensuring that the trajectory flows in this group are concentrated in the same section of community road space or place space; Specifically, confidence relaxation is performed on the trajectory set using buffers. This process requires expanding the set of each trajectory flow with a radius of R. If the expanded buffers are adjacent to each other, they are merged, and the low-confidence trajectory sets contained in the merged buffers are merged into the same group. S43: Calculate the similarity of trajectory behavior among different individuals within the group, only consider the top K individuals closest to the target individual i, and assume that these K individuals are located in the neighborhood of i. By setting the group correlation threshold γ, divide them into two types of trajectory patterns. Specifically, the velocity similarity between individuals within a neighborhood is calculated using the following expression:

[0022] Where C(i,j) represents the velocity correlation between individuals i and j in the neighborhood; Construct a similarity matrix W using the speed correlation between different individuals, and calculate the group similarity Φ. The expression for calculating group similarity is:

[0023] Where z is the regularization coefficient, used to converge the final result; Set z to 0.5 / K to make the range of Φ converge to between 0 and 1; by setting the group correlation threshold γ, the two different types of trajectory sets are distinguished. The trajectory flow set with group similarity greater than γ is classified as the road mode, and the rest of the trajectory flow set is classified as the free passage mode.

[0024] S5. Extract community road network information: Analyze the morphology of trajectory flow set classification results under different modes, extract linear road elements and area road area elements in the community, and organize them into community road segments and free passage area elements. As one embodiment, step S5 specifically includes: Based on the classification results of step S4, the community linear roads are generated using the pedestrian trajectory flow set of the road pattern according to steps S51 to S31, and the community surface roads are generated using the trajectory flow set of the free passage pattern according to step S53. S51: Generate a grayscale image with density features using the sparsity and density of the trajectory stream. Process the GPS dataset into trajectory streams and perform preliminary outlier filtering on these trajectory streams based on building footprint data. Remove unreasonable trajectory streams, such as those that show as passing through buildings, by filtering out the outliers. Use S4 to extract the road pattern part of the filtered trajectory streams and convert it into a density-based discretized image through kernel density estimation. That is, compress the continuous GPS trajectory data into a two-dimensional density map. Specifically, the GPS dataset processed into trajectory streams first undergoes an outlier filtering based on building footprint data to remove abnormal trajectory streams that traverse buildings. The filtered trajectory streams are then converted into density-based discretized images through kernel density estimation. During kernel density discretization, the complete GPS trajectory is compressed into a two-dimensional density estimate. This process divides the experimental area into 1×1 meter cells and calculates the number of trajectory streams passing through each cell to generate a two-dimensional histogram. This histogram is then convolved with a normal distribution function N(0; σ²) to simulate the expected GPS error distribution. S52: The image after Gaussian kernel smoothing is binarized. A set of grayscale thresholds is used to generate a binary mask for the road. The algorithm repeatedly performs binary skeletonization operations with different density thresholds. The skeleton of the road network is calculated by a thinning algorithm on the binarized image. Each time, the newly generated road segment is added to the skeleton. In the topology thinning step, the initial map generated by density processing is updated according to the trajectory matched by the map. All edges that cross zero times or only once are discarded. When there are multiple crossings of an edge generated based on the trajectory, the two intersections are merged. Each pixel is associated with an edge, and the Douglas-Peucker algorithm is used to generate the edges that make up the shape of each road segment until all grayscale thresholds are completed. Finally, a complete linear road network is obtained. S53: Extract the OD vertices of each trajectory in free-travel mode, and construct a vertex index for each vertex. Calculate a simple set of boundaries based on the vertex position and alpha value. Obtain the circumcircle radius r from the vertex set. Set the edges of the alpha shape within the range of r. Iterate through the vertex index and circumcircle radius, and add the edge set to the parameter edges using a radius filter. If the number of vertices is greater than 3, return the set of surrounding edges and construct a polygon using the edge set. Otherwise, directly create a polygon using the vertices. The area covered by the polygon is the free-travel area.

[0025] Finally, please see Figure 2 and Figure 3 , Figure 2 This is a schematic diagram of refined linear road extraction in a community according to an embodiment of the present invention; Figure 3 This is a schematic diagram of refined planar road extraction in a community according to an embodiment of the present invention; Please see Figure 4 , Figure 4 This is a schematic diagram of the hardware device in operation according to an embodiment of the present invention. The hardware device specifically includes: a community refined road network extraction device 401, a processor 402, and a storage medium 403.

[0026] A community-level refined road network extraction device 401: The community-level refined road network extraction device 401 implements the community-level refined road network extraction method.

[0027] Processor 402: The processor 402 loads and executes the instructions and data in the storage medium 403 to implement the community refined road network extraction method.

[0028] Storage medium 403: The storage medium 403 stores instructions and data; the storage medium 403 is used to implement the community refined road network extraction method.

[0029] The beneficial effects of this invention are: 1. This invention establishes a multi-level confidence relaxation constraint strategy to flexibly handle the interference of trajectory noise. By constraining pedestrian trajectory data with high-confidence partitioning, the interference of low-confidence trajectory noise is suppressed, enabling trajectory flow clustering to highlight the set of potential road objects with significant linear characteristics. Low-confidence level data relaxation is performed based on the cluster range to more accurately quantify and calculate pattern classification features. By modulating the input trajectory data source, the trajectory data is layered to adapt to tasks at different semantic scales, making the distinction between two different types of pedestrian trajectories more accurate and achieving an optimal balance between accuracy and coverage. 2. This invention introduces pedestrian trajectory flow semantics for road network generation to overcome the semantic ambiguity problem of low-frequency trajectory data. By enhancing the representation of pedestrian motion features through trajectory flow, it can more accurately depict the structure and connectivity of the road network than traditional methods, and is not affected by individual trajectory data. Furthermore, it can better adapt to the complexity brought about by the random movement of pedestrians in special scenarios at the community scale.

[0030] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for extracting a refined road network in a community, characterized in that: The method includes the following steps: S1. Data Acquisition: Acquire crowdsourced trajectory data sources; The crowdsourced trajectory data source consists of pedestrian trajectory data, including: pedestrian ID and pedestrian position coordinate information (x, y) at different times. The pedestrian trajectory data is represented as: pti (ID, x, y, t), where t represents the time corresponding to the i-th record; S2. Initial trajectory stream set construction: Construct an initial trajectory stream set based on the crowdsourced trajectory data source; S3. Trajectory Flow Filtering: Determine the confidence level of the trajectory flow set; establish a speed confidence level setting based on the speed value of the trajectory flow segment, and extract the trajectory flow set with high confidence; Step S3 is as follows: S31: Establish a formula for calculating the confidence level of velocity, and normalize the confidence level value to the range of 0 to 1; S32: Calculate the velocity value of the trajectory flow, substitute it into the velocity confidence formula to obtain the set of velocity confidence values ​​for all trajectory flows; S33: Extraction speed confidence value Greater than The trajectory flow, as a high-confidence trajectory flow, in which The set speed confidence threshold; In step S31, the formula for calculating the velocity confidence level is as follows: Where v represents the velocity of the trajectory flow, speed min and speed max The value is determined based on the concentrated range of the population's movement speed within the community; S4. Identify trajectory flow patterns: By setting thresholds for pedestrian group correlation indicators, trajectory flow patterns are classified into road patterns and free-traffic patterns. Step S4 is as follows: S41: Calculate the spatial geometric similarity of high-confidence trajectory flows, and perform trajectory flow clustering and merging based on the similarity; use the KNN clustering method to spatially cluster the trajectory flows, and group high-confidence trajectory sets that are geographically close and have similar geometric features into the same trajectory cluster; S42: Select a certain number of trajectory flows that are spatially clustered and group them into a group, ensuring that the trajectory flows in this group are concentrated in the same section of community road space or place space; S43: Calculate the similarity of trajectory behavior among different individuals within the group, considering only the top K individuals closest to the target individual i, and assuming that these K individuals are located in the neighborhood of i. By setting the group correlation threshold γ, the groups are divided into two types of trajectory patterns. S5. Extract community road network information: Analyze the morphology of trajectory flow set classification results under different modes, extract community linear road elements and area road area elements, and organize them into community road segments and free passage area elements.

2. The method for extracting a refined road network in a community as described in claim 1, characterized in that: In step S2, the pedestrian trajectory data is converted into trajectory stream records, and trajectory stream records with abnormal time intervals are removed to form an initial trajectory stream set.

3. The method for extracting a refined road network in a community as described in claim 2, characterized in that: Step S2 is as follows: S21: Extract the set of all pedestrian IDs from the pedestrian trajectory and perform deduplication; S22: for each pedestrian ID, obtain the time and position coordinate information of all track points belonging to the ID, and organize them into the pedestrian trajectory tr = {pt₁, pt₂,…, pt<0xE2><0x82><0x99>} of the ID, wherein pti (1 ≤ i<n) represents the space-time coordinate of the i-th track point, and all track points in tr are arranged in chronological order to obtain tr ’ = {pt₁, pt₂,…, pt<0xE2><0x82><0x99>}, wherein when i<j, pti(t) ≤ ptj(t); S23: extract tr ’ all adjacent trajectory point pairs {<pti, pti+1>| 1 ≤ i<n} therein, where the spatio-temporal coordinate of the previous trajectory point is pti, and the spatio-temporal coordinate of the next trajectory point is pti+1, and obtain all trajectory flows {f (pti, pti+1) | 1 ≤ i<n} of the ID; S24: Repeat S22 and S23 until all pedestrian IDs have been processed and the initial trajectory stream set is completed.

4. The method for extracting a refined road network in a community as described in claim 1, characterized in that: Step S5 is as follows: S51: Generate a density feature grayscale map based on the density of the trajectory flow; S52: The density feature grayscale image is smoothed and binarized using the Gaussian kernel image processing algorithm. A set of grayscale thresholds is used to generate a binary mask for the road. The algorithm repeatedly performs the binary skeletonization operation, iteratively adding the detected road segments to the skeleton, associating each pixel with an edge, and using the Douglas-Peucker algorithm to simplify and reconstruct each road segment until all grayscale thresholds have been executed, finally obtaining a linear road network. S53: Extract the OD vertices of each trajectory in free-travel mode, and construct a vertex index for each vertex. Calculate a simple set of boundaries based on the vertex position and alpha value. Obtain the circumcircle radius r from the vertex set. Set the edges of the alpha shape within the range of r. Iterate through the vertex index and circumcircle radius, and add the edge set to the parameter edges using a radius filter. If the number of vertices is greater than 3, return the set of surrounding edges and construct a polygon using the edge set. Otherwise, directly create a polygon using the vertices. The area covered by the polygon is the free-travel area.

5. A storage medium, characterized in that: The storage medium stores instructions and data to implement the community refined road network extraction method according to any one of claims 1 to 4.

6. A community-based refined road network extraction device, characterized in that: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the community refined road network extraction method according to any one of claims 1 to 4.

Images