Road network-track cooperative intersection type multimode integrated discrimination method
By integrating road network and trajectory data, using the supervised learning of graph convolutional network and high confidence samples, a multi-mode integrated discrimination method for intersection type of road network-trajectory collaboration is realized, solving the problem that intersection type discrimination in the existing technology depends on a single data source and manual annotation, and improving the discrimination accuracy and automation level.
Patent Information
- Application Number
- CN202510282664.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-11
- Publication Date
- 2025-06-27
AI Technical Summary
The prior art mainly relies on a single data source in the judgment of intersection types, ignores the fusion of multimodal data and the coordinated analysis of road network-trajectory, and the training sample data relies on manual annotation, which poses a risk of subjective misjudgment.
By integrating road network structure data and vehicle trajectory data, a multi-mode integrated discrimination method of intersection type with road network-trajectory cooperation is adopted, and an intersection type discrimination model is constructed using graph convolutional networks, and supervised learning based on high confidence samples is reduced to manual annotation dependence.
It significantly improves the accuracy of intersection classification, especially the identification ability of complex three-dimensional intersections, realizes the full process automation of intersection classification, reduces the risk of subjective misjudgment, and solves the problems of limitations of a single data source and high labeling costs.
Smart Images
Figure CN120220394A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of road network information mining, and in particular relates to a road network-trajectory collaborative intersection type multi-mode integrated discrimination method. Background Art
[0002] Intersections are key nodes in traffic networks. Different types of intersections have different impacts on traffic flow. By distinguishing and analyzing different types of intersections, traffic flow can be better dispatched and traffic congestion can be reduced. At present, existing studies mainly focus on the determination of intersection categories under a single data source, often ignoring the fusion of multimodal data and the collaborative analysis of road network-trajectory. In addition, the training sample data relies on manual annotation, which poses a risk of subjective misjudgment. Therefore, it has become an important research direction to integrate different types of traffic data (such as vehicle trajectories, road networks, etc.), conduct supervised learning of high-confidence samples, reduce reliance on manual annotation, enhance the intelligent identification and decision-making capabilities of intersection types, and realize the full process automation of intersection classification. Summary of the invention
[0003] The purpose of the present invention is to provide a road network-trajectory collaborative intersection type multi-mode integrated discrimination method to solve the problems existing in the above-mentioned prior art. By integrating road network structure data and vehicle trajectory data, the present invention can effectively capture the topological characteristics and dynamic behavior patterns of intersections, significantly improve classification accuracy, especially the recognition ability of complex three-dimensional intersections; and supervised learning based on high-confidence samples reduces the dependence on manual annotation, realizes the full process automation of intersection classification, and reduces the risk of subjective misjudgment. On the one hand, the present invention breaks through the limitation of a single data source, and on the other hand, alleviates the problem of high annotation cost in supervised learning, realizes the leap from "artificial experience driven" to "data intelligence driven" in intersection classification, and provides a core tool for urban traffic digitization. Accurate intersection type discrimination results provide important information for multi-scale map synthesis, high-precision map generation, automatic driving path planning, congestion analysis, etc.
[0004] To achieve the above object, the present invention provides a road network-track collaborative intersection type multi-mode integrated discrimination method, comprising:
[0005] Identifying a number of pending intersections based on the road network data, determining a real intersection based on the intersection characteristics of each of the pending intersections, and determining an intersection range corresponding to each real intersection;
[0006] Preliminarily judging the categories of each real intersection, obtaining the initial intersection type of each real intersection and marking it in the road network; the initial intersection type includes a high-confidence plane intersection, a high-confidence three-dimensional intersection and an intersection to be classified;
[0007] Construct a dual graph of the intersection road network based on the intersection ranges of various types of intersections, and assign graph node features to each node in the dual graph to obtain a quantitative expression of the graph structure of the intersection road network;
[0008] Construct a discriminant model for the type of intersection to be classified based on a graph convolutional network. Use the graph structures of high-confidence plane intersections and high-confidence three-dimensional intersections as training data, and train the discriminant model for the type of intersection based on the training data to obtain a trained discriminant model for the type of intersection;
[0009] Execute the task of discriminating the type of intersection for the intersection road network data based on the trained discriminant model for the type of intersection.
[0010] Optionally, the process of determining the real intersection specifically includes:
[0011] Identify the pending intersection T1 and the pending intersection T2 in the road network data based on road network topology inspection and morphological refinement of high road network density areas;
[0012] Calculate the features of each pending intersection, and the intersection features include topological connection features, reverse intersection point density features, and trajectory point density features;
[0013] Perform real intersection discrimination on each pending intersection based on multi-level discrimination rules and the calculated intersection features, and calculate the intersection range corresponding to each real intersection.
[0014] Optionally, the identifying the pending intersection T1 and the pending intersection T2 in the road network data based on road network topology inspection and morphological refinement of high road network density areas specifically includes:
[0015] Traverse the road network data and identify the nodes associated with three or more arcs as the pending intersection T1;
[0016] Perform kernel density analysis on the road network, extract high-density areas for morphological refinement, and identify the units with the number of pixel points in the 8-neighborhood greater than 2 as the pending intersection T2.
[0017] Optionally, the multi-level discrimination rules specifically include:
[0018] If both the pending intersection T1 and T2 exist, the pending intersection T1 is determined to be a real intersection;
[0019] If the pending intersection T1 does not exist, but the pending intersection T2 exists, and if the topological connection feature condition is met and one of the other intersection features is satisfied, the pending intersection T2 is determined to be a real intersection;
[0020] If neither the undetermined intersection T1 nor the undetermined intersection T2 exists, or the characteristics of each intersection do not meet the conditions, it is impossible to determine it as a true intersection.
[0021] Optionally, the process of obtaining the intersection range corresponding to each true intersection specifically includes:
[0022] Construct a Delanuary triangulation corresponding to each true intersection;
[0023] Identify the shortest side in the constructed Delanuary triangulation as the buffer circle radius of the intersection radiation area to obtain the corresponding intersection range.
[0024] Optionally, the preliminary judgment of each true intersection category specifically includes:
[0025] If the current intersection has the reverse convergence point density feature and exists in the set of undetermined intersections T1, it is judged as a high-confidence plane intersection;
[0026] If the current intersection does not have the reverse convergence point density feature, but has the topological connection feature and the trajectory point density feature, and does not exist in the set of undetermined intersections T1 but exists in the set of undetermined intersections T2, it is judged as a high-confidence three-dimensional intersection;
[0027] Mark the remaining intersections as intersections to be classified.
[0028] Optionally, each node in the dual graph is given graph node features, specifically including:
[0029] Based on the minimum bounding rectangle corresponding to the intersection range of each type of intersection as a mask, the road network is cropped to obtain the cropped intersection road network, and the dual graph of the intersection road network is constructed;
[0030] Design graph node features based on the road network and trajectory data, and the graph node features include road geometric features, road distribution features, road topological connection features, road traffic flow, and road operating speed;
[0031] Assign each graph node feature to the dual graph node to obtain a quantitative expression of the intersection road network graph structure.
[0032] Optionally, the training of the intersection type discrimination model based on the training data specifically includes:
[0033] Input the training data of the high-confidence plane intersection and high-confidence three-dimensional intersection road network graph structure into the intersection type discrimination model to be classified and predicted, and train according to the target loss function to obtain a trained intersection type discrimination model.
[0034] The technical effect of the present invention is:
[0035] Through multimodal data collaboration (road network + trajectory), the present invention enhances the semantic understanding of intersection functions; through the strategy of "high-confidence sample screening + retraining of samples to be classified", it overcomes the problem of insufficient utilization of unlabeled data; through multi-level discriminant rules, the application of graph convolutional networks, and rich feature engineering, it realizes efficient and accurate discrimination of intersection types. The present invention makes up for the deficiencies in the research on intersection type discrimination in existing work, can effectively solve the problems of long cycle and high cost in intersection type discrimination. The method proposed by the invention provides an effective technical reference for intersection type discrimination and can be applied to auxiliary traffic management decisions such as intersection semantic modeling, urban road network update, traffic flow simulation, and congestion governance. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.
[0037] The drawings forming a part of this application are used to provide a further understanding of this application. The illustrative embodiments and descriptions thereof of this application are used to explain this application and do not constitute an improper limitation of this application. In the drawings:
[0038] Figure 1 is the implementation flowchart in the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0039] The various exemplary embodiments of the present invention will now be described in detail. This detailed description should not be considered as a limitation of the present invention, but rather as a more detailed description of certain aspects, features, and implementation schemes of the present invention.
[0040] It should be understood that the terms used in the present invention are only for describing specific embodiments and are not used to limit the present invention. Additionally, for the numerical ranges in the present invention, it should be understood that each intermediate value between the upper and lower limits of the range is also specifically disclosed. Each intermediate value within any stated value or stated range, as well as each smaller range between any other stated value or intermediate value within the stated range, is also included in the present invention. The upper and lower limits of these smaller ranges can be independently included or excluded from the range.
[0041] Without departing from the scope or spirit of the present invention, various improvements and changes can be made to the specific embodiments of the present invention description, which are obvious to those skilled in the art. Other embodiments obtained from the description of the present invention are obvious to those skilled in the art. The description and embodiments of this application are only exemplary.
[0042] The words “include,” “including,” “have,” “contain,” etc. used in this article are open-ended terms, meaning including but not limited to.
[0043] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0044] like Figure 1 As shown, in this embodiment, a multi-mode integrated identification method for intersection types of road network-trajectory collaboration is provided, including: identifying a number of pending intersections based on road network data, determining real intersections based on the intersection features of each pending intersection and determining the intersection range corresponding to each real intersection; making a preliminary judgment on the category of each real intersection, obtaining the initial intersection type of each real intersection and marking it in the road network; the initial intersection type includes high-confidence plane intersections, high-confidence three-dimensional intersections and intersections to be classified; constructing a dual graph of the intersection road network based on the intersection ranges of each category of intersections, assigning graph node features to each node in the dual graph, and obtaining a quantitative expression of the intersection road network graph structure; constructing a model for distinguishing the type of intersection to be classified based on a graph convolutional network, taking the road network graph structures of high-confidence plane intersections and high-confidence three-dimensional intersections as training data, training the intersection type distinction model based on the training data, and obtaining a trained intersection type distinction model; executing the intersection type distinction task of the road network data of the intersection to be classified based on the trained intersection type distinction model.
[0045] The static geometry and spatial topology of the road network reflect the basic structure and layout of the road, while the vehicle trajectory provides strong data support for understanding the dynamic connection of the road network. The combination of the two can more comprehensively reflect the morphology of different types of intersections. In view of this, the present embodiment integrates the road network and trajectory information to perform multi-mode integrated discrimination of the intersection type, and discloses a multi-mode integrated discrimination method and system for the intersection type of the road network-trajectory collaboration, including: Determination of intersection location and spatial range: Analyze and mine the multi-mode features of the vehicle trajectory in the vector grid space about the intersection, design the multi-level discrimination rules of the intersection of the road network-trajectory collaboration, and carry out the determination of the intersection location and spatial range; Accurate discrimination of intersection type: Based on the result of multi-mode integrated identification of intersections, a graph convolutional neural network (GCN) that can effectively process complex graph structure data and automatically learn the relationship between graph nodes is used in the vector space to carry out accurate discrimination of intersection types of the road network-trajectory collaboration. This embodiment integrates different types of traffic data (such as vehicle trajectories, road networks, etc.) to enhance the intelligent discrimination and decision-making capabilities of intersection types, and provides technical support for the identification of plane intersections and three-dimensional intersections.
[0046] The specific implementation process of this embodiment includes:
[0047] Step 1, determination of intersection location and spatial range: First, identify the pending intersections based on the road network data, then design discrimination rules using the trajectory data to conduct multi-level discrimination on the pending intersections to obtain the real intersections, and finally determine the spatial range based on the real intersections.
[0048] Step 2, precise discrimination of intersection types: First, based on the intersection identification, conduct a preliminary category judgment to obtain high-confidence plane intersections, high-confidence three-dimensional intersections, and intersections to be classified; then use the intersection range to crop the road network to obtain the intersection road network, and conduct a dual transformation to complete the construction of the intersection road map structure; then design the graph node features based on the road network and trajectory data, and use the GCN network to accurately judge the types of intersections to be classified.
[0049] Implementable, the specific implementation method of Step 1 is as follows:
[0050] Step 1.1, identification of pending intersections: First, conduct a topological check on the road network, identify the nodes associated with three or more arcs as pending intersections T1. At the same time, conduct a kernel density analysis on the road network, extract the high-density areas for morphological thinning, and identify the units with more than 2 pixel points in the 8-neighborhood as pending intersections T2.
[0051] Step 1.2, discrimination of real intersections: Based on the topological connection features, reverse intersection point density features, and trajectory point density features, intersection results A, B, and C can be obtained. Then, based on the above topological connection features, reverse intersection point density features, and trajectory point density feature recognition results (A, B, C), design multi-level discrimination rules to discriminate the pending intersections (T1, T2) to obtain real intersections, and merge the intersections with similar positions of the identified intersections to obtain the final identification result.
[0052] Step 1.3, determination of intersection range: Construct a Delanuary triangulation for the identified intersections, and identify the shortest side in the triangle as the buffer circle radius of the intersection radiation area.
[0053] Implementable, the specific implementation method of Step 1.2 is:
[0054] If the pending intersection T1 exists and the intersection results A, B, and C satisfy the corresponding multi-level discrimination rules listed in Table 1, then the pending intersection T1 can be discriminated as a real intersection. If the pending intersection T1 does not exist and T2 exists, and the intersection results A, B, and C satisfy the corresponding multi-level discrimination rules listed in Table 1, then the pending intersection T2 can be discriminated as a real intersection.
[0055] Table 1 Multi-level discrimination rules for intersections
[0056]
[0057] Implementable, the specific implementation of step 2 is as follows:
[0058] Step 2.1, preliminary judgment of intersection category: Based on the identified intersections, judge high-confidence plane intersections, high-confidence grade-separated intersections, and intersections to be classified, and mark the recognition results.
[0059] Step 2.2, construction of the intersection road network structure: Use the minimum bounding rectangle of the intersection spatial range as a mask to crop the road network, obtain the cropped intersection road network, and construct the dual graph of the intersection road network.
[0060] Step 2.3, extraction of graph node features: Endow the dual graph nodes with road geometric features, road distribution features, road topological connection features, road traffic flow, road operating speed, etc. designed based on the road network and trajectory data to obtain a quantitative expression of the intersection road network graph structure.
[0061] Step 2.4, discrimination of intersection types: Input the training sample sets of marked plane intersections and grade-separated intersections into the GCN network for training to obtain an intersection type discrimination model. Input the dual graph of the intersection road network to be detected into the intersection type discrimination model to identify the classification information of the dual graph of the intersection road network to be detected.
[0062] Implementable, the specific implementation of step 2.1 is:
[0063] ① If an intersection has the reverse convergence point density feature and exists in the T1 set, then this intersection is judged as a high-confidence plane intersection;
[0064] ② If an intersection does not have the reverse convergence point density feature, but has topological connection features and trajectory point density features, and this intersection does not exist in the T1 set but exists in the T2 set, then this intersection is judged as a high-confidence overpass.
[0065] This embodiment also provides a multi-modal integrated discrimination system for intersection types based on road network-trajectory collaboration, including the following modules:
[0066] Intersection position and spatial range determination module, used for identifying the intersection position and range: First, identify the intersection to be determined based on the road network data, then use the trajectory data to design discrimination rules to perform multi-level discrimination on the intersection to be determined to obtain the real intersection, and finally determine the spatial range based on the real intersection.
[0067] An intersection type accurate discrimination module is used for discriminating the intersection type: First, based on the intersection recognition, a preliminary category judgment is carried out to obtain high-confidence plane intersections, high-confidence grade-separated intersections, and intersections to be classified; then, the road network is cropped by the intersection range to obtain the intersection road network, and a dual transformation is carried out to complete the construction of the intersection road map structure; then, based on the road network and trajectory data, the graph node features are designed, and the GCN network is used to accurately judge the types of intersections to be classified.
[0068] Compared with the prior art, in this embodiment, first, the topological connection features, density features, and reverse intersection point features reflected by the intersection in the road network and trajectory data are fully mined, and high-confidence plane intersections, high-confidence grade-separated intersections, and intersections to be classified are discriminated based on the design rules. Then, by comprehensively considering the road geometry, road distribution, road topology presented by the intersection road network and the road traffic flow, road operating speed, etc. reflected by the trajectory data, and taking into account the adjacent information of the road sections, the research on discriminating the types of intersections to be classified is carried out with the GCN network classification model as the basic classifier. It makes up for the deficiencies in the research on intersection type discrimination in the existing work, and can effectively solve problems such as the limitations of single data sources, the underutilization of unlabeled data, the long cycle and high cost of intersection type discrimination. The method proposed in this embodiment provides an effective technical reference for intersection type discrimination mining.
[0069] As described above, the above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
Claims
1. A road network-trajectory collaborative intersection type multi-mode integrated discrimination method, characterized in that: include: Identifying a number of pending intersections based on the road network data, determining a real intersection based on the intersection characteristics of each of the pending intersections, and determining an intersection range corresponding to each real intersection; Preliminarily judging the categories of each real intersection, obtaining the initial intersection type of each real intersection and marking it in the road network; the initial intersection type includes a high-confidence plane intersection, a high-confidence three-dimensional intersection and an intersection to be classified; Based on the intersection ranges of each type of intersection, a dual graph of the intersection road network is constructed, and each node in the dual graph is assigned a graph node feature to obtain a quantitative expression of the intersection road network graph structure; Constructing a model for distinguishing the type of intersection to be classified based on a graph convolutional network, taking a high-confidence planar intersection and a high-confidence stereoscopic intersection road network graph structure as training data, training the intersection type distinguishing model based on the training data, and obtaining a trained intersection type distinguishing model; Based on the trained intersection type discrimination model, the intersection type discrimination task is performed on the intersection road network data to be classified.
2. A road network-track collaborative intersection type multi-mode integrated identification method according to claim 1, characterized in that: The process of determining the real intersection specifically includes: Identify the pending intersection T1 and the pending intersection T2 in the road network data based on the road network topology check and the morphological refinement of the high road network density area; Calculating the features of each pending intersection, wherein the intersection features include topological connection features, reverse intersection density features, and trajectory point density features; Based on the multi-level identification rules and calculated intersection features, the real intersection of each pending intersection is identified, and the intersection range corresponding to each real intersection is calculated.
3. A road network-track collaborative intersection type multi-mode integrated identification method according to claim 2, characterized in that: The identifying of the pending intersection T1 and the pending intersection T2 in the road network data based on the road network topology check and the high road network density area morphology refinement specifically includes: Traverse the road network data and identify nodes associated with three or more arcs as pending intersections T1; Kernel density analysis was performed on the road network, high-density areas were extracted for morphological refinement, and units with more than 2 pixels in the 8-neighborhood were identified as pending intersections T2.
4. A road network-track collaborative intersection type multi-mode integrated identification method according to claim 2, characterized in that: The multi-level discrimination rules specifically include: If both pending intersections T1 and T2 exist, then pending intersection T1 is determined to be a true intersection; The pending intersection T1 does not exist, but the pending intersection T2 exists. If the topological connection feature condition is met and one of the conditions in the other intersection features is met, the pending intersection T2 is determined to be a true intersection; If both pending intersection T1 and pending intersection T2 do not exist, or the characteristics of each intersection do not meet the conditions, it cannot be determined as a true intersection.
5. The multi-mode integrated identification method of intersection type based on road network-trajectory collaboration according to claim 1 is characterized in that: The process of obtaining the intersection range corresponding to each of the real intersections specifically includes: Construct the Delanuary triangulation corresponding to each real intersection; The shortest side in the constructed Delanuary triangulation is identified as the radius of the buffer circle of the intersection radiation area to obtain the corresponding intersection range.
6. A road network-track collaborative intersection type multi-mode integrated identification method according to claim 1, characterized in that: The preliminary judgment of each real intersection type specifically includes: If the current intersection has the reverse intersection density feature and exists in the set of pending intersections T1, it is judged as a high-confidence planar intersection; If the current intersection does not have the reverse intersection density feature, but has the topological connection feature and the trajectory point density feature, and does not exist in the pending intersection T1 set but exists in the pending intersection T2 set, it is judged as a high-confidence three-dimensional intersection; The remaining intersections are marked as intersections to be classified.
7. A road network-track collaborative intersection type multi-mode integrated identification method according to claim 1, characterized in that: Each node in the dual graph is endowed with graph node features, specifically including: The road network is clipped based on the minimum circumscribed rectangle corresponding to the intersection range of each type of intersection as a mask to obtain a clipped intersection road network, and a dual graph of the intersection road network is constructed; Designing graph node features based on road network and trajectory data, wherein the graph node features include road geometry features, road distribution features, road topology connection features, road flow and road running speed; The characteristics of each graph node are assigned to the dual graph node to obtain a quantitative expression of the intersection road network graph structure.
8. The multi-mode integrated identification method of intersection type based on road network-trajectory collaboration according to claim 1 is characterized in that: The training of the intersection type discrimination model based on the training data specifically includes: The training data of the road network graph structure of high-confidence plane intersections and high-confidence three-dimensional intersections are input into the intersection type discrimination model to be classified for classification prediction, and training is performed according to the target loss function to obtain a trained intersection type discrimination model.