Multi-source traffic GIS road network multi-level real-time fusion update method

Abstract

Aiming at the application demand of high-precision vector data fusion, this application proposes a five-layer data fusion model based on CityGML, and realizes a multi-level real-time fusion and update method covering the transportation GIS road network; first, a multi-level, object-oriented, lane-level fusion model based on the OGC standard is proposed, which defines and describes the model at the conceptual, logical and physical levels respectively, and has a complete explanation of the geometry, semantics and topology involved in the model; second, an adaptive data fusion method is designed to realize clustering analysis, linearization and topological reconstruction of the probe car, vector road network matching based on clustering analysis of the vector road network and fusion of POI data based on location and semantic proximity information; third, an LOD‑s model management module is designed in an integrated manner; experimental results show that the tool meets the needs of multivariate vector data fusion, and compared with manual adaptive fusion, it can greatly improve efficiency while meeting the multi-level and high-precision requirements.

Description

Technical Field

[0001] The present application relates to a real-time fusion and updating method for a GIS road network, and in particular to a real-time fusion and updating method for a multi-source transportation GIS road network at multiple levels, belonging to the technical field of multi-level fusion of transportation road networks. Background Art

[0002] With the rapid development of urbanization and transportation, the road network is becoming more and more complex, and various types of road traffic network information are characterized by hugeness, diversity and complexity. At the same time, the demand for road networks in various fields is becoming more and more refined, and different levels of industries have different requirements for road networks. At this stage, more detailed expression and simulation of huge traffic vector data at a more microscopic level (sub-meter level) have been put into use. Fine (sub-meter level) road network data has the characteristics of richer content, more diverse features, faster update speed, more microscopic and finer expression level, and the lane-level navigation application combined with more refined lane-level data is also developing rapidly. With the aim of solving the problems of traffic safety, efficiency and environment in the field of transportation, the "Internet of Vehicles" has also come into being. Its important development branch is the fine position matching between vehicles and roads. The highway informatization involved in the "Internet of Vehicles" has put forward higher requirements for high-precision navigation and positioning. The key basis for the application of technologies such as unmanned driving and unmanned vehicles lies in the high-precision guidance and judgment of accurate lane-level (sub-meter level) vector network data.

[0003] At the same time, with the explosive growth of basic geographic data, the original accumulation of geographic data has gradually been completed, and the core of multi-source vector data has shifted from data production to data update, and data sharing, application and update have become a hot topic. Currentness has become an important indicator of data evaluation today. At present, the acquisition of large-scale, high-precision multi-source vector data has been realized. An important way to ensure data timeliness is to integrate multi-source data, adopt the advantages of each multi-source data, and complete the timely update of data. However, due to the lack of effective mechanisms and technical methods for data update, slow update speed and poor effect quality have become urgent problems that are difficult to ensure data timeliness.

[0004] At present, the four major mainstream vector data, point of interest data, DLG, navigation data, and rover data, have different data characteristics. In terms of data content, point of interest data consists of all points of interest included in the features, including the basic name, coordinates and associated attribute information of the features. The digital line drawing of the data includes all vector data of the features that can be abstracted as points, lines, surfaces, and bodies. The data format is consistent and easy to share. At present, navigation data has multiple data formats and rich content; rover data includes the data number, location, time, speed and other information sent back at all times. In terms of data update, POI data adopts manual collection and update, user update review and other methods, DLG adopts linkage update and dynamic update methods, and its accuracy is good, but due to the inherent poor timeliness of its metadata, the data timeliness is poor, while navigation data adopts real-time update method, which has good timeliness and high accuracy; in terms of data update method, DLG data adopts the method of re-measurement and storage by professionals, so the data accuracy is high, which also causes the disadvantages of high consumption and long cycle, while navigation data can be updated centrally with the help of volunteers, with fast update speed, and the accuracy needs to be further improved. The probe car can accurately describe the location and time of the platform every 1-10 seconds through hardware measures such as the satellite navigation receiver installed on the taxi, but there is still a sub-meter positioning accuracy error. In summary: multi-source traffic vector road network data has its own advantages and conflicts. Therefore, how to have the advantages of multi-source data and abandon its disadvantages at the same time becomes the key to solving the problem.

[0005] Current status of road network model: Traditional road network model represented by arc-node model has been widely used in GIS software due to its simple model and easy application. However, with the increase of data volume, the number of route segments has expanded rapidly, the attribute changes of arc segments have not been considered, and the disadvantages of not being able to handle one-to-many relationships have become increasingly prominent. In view of the above problems, the comprehensive development of multi-dimensional LRS data model requires the maintenance of the base layer, has poor flexibility, and the later design tends to be more complex and refined, and it is becoming more and more difficult to implement; UNETRANS model lacks three-dimensional temporal object management and does not take into account the actual lane topology; the concepts of "virtual lane" and "lane belt" proposed by the lane-oriented multi-level road network model do not conform to people's normal cognition, and there is a lack of clear modeling standards for complex objects such as intersections and viaducts. In summary: There are many existing road network models with different advantages and disadvantages, but in terms of meeting the fusion of multi-source data, they fail to meet the fusion requirements of multi-level refinement (lane level).

[0006] Multi-source vector data have different advantages. On the basis of completing data integration, how to fuse multi-source vector data is the key issue to solve the multi-level fusion of multi-source vectors. The objects involved in the road network are independent of each other, and objects are interconnected. There is a lack of unified model standards for comprehensive objects. Due to the diverse data audiences, the requirements for data accuracy also vary greatly. Appropriate fusion levels are constructed to meet diverse needs. Combined with the development trend of vector network models and user needs, an object-oriented, multi-level, sub-meter-level vector network fusion model is urgently needed.

[0007] In general, the existing multi-source road network fusion update technology has obvious shortcomings, and its main defects and design difficulties include:

[0008] First, the existing technologies, represented by basic geographic data, navigation data from Amap, Baidu, etc., do not truly realize the real-time fusion application and update of data from the coarse level (points and line objects) to the fine level (actual lanes). There are problems such as data not being mined in time, lack of correspondence between data model levels, and inability to meet the requirements of connecting to different data service providers based on the OGC standard. First, there are many existing data models, and the data are incompatible with each other. There is a lack of data models that improve their classification definitions at the detailed level, and a lack of multi-level real-time fusion framework. Second, there are problems such as the inconsistency between the data coordinate systems and map projections of different data sources, the inconsistency between attribute classification and attribute coding, and the deviation of data visualization caused by different data formats. There is a lack of a unified geographic framework for real-time fusion and update of multi-level road networks, and it is impossible to achieve cluster analysis for POI data, probe car data, road networks, and digital line maps, and it is impossible to meet the requirements of static and dynamic update fusion of vector data. Third, there is a lack of multi-source vector road network multi-level fusion and update tools.

[0009] Second, the acquisition of large-scale, high-precision multi-source vector data has been achieved, but the fusion of multi-source data updates still lacks effective mechanisms and technical methods, the update speed is slow, the effect quality is poor, and it is difficult to ensure the timeliness of the data. At present, the four major mainstream vector data, point of interest data, DLG, navigation data, and probe car data have different data characteristics. POI data adopts manual collection and update, user update review and other methods, and DLG adopts linkage update and dynamic update methods. Its accuracy is good, but its metadata inherent timeliness is poor; in terms of data update methods, DLG data adopts the method of re-measurement and storage by professionals, so the data accuracy is high, which also causes the disadvantages of high consumption and long cycle, while navigation data can be updated centrally with the help of volunteers, with fast update speed, and the accuracy needs to be further improved. The probe car currently still has sub-meter positioning accuracy errors. Multi-source traffic vector road network data have their own advantages, and their respective advantages have conflicting points. The existing technology cannot have the advantages of multi-source data at the same time, abandon its disadvantages, and cannot fuse and update multiple sets of data. The labor consumption is large, and the advantages of different data sources cannot be fully utilized, and the data reuse rate is low;

[0010] Third, the road network model of the existing technology does not take into account the attribute changes of arc segments and cannot handle one-to-many relationships. The drawbacks are becoming increasingly prominent. The comprehensive development of multi-dimensional LRS data models requires the maintenance of the base layer, which has poor flexibility. The later design tends to be more complex and refined, and it is becoming more and more difficult to implement. The UNETRANS model lacks three-dimensional temporal object management and does not take into account the actual lane topology. The lane-oriented multi-level road network model lacks clear modeling standards for complex objects such as intersections and viaducts. In terms of meeting the fusion of multi-source data, the existing road network model lacks the definition of road vector levels and the expression of complex objects due to the data structure barriers generated by data producers for data monopoly. It fails to meet the fusion requirements of multi-level refinement (lane level) and cannot achieve the fusion of multi-source traffic vector road network data.

[0011] Fourth, the existing technology faces several difficulties in the fusion of multi-source geographic data. First, it is necessary to integrate the data produced on demand by different production departments. Second, DLG, probe car data, road network, and POI data all have their own advantages and disadvantages. It is difficult to achieve lane-level road network data fusion, and it is impossible to make the fused data attributes rich, update quickly, and be accurate in real time. There is a lack of a unified geographic framework for multi-source traffic vector road network data; it is impossible to represent the needs of road network data fusion in computers, and there is a lack of a third road network fusion model that connects multiple heterogeneous data; there is a lack of fusion algorithms such as precise road segment matching of multiple data and construction of the relationship between road network and POI. There is a lack of unified model standards for comprehensive objects, and there is a lack of an object-oriented, multi-level, sub-meter-level vector network fusion model. Summary of the invention

[0012] In view of the fact that vector data has not really realized the real-time fusion application and update of data from the coarse level (points and line objects) to the fine level (actual lanes), the data has not been mined in time, the hierarchical correspondence of the data model is lacking, and it cannot meet the requirements of connecting to different data service providers based on the OGC standard, this application proposes a multi-source road network vector multi-level real-time fusion and update method; first, in view of the problem that there are many existing data models and the data are incompatible with each other, a multi-source vector road network multi-level real-time fusion model based on the CityGML standard is proposed, the data model of its classification definition is improved at the detail level, and a multi-level real-time fusion framework is designed; second, for different In order to solve the problems of inconsistency between the data coordinate system of the data source and the map projection, inconsistency between the attribute classification and attribute coding, and deviation in data visualization caused by different data formats, a unified geographic framework for multi-level real-time fusion and update of road networks is proposed to realize cluster analysis of POI data, probe car data, road networks, and digital line maps, so as to achieve high-precision matching and fusion within and between multi-source data types, and meet the static and dynamic update fusion of vector data; thirdly, a multi-level fusion and update tool for multi-source vector road networks is realized, which achieves a better effect of multi-level road network fusion and provides new data fusion ideas and methods for multi-level vector data applications.

[0013] In order to achieve the above technical effects, the technical solutions adopted in this application are as follows:

[0014] The multi-level real-time fusion and update method of multi-source traffic GIS road network is: first, a multi-level, object-oriented, lane-level road model is set based on the OGC standard; second, a five-layer data fusion model based on CityGML is proposed by adopting the high-precision fusion of vector network and POI data, and a multi-level real-time fusion and update method covering the traffic GIS road network is realized;

[0015] First, a multi-source vector road network multi-level real-time fusion model: the first is the geospatial cognitive conceptual model, including the five-level hierarchical division of the fusion model and the fusion model design framework; the second is the geospatial data logical model; and the third is the geospatial structure physical model.

[0016] A multi-source vector road network multi-level real-time fusion model based on the CityGML standard is proposed. Its classification definition is improved at the detail level, and a multi-level real-time fusion framework is designed. Based on the abstract expression method of different elements covered by vector data and the geometric, semantic, and topological relationships between elements and between different levels of elements, a three-layer architecture design of the data model is realized.

[0017] Second, multi-source road network multi-level real-time fusion update model algorithm: First, consistency coordination fusion, including multi-source spatial vector data consistency processing, unified combination of spatial coordinate system and map projection, unified combination of classification and attribute coding, and conversion and unification of multi-source road network data formats; second, real-time fusion and unification of multi-source road network multi-level models, including multi-level real-time unified model design, real-time fusion model and algorithm combination; third, vector road network data adaptive matching fusion, including POI matching fusion based on location and semantics, road network matching fusion based on road network change mode; fourth, probe car matching and aggregation, including calculation of probe car feature elements for multi-level fusion update of road network, matching of probe car and road network information, and calculation of probe car hotspot path based on cluster analysis;

[0018] A unified geographic framework for real-time fusion and updating of multi-level road networks is proposed to carry out preliminary data integration and subsequent fusion analysis of multi-source traffic vector road network data involved in multi-source data fusion and matching, realize clustering analysis of POI data, probe car data, road networks, and digital line maps, achieve high-precision matching and fusion within and between multi-source data types, and meet the static and dynamic update fusion of vector data.

[0019] Multi-source traffic GIS road network multi-level real-time fusion update method, further, the fusion model is divided into five levels: vector geographic data mainly involves various types of data on the traffic road network, which are generally summarized as six basic elements: lanes, intersections, tunnels, viaducts, roundabouts, and squares. The following hierarchical division rules are formulated:

[0020] LOD-0: Basic description of the location of various features using points or lines;

[0021] LOD-1: Shape description is performed based on various features;

[0022] LOD-2: Use surfaces with interfaces to describe the connectivity of various features;

[0023] LOD-3: Use lane-level surfaces to describe the connectivity of various features and lane shapes;

[0024] LOD-4: Lane-level faces are used to describe the connectivity of various features and the shape of lanes;

[0025] Based on the above definitions of model modules and hierarchical division, a multi-level conceptual hierarchical division diagram is obtained.

[0026] Multi-source traffic GIS road network multi-level real-time fusion update method, further, the fusion model design framework: the multi-level road network model is grouped and packaged according to logical properties, and the following features are set:

[0027] (1) Feature 1: The objects in the network feature package include rich multi-scale geometric features, and the multi-scale geometric expression is constructed through internal coding representation association;

[0028] (2) Feature 2: Events include static facilities and random dynamic time, and are manifested as multi-scale forms of points;

[0029] (3) Feature 3: Mobile objects include probe vehicle objects related to traffic behavior in the vector road network;

[0030] (4) The model design is based on query, path planning, location reference, events, network features, mobile objects, domain knowledge, and relationship modules;

[0031] (5) Domain knowledge includes prescriptive classification of vehicle types and road functions in probe car data and giving specific labels;

[0032] (6) Query includes internal query and mutual query between elements.

[0033] Multi-source traffic GIS road network multi-level real-time fusion update method, further, the geographic spatial data logical model design: for each level of the multi-level logical model, the geometry, semantics and topology are described:

[0034] (1) LOD-0 abstracts the vector network at the most macro level, describing the simplest geometric form of the road with a point or a single line, which is located at the center of the object. LOD-O geometric objects are used to describe the road network layout at a macro level;

[0035] (2) LOD-1 does not have any semantic meaning in describing the shapes of objects divided in the vector network;

[0036] (3) LOD-2 divides the six categories of objects in the vector network into object exits according to the connection nodes between them, and adds semantic relationships by adding nodes between objects;

[0037] (4) LOD-3 abstracts the lane level of the vector network system, using double lines to describe the geometric form of the vector network data. The double lines are located at the center of the lane. For roads without a dedicated central isolation zone, they are still expressed as a single line.

[0038] (5) LOD-4 is the most detailed abstraction of the vector network system, describing lane-level abstract features, describing the geometric shape of the road with multiple lines, each line segment has directionality, and each line segment has different traffic semantic characteristics;

[0039] Design of physical model of geographic spatial structure: objects involved in the vector road network are divided into six categories of road network elements and one category of POI elements according to their main usage frequency, the road network is segmented and the topological construction method of each layer of network is described, and the following data expression rules are set:

[0040] (1) Rule 1: For LOD0 networks, objects are located at their intersections and road segments are located at the logical intersections of streets. Objects are bidirectionally connected without semantic attribute constraints.

[0041] (2) Rule 2: Viaducts and roundabout buildings are represented as a point feature at the LOD0 level;

[0042] (3) Rule 3: To ensure the semantic association between multiple levels in the multi-level road network model, additional road segment nodes are added.

[0043] Multi-source traffic GIS road network multi-level real-time fusion update method, further, multi-source spatial vector data consistency processing: divided into the unification of spatial coordinates and map projection, the unification of classification and grading and attribute coding, the unification of classification and grading and attribute coding, and the conversion and unification of data formats;

[0044] Unification of spatial coordinate system and map projection: WGS84 spatial coordinate system and Gauss Krüger projection coordinates are uniformly adopted, the elevation coordinate system adopts the 1985 National High-Level Datum, the conversion method adopts the form of code conversion combined with ArgGIS and ENVI image processing software, large-scale spatial coordinate system conversion and projection coordinate conversion are unified by code, and single image is converted by image processing software.

[0045] Multi-source traffic GIS road network multi-level real-time fusion update method, further, unified combination of classification and attribute coding: different classification and attribute coding methods are used for vector road network and POI data, among which the classification and grading of vector road network is based on the six-category five-layer classification system in the road network model;

[0046] Unification of multi-source road network classification and attribute coding: According to the data sources that need classification, classification and attribute coding, they are unified into two categories: road network data and POI data. The coding method will follow the following rules:

[0047] 1) Road network classification and attribute coding: Based on the design of a multi-level road vector model, the road network classification is classified according to six major elements and graded according to a five-level model. Its attribute coding will follow the coding method of the geographic spatial structure physical model;

[0048] 2) POI data classification and attribute coding: Classification is carried out according to a four-level classification system.

[0049] Multi-source transportation GIS road network multi-level real-time fusion update method, further, multi-level real-time unified model design: abstract the real-world geographic entities and their relationships, build several windows for understanding geographic space with geographic regions as boundaries, the data region includes several data blocks, each data block includes several geographic element layers, each element layer is relatively independent in data structure and organization, data update, query, analysis and display operations are based on element layers as basic units;

[0050] The multi-source road network multi-level unified model adopts a hierarchical structure to organize data, and adopts a topological structure to achieve integrated management of attribute and geometric data based on the overall target operation of the surface object, meeting the requirements of multi-source data integration;

[0051] The multi-source road network multi-level unified model takes geographic feature objects as the most basic feature unit, and stores its attribute data and spatial data in an integrated manner, adopting the entire process of object-oriented data storage and GIS data operation;

[0052] Joint unification of real-time fusion models and algorithms: Fusion of different algorithms at different levels:

[0053] Extraction of LOD-O level: POI data is fused at the LOD-O level using the relationship between position and semantic proximity in model design, and the fused vector network data of POI data is realized at the LOD-0 level. The node and geometric center position are obtained based on the DLG data with the highest accuracy level to realize its 0-level fusion;

[0054] The LOD-1 level obtains its shape features according to the attribute table to achieve information extraction at level 1;

[0055] The LOD-2 level obtains the connectivity of its domain by intersecting objectified elements;

[0056] The LOD-3 level realizes the linearization of the probe car data by directional aggregation and topological reconstruction of the probe car data and performs probabilistic matching of the navigation data to achieve LOD-3 level fusion;

[0057] The LOD-4 level realizes the linearization of the probe car data through directional road network aggregation and topology reconstruction, and achieves LOD-4 level fusion through probabilistic matching with the navigation data.

[0058] Multi-source traffic GIS road network multi-level real-time fusion update method, further, POI matching fusion based on the combination of location and semantics: the accuracy of POI fusion set is improved by combining spatial location and attribute information;

[0059] Road network matching fusion based on road network change patterns: includes three categories:

[0060] The first category is that in the same area, data with better currentness contain more spatial elements than data with worse currentness, which belongs to the 0:1 matching in spatial data matching;

[0061] The second category is that in the same area, data with better currentness include fewer spatial elements than data with worse currentness due to spatial changes, which belongs to 1:0 matching in spatial data matching;

[0062] Category 3: In the same region, data with better current status replaces spatial elements with poor current status, and the situation is complex and diverse;

[0063] According to the change mode of vector data, it satisfies the real-time and dynamic vector data fusion, and matches the map change at time t from a certain time point t of the road network. There are two modes: one is to match the road network data A at time t to the map change at time t. i The first is to use the map matching of road network data B, and the second is to use the real-time transmitted POL data to update the map matching at a certain moment.

[0064] Multi-source traffic GIS road network multi-level real-time fusion update method, further, probe car matching and aggregation: the probe car data uses satellite navigation positioning technology, and according to its sampling rate, regularly records its driving longitude and latitude, time information at regular intervals. At the same time, the potential instantaneous speed longitude and latitude information is processed by the probe car clustering analysis method and model algorithm, and by converting the probe car text information into spatial data format information, it is associated with the road network data in time and space, and the road network information changes are detected in real time and dynamically, and the connection relationship between lanes is discovered, thereby extracting the geometric, semantic, and topological information of the complex urban road network;

[0065] 1. Calculation of probe car feature elements for multi-level fusion and update of road network: The definition and calculation of probe car data involved in multi-level fusion and update of road network include:

[0066] 1) Determine the traffic volume ps in the direction:

[0067] p s =(X d +Y e ) / (r d +r e ) Formula 1

[0068] in:

[0069] p s : Traffic volume in the direction to be tested at LOD-3 level, unit: vehicles / min;

[0070] X d: The number of vehicles traveling in the opposite direction of the probe vehicle (in the test direction) at the LOD-3 level, unit: vehicles / min;

[0071] Y e : At the LOD-3 level, when a vehicle to be tested is traveling in the undetermined direction, the number of vehicles that overtake the probe vehicle minus the number of vehicles overtaken by the probe vehicle;

[0072] r d : The driving time of a vehicle to be tested at the LOD-3 level when driving in the direction and reverse direction of the measured traffic flow, unit: vehicle / min;

[0073] r s : At the LOD-3 level, the driving time of a vehicle to be tested traveling in the direction of the flow to be measured, unit: min;

[0074] 2) Average travel time T s :

[0075] R s =r s -Y s / p s Formula 2

[0076] Where: R s To determine the average travel time of a road section;

[0077] 3) Average vehicle speed U s :

[0078] U s =(J / R s )×60 Formula 3

[0079] Among them: U s is the average speed of the observed section, in km / h, J is the length of the observed section, in km;

[0080] When using equations 1 to 3 for calculation, X d , Y e , r d , R s The arithmetic mean is taken for calculation. The data set is cleaned and the abnormality of vehicle positioning information is judged by taking the standard deviation of the first-order difference of the positioning information. The activity range is judged by the variance of the longitude and latitude to clean the directional outliers. By eliminating the gross error data, the erroneous statistics of the data process caused by the error data are avoided.

[0081] A multi-source traffic GIS road network multi-level real-time fusion update method, further, the probe car and road network information matching: based on the prior knowledge of high-precision lane-level road network data, the positioning error of the probe car data is corrected, and the location information of the probe car data is corrected to the correct road network, and the probe car text data information is associated with the road network data information through the map matching algorithm;

[0082] Calculation of hot spots of probe vehicles based on cluster analysis: On the basis of undirected, by adding the same-direction condition basis, we can find the probe vehicle trajectory data connected with the same density, and by mining the distribution law of the probe vehicle data, we can generate directed edges according to the clustering class, complete the data extraction of LOD-4 level, and obtain the topological structure diagram of the complex road network;

[0083] By analyzing the approximation and clustering characteristics of its spatial data, we further analyze the correlation between the spatial distribution pattern of the probe car trajectory data and the traffic geographical characteristics. Finally, through cluster analysis, we can explore the spatial distribution pattern and traffic geographical correlation of the probe car data. By using approximate cluster analysis of directed data, we can achieve the road network clustering analysis effect of the probe car data and realize the dynamic and timely fusion of road network and probe car data.

[0084] Compared with the prior art, the innovations and advantages of this application are:

[0085] First, this application is aimed at the application needs of high-precision vector data fusion, and aims at the problems existing in existing methods: there is no multi-level, object-oriented, lane-level road model based on OGC standards that can be provided to different map suppliers; it is impossible to achieve high-precision fusion of vector network and POI data, and proposes a five-layer data fusion model based on CityGML, and implements a multi-level real-time fusion and update method covering transportation GIS road networks; first, a multi-level, object-oriented, lane-level fusion model based on OGC standards is proposed, and the model is defined and described at the conceptual, logical, and physical levels, and there is a complete explanation of the geometry, semantics, and topology involved in the model; second, the data to which it is adapted is designed. The fusion method realizes clustering analysis, linearization and topological reconstruction of the probe car, vector road network matching based on clustering analysis of the vector road network, and fusion of POI data based on location and semantic proximity information; the third is to integrate the design of the LOD-s model management module, and design six modules according to its function, including coordinate system conversion, classification and grading, format consistency processing, clustering analysis and topological reconstruction of probe car data, probabilistic registration of road network data, and LODs model generation management; finally, using Amap and Tiandi map data as experimental data, combined with the analysis of probe car data, the experimental results show that the tool meets the needs of multivariate vector data fusion. Compared with manual adaptive fusion, it can greatly improve efficiency while meeting multi-level and high-precision requirements;

[0086] Second, currently vector data has not really realized the real-time fusion application and update of data from the coarse level (points and line objects) to the fine level (actual lanes). In view of the problems that data has not been mined in time, the correspondence between the data model levels is lacking, and it is not possible to connect to different data service providers based on the OGC standard, this application proposes a multi-source road network vector multi-level real-time fusion update method; first, in view of the problem that there are many existing data models and data are incompatible with each other, a multi-source vector road network multi-level real-time fusion model based on the CityGML standard is proposed, the data model of its classification definition is improved at the detail level, and a multi-level real-time fusion framework is designed; second, for different data In order to solve the problems of inconsistent data coordinate system with map projection, inconsistent attribute classification and attribute coding, and deviation in data visualization caused by different data formats, a unified geographic framework for real-time fusion and update of multi-level road networks is proposed to realize cluster analysis of POI data, probe car data, road networks, and digital line maps, so as to achieve high-precision matching and fusion within and between multi-source data types, and meet the static and dynamic update fusion of vector data; thirdly, a multi-source vector road network multi-level fusion update tool is realized, which is suitable for the fusion processing of multi-source vector data, and achieves a good multi-level road network fusion effect, providing new data fusion ideas and methods for multi-level vector data applications;

[0087] Third, this application is aimed at the fusion and application needs of multi-source traffic vector road network data. It is based on the current status of the diversification of existing vector data and the data structure barriers caused by data producers for data monopoly. It aims at the problems of different advantages of existing vector geographic data, lack of road vector level definition, and lack of complex object expression. It comprehensively considers the development status of multi-source vector multi-level fusion, multi-source vector data model, and algorithm, and studies the design and implementation of multi-source vector multi-level fusion tools, realizing the fusion of multi-source traffic vector road network data, and achieving the purpose of enhancing the timeliness and accuracy of multi-source fusion vector geographic data; on the one hand, it realizes the integration of data produced on demand by different production departments, and on the other hand, it realizes the fusion of lane-level road network data based on the advantages and disadvantages of DLG, detection vehicle data, road network, and POI data, so that the fused data has the characteristics of rich attributes, fast update speed, and real-time accuracy. The built fusion system can fuse and update multiple sets of data with very little manual consumption, and can fully utilize the advantages of different data sources to improve the reuse rate of data.

[0088] Fourth, the innovation of this application is also reflected in three aspects: first, for multi-source traffic vector road network data, a unified geographic framework; second, to solve the representation of road network data fusion needs in computers, propose a road network fusion model, and connect a variety of heterogeneous data. The third is the fusion algorithm of accurate road section matching of multiple data, construction of road network and POI association relationship; it not only improves the timeliness, but also updates the road network data through real-time updated probe car data; it also improves the accuracy, and improves the accuracy of the probe car data in real time through high-precision road network and POI data; by superimposing the probe car data on the multi-source road network data to achieve the purpose of dynamically improving the timeliness and spatial accuracy, a multi-level road network model is designed, and multi-level, high-precision, and time-honored multi-source traffic vector road network data is obtained, which has good practical value; BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 It is a schematic diagram of three expression levels of the multi-source vector road network multi-level real-time fusion model.

[0090] Figure 2 It is a classification diagram of the six basic elements of the multi-source vector road network multi-level real-time fusion model.

[0091] Figure 3 It is a multi-level conceptual division diagram of the multi-source vector road network real-time fusion model.

[0092] Figure 4 It is a flowchart of multi-source spatial vector data consistency processing.

[0093] Figure 5 It is a flow chart for designing code for classification and grading rules of multi-source transportation GIS data.

[0094] Figure 6 It is a schematic diagram of the classification, grading and attribute coding system of multi-source POI data.

[0095] Figure 7 It is a schematic diagram of the integration and fusion of multi-source road network data formats.

[0096] Figure 8 It is a flow chart of the POI matching fusion solution based on the combination of location and semantics.

[0097] Fig. 9 This is the directed clustering effect and topological structure diagram of the detection vehicle data.

[0098] Fig.10 These are the DEM and DOM data maps before and after clipping and splicing in this application.

[0099] Fig.11 This is a schematic diagram of the superposition effect of the detection car and road network data.

[0100] Fig.12This is the effect diagram of multi-source data fusion and update at the lane level. Specific implementation methods

[0101] The following, in conjunction with the accompanying drawings, further describes the technical solution of the multi-source transportation GIS road network multi-level real-time fusion update method set up in this application, so that technical personnel in this field can better understand the application and implement it.

[0102] At present, vector data represented by basic geographic data, Amap, Baidu and other navigation data have not truly realized the real-time fusion application and update of data from the coarse level (points and line objects) to the fine level (actual lanes). In view of the problems of data not being mined in time, lack of hierarchical correspondence of data models, and inability to meet the requirements of connecting to different data service providers based on the OGC standard, this application proposes a multi-source road network vector multi-level real-time fusion update method. It mainly includes:

[0103] (1) In order to solve the problem of numerous existing data models and incompatible data, a multi-source vector road network multi-level real-time fusion model based on the CityGML standard is proposed. The data model with classification definition is improved at the detail level, and a multi-level real-time fusion framework is designed. Based on the abstract expression method of different elements covered by vector data and the geometric, semantic, and topological relationships between elements and between different levels of elements, a three-layer architecture design of the data model is realized.

[0104] (2) In order to address the problems of inconsistency between data coordinate systems and map projections of different data sources, inconsistency between attribute classification and attribute coding, and deviation in data visualization caused by different data formats, a unified geographic framework for real-time fusion and updating of multi-level road networks is proposed. The differences in geometry, semantics, and topology of multi-source geographic spatial road network data fusion are analyzed. The early data integration involved in multi-source data fusion and matching and the fusion analysis of multi-source traffic vector road network data are carried out. Cluster analysis of POI data, probe vehicle data, road networks, and digital line maps is implemented to achieve high-precision matching and fusion within and between multi-source data types, satisfying the static and dynamic update fusion of vector data.

[0105] (3) Design and implement a multi-source vector road network multi-level fusion update tool to verify the effectiveness of the model, the accuracy of the fusion, and the efficiency of the algorithm. The experimental results show that the fusion model is suitable for the fusion processing of multi-source vector data and achieves a good multi-level road network fusion effect, providing new data fusion ideas and methods for multi-level vector data applications.

[0106] The specific steps of the multi-source traffic GIS road network multi-level real-time fusion update method are as follows:

[0107] Step 1: Multi-source data preprocessing: Preprocessing the geometry, attributes, and semantics of multi-source traffic vector road network data;

[0108] Step 2: Model loading: According to the concept of multi-source vector road network multi-level real-time fusion model, the multi-source vector data are loaded into the database according to the multi-level concept.

[0109] Step 3: For multi-source spatial data, the road network and POI points are matched according to the classical algorithm probability relaxation method, and the road network data is cleaned by spatial cluster analysis, and the detection vehicle data is integrated with the matching fusion method based on space and attributes to implement static road network update and fusion of multi-source POI data;

[0110] Step 4: Spatialize the rover data. The vector data is in a unified format, processed for vector consistency, and loaded into the database;

[0111] Step five: Based on lane-level probe car correction and fusion of high-precision vector data, the probe car data is superimposed on the road network data. Through real-time superposition and cluster analysis of the probe car and the static road network, a probe car road network is constructed and its topological relationship is analyzed, ultimately achieving dynamic real-time fusion of the probe car data and road network information.

[0112] 1. Multi-source vector road network multi-level real-time fusion model

[0113] A multi-source vector road network multi-level real-time fusion model for vector fusion is established, with a five-level fusion hierarchical model. The multi-level associations of the fusion model in terms of module division, geometry, semantics, and topology are designed in detail from the perspectives of the geographic space cognitive concept model, the geographic space data logical model, and the geographic space structure physical model.

[0114] Based on the geospatial cognitive conceptual model, geospatial data logical model, and geospatial structural physical model, the geographical entities, representations, and relationships of the road network are abstracted and described, realizing the three levels of expression of the road network from geographical reality to computer. The three levels of abstract expression are from top to bottom, from abstract to concrete, such as Figure 1 shown.

[0115] This application uses UML language description to establish a multi-source vector multi-level fusion model as the logical model of traffic GIS vector fusion. The data structure is explained using Oracle's design language PLSql. Among them, multi-source traffic vector road network data modeling is divided into five steps: demand modeling, definition of entity relationships, correct expression of geographic elements, matching database elements, and physical design.

[0116] 1. Design of conceptual model of geographic spatial cognition

[0117] 1. Five-level division of the fusion model

[0118] Vector geographic data mainly involves various types of data on the traffic road network, which are generally summarized as six basic elements: lanes, intersections, tunnels, viaducts, roundabouts, and squares. Figure 2 shown.

[0119] The hierarchy of the fusion model is based on the CityGML hierarchical division rules, and at the same time meets the requirements of multi-level applications. Different levels are supplemented with definitions, and the following hierarchical division rules are formulated:

[0120] LOD-0: Basic description of the location of various features using points or lines;

[0121] LOD-1: Shape description is performed based on various features;

[0122] LOD-2: Use surfaces with interfaces to describe the connectivity of various features;

[0123] LOD-3: Use lane-level surfaces to describe the connectivity of various features and lane shapes;

[0124] LOD-4: Lane-level faces are used to describe the connectivity of various features and the shape of lanes;

[0125] Under the above model module and hierarchical division definitions, a multi-level conceptual hierarchical division diagram is obtained, such as Figure 3 shown.

[0126] 2. Fusion model design framework

[0127] The multi-level road network model is grouped and packaged according to logical properties, and the following features are set:

[0128] (1) Feature 1: The objects in the network feature package include rich multi-scale geometric features, and the multi-scale geometric expression is constructed through internal coding representation association;

[0129] (2) Feature 2: Events include static facilities and random dynamic time, and are manifested as multi-scale forms of points;

[0130] (3) Feature 3: Mobile objects include probe vehicle objects related to traffic behavior in the vector road network;

[0131] (4) The model design is based on query, path planning, location reference, events, network features, mobile objects, domain knowledge, and relationship modules;

[0132] (5) Domain knowledge includes prescriptive classification of vehicle types and road functions in probe car data and giving specific labels;

[0133] (6) Query includes internal query and mutual query between elements.

[0134] 2. Design of Geospatial Data Logical Model

[0135] For a multi-level logical model, each level is described geometrically, semantically, and topologically:

[0136] (1) LOD-0 abstracts the vector network at the most macro level, describing the simplest geometric form of the road with a point or a single line, which is located at the center of the object. LOD-O geometric objects are used to describe the road network layout at a macro level;

[0137] (2) LOD-1 does not have any semantic meaning in describing the shapes of objects divided in the vector network;

[0138] (3) LOD-2 divides the six categories of objects in the vector network into object exits according to the connection nodes between them, and adds semantic relationships by adding nodes between objects;

[0139] (4) LOD-3 abstracts the lane level of the vector network system, using double lines to describe the geometric form of the vector network data. The double lines are located at the center of the lane. For roads without a dedicated central isolation zone, they are still expressed as a single line.

[0140] (5) LOD-4 is the most detailed abstraction of the vector network system, describing lane-level abstract features, describing the geometric shape of the road with multiple lines, each line segment has directionality, and each line segment has different traffic semantic characteristics;

[0141] Express the correspondence between the feature entities in the data model of the multi-level real-time fusion model of the road network and the features in Oracle.

[0142] 3. Design of physical model of geographic spatial structure

[0143] The objects involved in the vector road network are divided into six categories of road network elements and one category of POI elements according to their main usage frequencies. The road network is segmented and the topological construction method of each layer of the network is described. The following data expression rules are set:

[0144] (1) Rule 1: For LOD0 networks, objects are located at their intersections and road segments are located at the logical intersections of streets. Objects are bidirectionally connected without semantic attribute constraints.

[0145] (2) Rule 2: Viaducts and roundabout buildings are represented as a point feature at the LOD0 level;

[0146] (3) Rule 3: To ensure the semantic association between multiple levels in the multi-level road network model, additional road segment nodes are added.

[0147] 2. Multi-source road network multi-level real-time fusion update model algorithm

[0148] Multi-source vector multi-level fusion includes three aspects: fusion of spatial entity geometric features, fusion of attribute information, and reconstruction of topological relationships. Multi-source vector multi-level fusion mainly involves the following three aspects:

[0149] (1) Coordination and integration: including the unification of spatial subsystem and map projection, the unification of classification and attribute coding, the unification of data format conversion and the unification of data model;

[0150] (2) Matching and fusion algorithm of POI data and vector linear data: matching is achieved based on geometric, semantic and topological relationships and taking some source data as the benchmark to achieve spatial matching of data;

[0151] (3) Detection vehicle data clustering analysis algorithm: realizes the clustering analysis process of detection vehicle data from scattered points to constrained points in the road network.

[0152] 1. Consistency, coordination and integration

[0153] 1. Consistency processing of multi-source spatial vector data

[0154] The process of multi-source spatial vector data consistency processing is as follows: Figure 4 The consistency processing is divided into the unification of spatial coordinates and map projection, the unification of classification and grading and attribute coding, the unification of classification and grading and attribute coding, and the conversion and unification of data formats.

[0155] 2. Unification of spatial coordinate system and map projection

[0156] Since the coordinate systems used by multiple sources are inconsistent, it is easy to cause offset problems when superimposing. There are three types of offsets caused by the coordinate system: one is the deviation caused by the inconsistency of the spatial coordinate system, one is the deviation caused by the projection coordinate system, and the other is the deviation caused by the inconsistency of both the spatial coordinates and the projection coordinates.

[0157] (1) The geographic coordinate system is a coordinate system based on the center point of the earth's ellipsoid (center of the sphere / geocenter), including WGS84, Xi'an 80 coordinate system, and National 2000 coordinate system. To ensure consistency, the geographic coordinate systems are unified.

[0158] (2) The projection coordinate system is a coordinate system that converts spatial coordinates into plane coordinates according to a certain mathematical projection method. It includes Mercator projection, Gauss-Krüger projection, and UTM projection. The latter two are generally variants of the former. In order to ensure the correctness of the map when displaying it on the plane coordinate system, the projection coordinate system is unified on the premise of unifying the spatial coordinate system.

[0159] This application uniformly adopts the WGS84 spatial coordinate system and Gauss Krüger projection coordinates. The elevation coordinate system adopts the 1985 National High-Level Datum. The conversion method adopts the form of combining code conversion with ArgGIS and ENVI image processing software. Large-scale spatial coordinate system conversion and projection coordinate conversion are unified by code, and single images are converted using image processing software.

[0160] 3. Unified combination of classification and attribute coding

[0161] On the basis of complying with OGC standards, the existing classification and grading systems are analyzed, their differences are compared, and a vector data fusion classification and grading system and attribute coding are constructed. Data classification and grading are important ways and contents of spatial data organization in GIS. The classification and grading of multi-source traffic vector road network data is the basis of multi-source traffic vector road network data coding. Coding is the ultimate embodiment of classification and grading. Through classification and grading and attribute coding, specific entities in the real world are uniquely stored in the computer, which is convenient for the organization and management of the database.

[0162] Based on the above data classification and grading rules and standards, the specific process is as follows Figure 5 For multi-source vector multi-level fusion, this application adopts different classification and attribute encoding methods for vector road network and POI data, where the classification and grading of vector road network follows the six-category five-level classification system in the road network model.

[0163] (1) Unification of multi-source road network classification and attribute coding

[0164] This application divides the data sources into two categories according to the need for classification, grading and attribute coding: road network data and POI data. The coding method will follow the following rules:

[0165] 1. Road network classification and attribute coding: Based on the design of a multi-level road vector model, the road network classification is classified according to six major elements and graded according to a five-level model. Its attribute coding will follow the coding method of the geographic spatial structure physical model.

[0166] 2. POI data classification and attribute coding: Classification is carried out according to a four-level classification system, and the number of each category is as follows Figure 6 shown.

[0167] 4. Unification of conversion of multi-source road network data formats

[0168] Based on the needs of existing data fusion, for the large amount of exploration vehicle data, this application adopts AE secondary development, and the development language is C# to convert the data format into shapefile files in batches. At the same time, for a single file, the application software FME that accepts multiple format conversions is used to convert the data into shapefile files, thereby realizing the unified integration of multi-source data.

[0169] Combined with the analysis of fused data, FME software and code are used to realize the conversion between data formats and the integration of multi-source data, such as Figure 7 shown.

[0170] 2. Real-time fusion and unification of multi-source road network multi-level models

[0171] 1. Multi-level real-time unified model design

[0172] The multi-source road network multi-level unified model abstracts the real-world geographic entities and their relationships, and constructs several windows for understanding geographic space bounded by geographic regions. The data region includes several data blocks, each data block includes several geographic element layers, and each element layer is relatively independent in data structure and organization. Data update, query, analysis and display operations are based on the element layer as the basic unit. The geographic element layer includes several geographic elements, which is the basic representation of geographic entities and representations.

[0173] The multi-source road network multi-level unified model adopts a hierarchical structure to organize data, and at the same time adopts a topological structure. Based on the overall target operation of the surface object, it realizes the integrated management of attribute and geometric data to meet the requirements of multi-source data integration.

[0174] The multi-source road network multi-level unified model takes geographic feature objects as the most basic feature units, and stores their attribute data and spatial data in an integrated manner. It adopts object-oriented data storage and GIS data operation throughout the entire process, which not only simplifies data operation, but also makes it more flexible to add and delete spatial objects and expand various new data types.

[0175] 2. Joint unification of real-time fusion models and algorithms

[0176] The effectiveness and accuracy of the data model are improved through a multi-source vector multi-level fusion algorithm. According to different levels, this application implements the fusion of different algorithms at different levels.

[0177] Extraction of LOD-O level: POI data is fused at the LOD-O level using the relationship between position and semantic proximity in model design, and the fused vector network data of POI data is realized at the LOD-0 level. The node and geometric center position are obtained based on the DLG data with the highest accuracy level to realize its 0-level fusion.

[0178] The LOD-1 level obtains its shape features according to the attribute table to achieve information extraction at level 1;

[0179] The LOD-2 level obtains the connectivity of its domain by intersecting objectified elements;

[0180] The LOD-3 level realizes the linearization of the probe car data by directional aggregation and topological reconstruction of the probe car data and performs probabilistic matching of the navigation data to achieve LOD-3 level fusion;

[0181] The LOD-4 level realizes the linearization of the probe car data through directional road network aggregation and topology reconstruction, and achieves LOD-4 level fusion through probabilistic matching with the navigation data.

[0182] 3. Adaptive matching and fusion of vector road network data

[0183] 1. POI matching fusion based on location and semantics

[0184] This application uses a POI matching fusion solution based on the combination of spatial location and attribute information to improve the accuracy of the POI fusion set. Figure 8 shown.

[0185] 2. Road network matching and fusion based on road network change patterns

[0186] The differences in spatial expression caused by changes in the geographical environment include three categories:

[0187] The first category is that in the same area, data with better currentness contain more spatial elements than data with worse currentness, which belongs to the 0:1 matching in spatial data matching;

[0188] The second category is that in the same area, data with better currentness include fewer spatial elements than data with worse currentness due to spatial changes, which belongs to 1:0 matching in spatial data matching;

[0189] Category 3: In the same region, data with better current status replaces spatial elements with poor current status, and the situation is complex and diverse;

[0190] According to the change mode of vector data, it satisfies the real-time and dynamic vector data fusion, and matches the map change at time t from a certain time point t of the road network. There are two modes: one is to match the road network data A at time t to the map change at time t. i The first is to use the map matching of road network data B, and the second is to use the real-time transmitted POL data to update the map matching at a certain moment.

[0191] 4. Rover matching and aggregation

[0192] The probe car data uses satellite navigation positioning technology. According to its sampling rate, it records its driving longitude and latitude, and time information at regular intervals (a few seconds). At the same time, the potential instantaneous speed longitude and latitude information is processed through analysis methods and model algorithms such as probe car clustering. The probe car text information is converted into spatial data format information to be associated with the road network data in time and space. The changes in road network information are detected in real time and dynamically, and the connectivity between lanes is discovered, thereby extracting the geometric, semantic, and topological information of the city's complex road network.

[0193] 1. Calculation of probe vehicle feature elements for multi-level fusion and update of road network

[0194] The definition and calculation of the probe car data involved in the multi-level fusion update of the road network include:

[0195] 1) Determine the traffic volume ps in the direction:

[0196] p s =(X d +Y e ) / (r d +r e ) Formula 1

[0197] in:

[0198] p s : Traffic volume in the direction to be tested at LOD-3 level, unit: vehicles / min;

[0199] X d : The number of vehicles traveling in the opposite direction of the probe vehicle (in the test direction) at the LOD-3 level, unit: vehicles / min;

[0200] Y e : At the LOD-3 level, when a vehicle to be tested is traveling in the undetermined direction, the number of vehicles that overtake the probe vehicle minus the number of vehicles overtaken by the probe vehicle (relative to the traffic volume in the direction measured by the test vehicle);

[0201] r d : The driving time of a vehicle to be tested at the LOD-3 level when driving in the direction and reverse direction of the measured traffic flow, unit: vehicle / min;

[0202] r s : At the LOD-3 level, the driving time of a vehicle to be tested traveling along the direction of the flow to be measured, unit: min.

[0203] 2) Average travel time T s :

[0204] R s =r s -Y s / ps Formula 2

[0205] Where: R s To determine the average travel time of a road section;

[0206] 3) Average vehicle speed U s :

[0207] U s =(J / R s )×60 Formula 3

[0208] Among them: U s is the average speed of the observed section, in km / h, J is the length of the observed section, in km;

[0209] When using equations 1 to 3 for calculation, X d , Y e , r d , R s The arithmetic mean is taken for calculation. In order to obtain a better correction fusion effect and avoid the influence of noise data on the fusion process, the data set is cleaned. The abnormality of vehicle positioning information is judged by obtaining the standard deviation of the first-order difference of the positioning information, and the activity range is judged by the variance of longitude and latitude to clean the directional outliers. By eliminating gross error data, the erroneous statistics of the data process caused by error data are avoided.

[0210] 2. Matching of probe car and road network information

[0211] The probe car data has a sub-meter positioning error. At the same time, in the urban environment, the influence of large-scale and dense urban buildings makes the satellite navigation positioning accuracy error even greater, resulting in the probe car data not being able to fall directly on the road range. At the same time, the urban road network is very dense and complex. In order to ensure that the probe car data is located on the correct road, map matching of the road network data and the probe car data is essential. Since the probe car text information only reflects the vehicle position (latitude and longitude coordinates) and time information, it is not directly associated with the road network data. Therefore, it is necessary to correct the positioning error of the probe car data based on the prior knowledge of high-precision lane-level road network data, correct the probe car data location information to the correct road network, and establish an association between the probe car text data information and the road network data information through the map matching algorithm.

[0212] 3. Calculation of hotspot paths of probe vehicles based on cluster analysis

[0213] On the basis of undirectedness, by adding the same-direction condition basis, we can find the probe vehicle trajectory data connected by the same density. By mining the distribution law of the probe vehicle data, we can generate directed edges according to the clustering class, complete the LOD-4 level data extraction, and obtain the topological structure diagram of the complex road network, so as to realize directed clustering such as Fig. 9As shown, (a) is a directed clustering effect diagram, and (b) is a topological structure diagram generated adaptively.

[0214] By analyzing the approximation and clustering characteristics of its spatial data, the correlation between the spatial distribution pattern of the probe car trajectory data and the traffic geographical characteristics is further analyzed. Finally, through cluster analysis, the spatial distribution pattern and traffic geographical correlation of the probe car data are excavated. By using approximate clustering analysis of directed data, there is no need for high-precision correction based on prior knowledge of lane-level road network data, so as to achieve the road network clustering analysis effect of the probe car data and realize the dynamic and timely fusion of road network and probe car data.

[0215] 3. Experiment and Analysis

[0216] (I) Constructing a 3D scene

[0217] The construction of the 3D scene includes the use of DEM and DOM to reconstruct the data background of the Beijing area, loading the road network data into the 3D background, and finally loading the probe car data. The purpose of the construction of the 3D scene is to verify the elevation accuracy of the road network and probe car data.

[0218] 1. 3D background modeling

[0219] The elevation data file of the 3D scene uses the DEM_1 file, which is converted into the .riff format through ENVI, namely DEM_1.riff. The texture map file uses the DEM.riff file of the same area. The terrain tool based on MFC is written by connecting to OSG to read DEM and generate a 3D scene model. Fig.10 As shown in the figure, it can be seen that the road surface is not very undulating.

[0220] 2. Loading rover data

[0221] First, the probe car data covering the area is selected, and the probe car data files that meet the conditions (the probe car data that have been converted to shp data) are imported into the database in time series. The drawn three-dimensional scene map is superimposed on the road network data to facilitate visual inspection of the changes in the two types of data due to time differences, such as Fig.11 , in preparation for verifying and analyzing the effectiveness, accuracy and efficiency of the fusion model.

[0222] 3. Loading of road network data.

[0223] Road network data can be roughly divided into national basic geographic information data and road network data. In terms of basic geographic data, only DLG data and vector data produced by vector data production departments at all levels are studied, while road network data includes a wider range of data, including data from various sources such as Tiandi Map, Tianxia Map, Amap, Baidu Map, etc. The data heterogeneity problems caused by different production requirements of these data are handled through the data consistency processing process, and multi-source road network data are included in the same geographic framework. Fig.11 This is the overlay effect diagram of the probe car data and the road network data under the unified geographic framework. It can be seen that under the unified geographic framework, there is no overlay error between the road network data loading and the probe car data.

[0224] 4. Simulate and verify the update

[0225] According to the obtained probe car data, a certain characteristic vector (density, direction, etc.) of the road section is extracted, and the linear features of the traffic vector and the structural relationship are constructed. Then, the vector is fused and associated, and compared with the actual road network conditions and DOM data of each node to verify the correctness and accuracy of the data fusion effect. After the experimental data is preprocessed, all the data are superimposed together, and the simulation and fusion results are verified by performing various statistical analyses on the probe car data on the road network. Among the POI points, 760 objects in the OSM map and Baidu map are selected. According to the rules, each POI data uniquely represents a geographic entity, and there is no duplication between them; in the road network data, the LOD-0 level is obtained by map digitization, which ensures the geometric connectivity between objects, the LOD-1 level is extended from the point structure to the surface structure by the attribute, the LOD-2 level obtains the surface structure with an exit by the flow direction of the probe car data, the LOD-3 level detects updates by calculating the number of opposite driving of the probe car data, and the LOD-4 is detected by the data (density) of the same direction driving of the probe car data.

[0226] Results of multi-source POI data fusion: Four sets of POI data were selected for testing. These POI sets have different degrees of overlap. The positive example is defined as two sets of interest points with matching items, and the negative example is defined as no matching items. After being processed by the POI data fusion algorithm and visually inspected, the two sets of data showed good fusion effects after being processed by the point fusion algorithm. Only a few failed to match correctly due to large coordinate errors between maps.

[0227] Road network data fusion results: The experiment verifies and analyzes the effect of data fusion and update when multi-source data is loaded into a multi-level data fusion model. Fig.12As shown in the figure, the dotted line is the 2021 AutoNavi data, the dotted data is the 2020 probe car data and the linear data connected and vectorized by the dotted data, and the scattered line is the connection line between AutoNavi and the vectorized points of the same name. Based on the data line characteristics of the multi-source road network data, the probe car data is fused, and the multi-source vector data is superimposed and information mined. According to the fusion model design, it is loaded into the database in sequence to further obtain multi-level road network information data. As can be seen from the figure, after the probe car data is corrected by the lane-level road network data, the probe car data can be topologically fused in real time at the lane level, meeting the needs of multi-level visualization.

[0228] (II) Effectiveness

[0229] Through the above experimental analysis, POI data, probe car data and road network data are loaded into the multi-level road network model designed in this application, and the fusion of multi-source vector data is realized. The experiment shows that the multi-source vector multi-level fusion achieves a good fusion effect, thus verifying the effectiveness of the road network model and the multi-source traffic vector road network data fusion method.

[0230] (III) Accuracy

[0231] The most significant feature of the multi-source traffic vector road network data fusion tool is that it incorporates sub-meter lane data into the fusion range, which can meet the needs of different levels and is significantly superior to other data model fusion tools in terms of accuracy.

[0232] The multi-source traffic vector network data fusion tool proposed in this application is based on the CityGML standard and takes the multi-level and multi-source traffic vector network data model as the core for the purpose of vector data fusion. It basically achieves the goal of ensuring the use of lane-level data while meeting the requirements of effectiveness and efficiency.

Claims

1. A multi-source traffic GIS road network multi-level real-time fusion update method, characterized in that: First, a multi-level, object-oriented, lane-level road model is set up based on the OGC standard; second, a five-layer data fusion model based on CityGML is proposed by adopting the high-precision fusion of vector network and POI data, and a multi-level real-time fusion update method covering the transportation GIS road network is realized; First, a multi-source vector road network multi-level real-time fusion model: the first is the geospatial cognitive conceptual model, including the five-level hierarchical division of the fusion model and the fusion model design framework; the second is the geospatial data logical model; and the third is the geospatial structure physical model. A multi-source vector road network multi-level real-time fusion model based on the CityGML standard is proposed. Its classification definition is improved at the detail level, and a multi-level real-time fusion framework is designed. Based on the abstract expression method of different elements covered by vector data and the geometric, semantic, and topological relationships between elements and between different levels of elements, a three-layer architecture design of the data model is realized. Second, multi-source road network multi-level real-time fusion update model algorithm: First, consistency coordination fusion, including multi-source spatial vector data consistency processing, unified combination of spatial coordinate system and map projection, unified combination of classification and attribute coding, and conversion and unification of multi-source road network data formats; second, real-time fusion and unification of multi-source road network multi-level models, including multi-level real-time unified model design, real-time fusion model and algorithm combination; third, vector road network data adaptive matching fusion, including POI matching fusion based on location and semantics, road network matching fusion based on road network change mode; fourth, probe car matching and aggregation, including calculation of probe car feature elements for multi-level fusion update of road network, matching of probe car and road network information, and calculation of probe car hotspot path based on cluster analysis; A unified geographic framework for real-time fusion and updating of multi-level road networks is proposed to integrate the early data involved in multi-source data fusion and matching and analyze the fusion of multi-source traffic vector road network data in the later stage, and cluster analysis of POI data, probe car data, road network, and digital line map is realized to achieve high-precision matching and fusion within and between multi-source data types, and meet the static and dynamic update fusion of vector data. Calculation of probe vehicle characteristic elements for multi-level fusion and update of road network: The definition and calculation of probe vehicle data involved in multi-level fusion and update of road network include: 1) Determine the traffic volume in the direction ps: p s =(X d +Y e ) / (r d +r s ) Formula 1 in: p s : Traffic volume in the direction to be tested at LOD-3 level, unit: vehicles / min; X d : The number of vehicles traveling in the opposite direction of the probe vehicle at the LOD-3 level, unit: vehicles / min; Y e : At the LOD-3 level, when a vehicle to be tested is traveling in the direction to be tested, the number of vehicles that overtake the probe vehicle minus the number of vehicles overtaken by the probe vehicle; r d : The driving time of a vehicle to be tested at the LOD-3 level when driving in the direction and reverse direction of the measured traffic flow, unit: vehicle / min; r s : At the LOD-3 level, the driving time of a vehicle to be tested traveling in the direction of the flow to be tested, unit: min; 2) Average travel time T s : R s =r s -Y e / p s Formula 2 Where: R s To determine the average travel time of a road section; 3) Average vehicle speed U s : U s =(J / R s )×60 Formula 3 Among them: U s is the average speed of the observed section, in km / h, J is the length of the observed section, in km; When using equations 1 to 3 for calculation, X d , Y e , r d , R s The arithmetic mean is taken for calculation. The data set is cleaned and the abnormality of vehicle positioning information is judged by taking the standard deviation of the first-order difference of the positioning information. The activity range is judged by the variance of the longitude and latitude to clean the directional outliers. By eliminating the gross error data, the erroneous statistics of the data process caused by the error data are avoided.

2. According to the multi-source transportation GIS road network multi-level real-time fusion update method of claim 1, it is characterized in that: The five-level hierarchical division of the fusion model: vector geographic data mainly involves various types of data on the traffic road network, which are generally summarized as six basic elements: lanes, intersections, tunnels, viaducts, roundabouts, and squares. The following hierarchical division rules are formulated: LOD-0: Basic description of the location of various features using points or lines; LOD-1: Shape description is performed based on various features; LOD-2: Use surfaces with interfaces to describe the connectivity of various features; LOD-3: Use lane-level surfaces to describe the connectivity of various features and lane shapes; LOD-4: Lane-level faces are used to describe the connectivity of various features and the shape of lanes; Based on the definition of the above model hierarchy, a multi-level conceptual hierarchy diagram is obtained.

3. According to the multi-source transportation GIS road network multi-level real-time fusion update method of claim 1, it is characterized in that: Fusion model design framework: The multi-level road network model is grouped and packaged according to logical properties, and the following features are set: (1) Feature 1: The objects in the network feature package include rich multi-scale geometric features, and the multi-scale geometric expression is constructed through internal encoding representation association; (2) Feature 2: Events include static facilities and random dynamic time, and are manifested as multi-scale forms of points; (3) Feature 3: Mobile objects include probe vehicle objects related to traffic behavior in the vector road network; (4) The model design is based on query, path planning, location reference, events, network features, mobile objects, domain knowledge, and relationship modules; (5) Domain knowledge includes prescriptive classification of vehicle types and road functions in probe car data and giving specific labels; (6) Query includes internal query and mutual query between elements.

4. According to claim 1, the multi-source transportation GIS road network multi-level real-time fusion update method is characterized in that: Geospatial data logical model design: Each level of the multi-level logical model is described in terms of geometry, semantics, and topology: (1) LOD-0 abstracts the vector network at the most macro level, describing the simplest geometric form of the road with a point or a single line, which is located at the center of the object. LOD-O geometric objects are used to describe the road network layout at a macro level; (2) LOD-1 does not have any semantic meaning in describing the shapes of objects divided in the vector network; (3) LOD-2 divides the six categories of objects in the vector network into object exits according to the connection nodes between each other, and adds semantic relationships by adding nodes between objects; (4) LOD-3 abstracts the lane level of the vector network system, using double lines to describe the geometric form of the vector network data. The double lines are located at the center of the lane. For roads without a dedicated central isolation zone, they are still expressed as a single line. (5) LOD-4 is the most detailed abstraction of the vector network system, describing lane-level abstract features, describing the geometric shape of the road with multiple lines, each line segment has directionality, and each line segment has different traffic semantic characteristics; Design of physical model of geographic spatial structure: objects involved in the vector road network are divided into six categories of road network elements and one category of POI elements according to their main usage frequency, the road network is segmented and the topological construction method of each layer of network is described, and the following data expression rules are set: (1) Rule 1: For LOD0 networks, objects are located at their intersections and road segments are located at the logical intersections of streets. Objects are bidirectionally connected without semantic attribute constraints. (2) Rule 2: Viaducts and roundabout buildings are represented as point features at the LOD0 level; (3) Rule 3: To ensure the semantic association between multiple levels in the multi-level road network model, additional road segment nodes are added.

5. According to the multi-source transportation GIS road network multi-level real-time fusion update method of claim 1, it is characterized in that: Consistency processing of multi-source spatial vector data: divided into the unification of spatial coordinates and map projection, the unification of classification and grading and attribute coding, the unification of classification and grading and attribute coding, and the conversion and unification of data formats; Unification of spatial coordinate system and map projection: WGS84 spatial coordinate system and Gauss Krüger projection coordinates are uniformly adopted, the elevation coordinate system adopts the 1985 National High-Level Datum, the conversion method adopts the form of code conversion combined with ArgGIS and ENVI image processing software, large-scale spatial coordinate system conversion and projection coordinate conversion are unified by code, and single image is converted by image processing software.

6. The multi-source transportation GIS road network multi-level real-time fusion update method according to claim 2 is characterized in that: Unified combination of classification and attribute coding: Different classification and attribute coding methods are used for vector road network and POI data. The classification and grading of vector road network is based on the six-category five-layer classification system in the road network model. Unification of multi-source road network classification and attribute coding: According to the data sources that need classification, classification and attribute coding, they are unified into two categories: road network data and POI data. The coding method will follow the following rules: 1) Road network classification and attribute coding: Based on the design of a multi-level road vector model, the road network classification is classified according to six major elements and graded according to a five-level model. Its attribute coding will follow the coding method of the geographic spatial structure physical model; 2) POI data classification and attribute coding: Classification is carried out according to a four-level classification system.

7. The multi-source transportation GIS road network multi-level real-time fusion update method according to claim 1 is characterized in that: Multi-level real-time unified model design: abstract the real-world geographic entities and their relationships, and build several windows for understanding geographic space with geographic regions as boundaries. The data region includes several data blocks, each data block includes several geographic element layers, and each element layer is relatively independent in data structure and organization. Data update, query, analysis and display operations are based on the element layer as the basic unit; The multi-source road network multi-level unified model adopts a hierarchical structure to organize data, and adopts a topological structure to achieve integrated management of attribute and geometric data based on the overall target operation of the surface object, meeting the requirements of multi-source data integration; The multi-source road network multi-level unified model takes geographic feature objects as the most basic feature unit, and stores its attribute data and spatial data in an integrated manner, using the entire process of object-oriented data storage and GIS data operation; Joint unification of real-time fusion models and algorithms: Fusion of different algorithms at different levels: Extraction of LOD-O level: POI data is fused at the LOD-O level using the relationship between position and semantic proximity in model design, and the fused vector network data of POI data is realized at the LOD-0 level. The node and geometric center position are obtained based on the DLG data with the highest accuracy level to realize its 0-level fusion; The LOD-1 level obtains its shape characteristics based on the attribute table. Realize information extraction at level 1; The LOD-2 level obtains the connectivity of its domain by intersecting objectified elements; The LOD-3 level realizes the linearization of the probe car data by directional aggregation and topological reconstruction of the probe car data and performs probabilistic matching of the navigation data to achieve LOD-3 level fusion; The LOD-4 level realizes the linearization of the probe car data through directional road network aggregation and topology reconstruction, and achieves LOD-4 level fusion through probabilistic matching with the navigation data.

8. The multi-source transportation GIS road network multi-level real-time fusion update method according to claim 1 is characterized in that: POI matching fusion based on location and semantics: The accuracy of POI fusion set is improved by combining spatial location and attribute information; Road network matching fusion based on road network change patterns: includes three categories: The first category is that in the same area, data with good currentness contain more spatial elements than data with poor currentness, which belongs to the 0:1 matching in spatial data matching; The second category is that in the same area, data with good currentness include fewer spatial elements than data with poor currentness due to spatial changes, which belongs to 1:0 matching in spatial data matching; The third category: in the same area, data with good current situation replaces spatial elements with poor current situation, and the situation is complex and diverse; According to the change mode of vector data, it satisfies the real-time and dynamic vector data fusion, and matches the map change at time t from a certain time point t of the road network. There are two modes: one is to match the road network data A at time t to the map change at time t. i The first is to use the map matching of road network data B, and the second is to use the real-time transmitted POL data to update the map matching at a certain moment.

9. The multi-source transportation GIS road network multi-level real-time fusion update method according to claim 1 is characterized in that: Probe car matching and aggregation: The probe car data uses satellite navigation positioning technology. According to its sampling rate, it records its driving longitude, latitude and time information at regular intervals. At the same time, the potential instantaneous speed longitude and latitude information is processed through the probe car clustering analysis method and model algorithm. By converting the probe car text information into spatial data format information, it is associated with the road network data in time and space, and the changes in road network information are detected in real time and dynamically, and the connection relationship between lanes is discovered, thereby extracting the geometric, semantic and topological information of the complex urban road network.

10. The multi-source transportation GIS road network multi-level real-time fusion update method according to claim 9 is characterized in that: Matching of probe car and road network information: Based on the prior knowledge of high-precision lane-level road network data, the positioning error of the probe car data is corrected, the location information of the probe car data is corrected to the correct road network, and the text data information of the probe car is associated with the road network data information through the map matching algorithm; Calculation of hot spots of probe vehicles based on cluster analysis: On the basis of undirected, by adding the same-direction condition basis, we can find the probe vehicle trajectory data connected with the same density, and by mining the distribution law of the probe vehicle data, we can generate directed edges according to the clustering class, complete the data extraction of LOD-4 level, and obtain the topological structure diagram of the complex road network; By analyzing the approximation and clustering characteristics of its spatial data, we further analyze the correlation between the spatial distribution pattern of the probe car trajectory data and the traffic geographical characteristics. Finally, through cluster analysis, we can explore the spatial distribution pattern and traffic geographical correlation of the probe car data. By using approximate cluster analysis of directed data, we can achieve the road network clustering analysis effect of the probe car data and realize the dynamic and timely fusion of road network and probe car data.

Images