A method and system for integrating multi-source and multi-structure spatial data

Abstract

The present invention discloses a method and system for fusing multi-source and multi-structure spatial data, which realizes the efficient fusion of low spatial precision business data and high-precision map data. First, the present invention designs a set of complete mileage pile coding rules that are compatible with the original system and oriented to high-precision maps, and an algorithm implemented using high-precision map data, which meets the needs of auxiliary spatial positioning and rapid calculation of road distances. Secondly, a processing method is proposed to effectively integrate the two types of data, using feature extraction and vector calculation methods, through quantitative distance analysis and shortest distance matching, to achieve efficient and automatic fusion. Finally, the fusion results are adapted, and the quality of the results is inspected by random sampling, circular voting, etc., to meet the accuracy requirements of the production system. The present invention processes and fuses the two types of data through the above series of technical means, realizes the associated fusion of business data and high-precision map data, can rely less on manual work and form reliable fusion results.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to a fusion method and system based on multi-source and multi-structure spatial data. Background Art

[0002] The construction of spatial data used by business systems in the field of transportation engineering is mostly based on original data from multiple sources and structures. The sources of these data include highway design, engineering construction, GPS point mapping, high-precision acquisition (based on LiDAR), aerial photography, and annotation data used by highway operation and maintenance systems. All kinds of (spatial-related) data are produced based on their own business, and there are differences in the technical specifications, production and storage methods, accuracy, coordinate systems, etc. In particular:

[0003] 1) Road engineering design and construction focuses on the use of the geodetic coordinate system, paying more attention to the construction technology of the road and the more accurate local geometry;

[0004] 2) Road mapping and high-precision maps focus on using the geographic coordinate system, deconstructing and organizing map data from a more intuitive perspective;

[0005] 3) The business system of the transportation industry focuses on the use of the mileage system for the positioning of spatial facilities and equipment. The system has very rough definitions and location descriptions of the geometric features of entities, and the spatial errors are usually lower than the level of standard precision maps. Some even have no spatial expression: ① It is only the spatial position expression of the physical mileage system, which is not measurable and calculable; ② The spatial position accuracy of the mileage system is low; ③ The coverage is incomplete, such as: some roads have mileage systems on only one side of the up and down roads, or data is only collected on one side; the mileage systems of some roads (the embodiment of spatial position in the data) are located in the median strip, and the up and down routes share them; ④ There is ambiguity. Due to factors such as changes in road names and reconstruction and expansion, there are multiple sets of pile number systems in some road sections, which cause ambiguity in expression and use.

[0006] The fusion of spatial data from different sources and structures can realize the integrated sharing, comprehensive application and value mining of data. However, the data under each system cannot eliminate ambiguity from each other due to differences in data structure, and an effective integration method has not yet been formed. Effective business map fusion cannot be achieved in business applications, which restricts the comprehensive application of data and the improvement of its utility. Summary of the invention

[0007] The first object of the present invention is to overcome the problems existing in the above-mentioned prior art and provide a method for fusing multi-source and multi-structure spatial data.

[0008] To achieve the above object, the present invention adopts the following scheme:

[0009] A method for fusing multi-source multi-component spatial data, wherein the multi-source multi-component spatial data includes business data with spatial attribute information and map data with high-precision spatial location information, the method comprising:

[0010] According to the spatial position of the physical kilometer piles in the high-precision map data and the road reference line, the projection kilometer piles on the road reference line are obtained to form a kilometer pile benchmark network;

[0011] The kilometer pile benchmark network is interpolated at meter level along the road reference line in the longitudinal direction of the road to obtain the meter-level mileage pile benchmark network. Each meter-level mileage pile is projected along the center line of each lane in the transverse direction of the road to obtain the lane-by-lane meter-level mileage pile benchmark network as the mileage pile reference system.

[0012] Based on the mileage pile reference system, the elements in the high-precision map data and business data are parameterized by mileage piles;

[0013] Merge the business data and high-precision map data into the same data set, select feature fields for each element record in the merged data set, calculate feature vector values ​​using embedding technology, and use them for similarity analysis and similarity distance calculation between element data, and match high-precision map data elements and business data elements according to the shortest distance to achieve the association and fusion of business data and high-precision map data; the feature fields at least include fields containing business meanings and fields containing positioning information;

[0014] Among them, each mileage post in the meter-level mileage post benchmark network is assigned a code, and the code composition includes: a route code describing the route of the road where the mileage post is located, a road code describing the type and direction of the road where the mileage post is located, a mileage code describing the mileage where the mileage post is located, and a lane code describing the lane where the mileage post is located; the data set includes mileage post parameterized data obtained by parameterizing the elements in the high-precision map data and business data based on the mileage post reference system.

[0015] As a preferred implementation, if there are one or more missing points between two adjacent kilometer posts, the kilometer posts projected before and after the missing points are used to perform average interpolation according to the number of missing points along the road reference line.

[0016] As a preferred implementation, the method further includes adjusting the positions of the projected kilometer piles along the road reference line so that the distance error between adjacent projected kilometer piles is globally minimized, and constructing a kilometer pile reference network using the adjusted projected kilometer piles as reference points of the pile number reference system.

[0017] As a preferred implementation, the route code includes a character bit representing a road grade and a character bit representing a road number;

[0018] The road code includes a character bit indicating the road type and a character bit indicating the driving direction; the road type includes a main road and a ramp, which are distinguished by different letters, and the driving direction includes forward and reverse, which are distinguished by different numbers;

[0019] The mileage code includes a character position indicating kilometers, a character position indicating a specific kilometer value, a character position indicating meters, a character position indicating a specific meter value and / or a character position indicating a decimal point and a character position indicating a centimeter value;

[0020] The lane code includes a separator character bit and a character bit representing a lane number. The lane number is: corresponding to the lane longitudinally, numbered incrementally from the inside to the outside by integers; when the value is 0, there is no lane information.

[0021] As a preferred implementation, the road code also includes character bits for road section disambiguation and expansion, where numbers are used when the road section where the mileage post is located is a main line, and letters are used for ramps.

[0022] As a preferred implementation, the method further includes establishing a conversion model between the mileage pile reference network and the geographic space position reference system;

[0023] For the conversion of mileage posts to spatial coordinates, the geometric coordinates are queried through the mileage post codes and converted to any geographic coordinate system;

[0024] For the conversion of geospatial coordinates to mileage posts, the geospatial coordinates are converted to the coordinate system corresponding to the geometric coordinates of the mileage post, and then the nearest mileage post is found in the mileage post reference network to convert the spatial coordinates to the mileage post.

[0025] As a preferred implementation, the mileage post parameterization of the element is to associate and map the point, line, surface, and body elements to the mileage post, so that the element has its own spatial coordinate value and mileage post coding value, and the element has mileage post parameterization data composed of its own spatial coordinate value and mileage post coding value.

[0026] As a preferred implementation, the merging of business data and high-precision map data into the same data set includes:

[0027] UUID is compiled for all feature data records in business data and high-precision map data;

[0028] Based on the unified data classification and coding standards, all element data are classified and coded, and based on the mileage reference system, all element data are parameterized by mileage; the classification coding and mileage parameterization data are included in the data set;

[0029] For the common information in the business data and high-precision map data, further field-level merging is performed; for the non-common information, they are recorded separately.

[0030] As a preferred implementation, the method further includes, after the data are matched, verifying the matched data by manual verification and / or circular voting to complete the data fusion.

[0031] Another object of the present invention is to provide a fusion system for multi-source and multi-component spatial data, comprising:

[0032] A data acquisition module is used to acquire business data with spatial attribute information and map data with high-precision spatial location information;

[0033] The mileage pile reference system construction module obtains the projected mileage piles on the road reference line according to the spatial position of the physical mileage piles in the high-precision map data and the road reference line, forming a mileage pile reference network; the mileage pile reference network is interpolated at the meter level along the road reference line in the longitudinal direction of the road to obtain a meter-level mileage pile reference network, and each meter-level mileage pile is projected along the center line of each lane in the transverse direction of the road to obtain a lane-by-lane meter-level mileage pile reference network as a mileage pile reference system;

[0034] Each milepost in the meter-level milepost reference network is assigned a code, the code comprising: a route code describing the route of the road where the milepost is located, a road code describing the type and direction of the road where the milepost is located, a mileage code describing the mileage where the milepost is located, and a lane code describing the lane where the milepost is located;

[0035] A data merging module, merging business data and high-precision map data into the same data set; the data set includes mileage post parameterized data obtained by parameterizing the elements in the high-precision map data and business data based on the mileage post reference system;

[0036] The data fusion module selects characteristic fields for each element record in the merged data set, calculates the characteristic vector value using embedding technology, and uses it for similarity analysis and similarity distance calculation between element data, and matches the high-precision map data elements and business data elements according to the shortest distance to achieve the association and fusion of business data and high-precision map data; the characteristic fields include at least a field containing business meaning and a field containing positioning information.

[0037] The present invention has the following beneficial effects:

[0038] 1) Through a series of algorithms with adjustable parameters, the fusion of multi-source and multi-structured spatial data (divided into business data and high-precision map data) is formed, which relies less on manual work and forms reliable fusion results. It provides a new solution to the problem of fusion of business data and high-precision map data, and provides a technical implementation that can balance efficiency and quality for data fusion of business systems.

[0039] 2) The present invention constructs a mileage post reference system as a bridge connecting business data and high-precision map data, so that business data and high-precision map data can exchange position expressions, realize the integration of business data and high-precision map data, and design a mileage post coding system based on high-precision maps, which is compatible with existing coding and covers the undefined ramp areas in traditional business, greatly improving spatial accuracy, horizontally expanding to the lane level, and vertically improving to the meter level and even higher. This coding system realizes the requirements of high spatial accuracy lane-level road network coverage and position distance calculability of mileage post coding, thereby achieving the purpose of replacing geographic coordinates, making it easier for business system operators to intuitively understand the spatial positioning of entities.

[0040] 3) Combine the current national and industry standards to build a unified classification and coding of spatial related data to facilitate automatic connection and fusion of data at the algorithm level.

[0041] 4) Use embedding technology to perform vectorized distance analysis on two types of data (data to be calibrated and reference data), so as to achieve automatic pre-calibration of business data and collected data combined with mileage post coding by fuzzy matching, replacing traditional spatial range retrieval. It is more suitable for high-precision map data elements with complex structures to be matched, improving efficiency and hit rate.

[0042] 5) Use manual verification and round-robin voting to ensure the final availability of the algorithm matching fusion results. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is the technical roadmap of the present invention. DETAILED DESCRIPTION

[0044] The technical solution of the present invention is further described below in conjunction with specific embodiments.

[0045] The high-precision map referred to in the present invention has a geographic geometric accuracy of: the absolute error is usually within 10 cm, and the relative error is below 1%.

[0046] The multi-source and multi-structure spatial data involved in this embodiment are divided into two types of data: 1) business data with spatial attribute information. The classification and definition of this type of data are based on the developed business model. The accuracy of spatial information is relatively low. It is usually the longitude and latitude coordinates attached to the business entity model. Its characteristics are a. low spatial position accuracy and b. abstract spatial geometric information (usually replaced by points); 2) map data with high-precision spatial position information. The classification of this type of data is usually based on the visualization characteristics of the spatial position, attachment relationship, geometric style, material composition, etc. of the entity object during data collection. The spatial position accuracy is usually within the decimeter level, and the geometric description of its entity is detailed and meticulous.

[0047] 1) The source of the first category of data is the underlying spatial master data of each business system, precision maps, and derived data or information in the business system; 2) The source of the second category of data is the data collected and processed by high-precision maps. The collected data includes: lidar point cloud, satellite and aerial images, vehicle-mounted cameras and roadside images, etc. The processed data mainly includes structured data, mainly various types of quantified elements, including spatial information and related attribute information obtained by surveying and mapping.

[0048] The embodiment realizes the fusion of the above two types of data through a series of technical means, and obtains fused data with high spatial accuracy and business definition based on business data.

[0049] Example 1

[0050] The process of realizing data fusion is as follows:

[0051] S1 Establish a high-precision mileage post network; Based on the current highway mileage post system, design (S1A) and construct (S1B) a set of universal location coding rules and spatial positioning methods (S1C) for the highway system, including:

[0052] S1A High-precision Mileage Pile Coding System Design

[0053] The total length of the high-precision mileage post code is 22 bits, which is composed of 4 groups of codes. Bits 1-5 are route codes, which describe the route of the road where the mileage post is located; bits 6-8 are road codes, which describe the type and direction of the road where the mileage post is located; bits 9-20 are mileage codes, which describe the mileage where the mileage post is located; bits 21-22 are lane codes, which describe the lane where the mileage post is located, as follows:

[0054] 1) Route codes are compatible with national standards. The first digit is a letter, indicating the road grade (such as G and S, indicating national highways and provincial highways, respectively), and the last four digits are numbers, indicating the road number. When a ramp appears at a toll station, interchange, or hub, its route code follows the main line code of the upstream road (i.e., the main line of the first divergence along the road reference line in the reverse direction). If there is no main line, the main line of the first confluence downstream is found according to the same rules, and its code is used as the route code of the section.

[0055] 2) Road code, used to distinguish and display the road type (including main line and ramp) and driving direction (direction or reverse) of the mileage post. The first digit is a letter, which represents the road type. M represents the main line, and R / T represents the ramp (R represents the ramp at the hub, and T represents the ramp at the toll station interchange); the second digit is a number, which indicates the direction of travel of the road route where the pile number is located. The driving direction increases in the direction of the mileage post as upward (indicated by 1), and increases in the direction of travel against the mileage post as downward (indicated by 0); the third digit is a number or letter, which is used for road section disambiguation and expansion. When the road section where the mileage post is located is the main line, a number is used. The usual scenario is: for separated roads, numbers are used and numbered from the inside to the outside in ascending order. For ramps, letters are used and numbered from the main line divergence point to the downstream in sequence according to the "along the left side and depth first" rule until all ramps merge into the main line. The left-side and depth-first rules give priority to coding the roads on the left side of the driving direction and code them according to the driving route until they cannot be coded (usually when they reach the downstream main line downstream), and then go back in reverse to the nearest diversion point to retrieve the mileage post codes on the remaining uncoded roads until all are completed.

[0056] 3) Mileage code: the first digit is the character K, indicating kilometers; the next four digits are numbers, indicating the specific kilometer value; the sixth digit is the "+" character, indicating meters; the 7th to 9th digits are numbers, indicating the meter value; the 10th digit is the ".", which is the decimal point symbol; the next two digits are numbers, indicating the centimeter value. If the mileage post accuracy only needs to be to meters, the last three digits can be left blank or not used.

[0057] 4) Lane coding: the first digit is the separator “|”, and the second digit is a numerical value, which corresponds to the lane number in the longitudinal direction and increases by integers from the inside to the outside. When the value is 0, there is no lane information, which can correspond to the position of the road reference line in geometric expression.

[0058] S1B high-precision mileage pile system construction;

[0059] Based on the coding rules in S1A, a high-precision mileage measurement system is constructed using high-precision collected map data. The main strategies are as follows:

[0060] 1) According to the spatial position of the physical kilometer mileage posts (kilometer posts) in the high-precision map data and the road reference line, the projected kilometer posts on the road reference line are obtained; if there are one or more missing points between two adjacent kilometer posts, the projected kilometer posts before and after the missing points are used, and average interpolation is performed along the road reference line according to the number of missing points;

[0061] Assume that in a certain direction of a road R with a length of m kilometers, the high-precision map collects n kilometer stakes (n≤m); any kilometer stake is expressed as S(x, y, z) i , i∈n, x, y, z are its coordinates and elevation in the CGCS2000 coordinate system. The elevation can be obtained based on high-precision map data. i is the mileage of the pile (i.e., kilometers, which can be obtained according to the mileage number in the code). First, this process projects all S onto the road reference line collected by its high-precision map, so that the kilometer pile S'(x', y', z') projected onto the road reference line i to S(x, y, z) i The distance from any point on the reference line to S(x, y, z) i Secondly, average interpolation is performed on the missing kilometer stakes along the road reference line to obtain the projection points S'(x', y', z') of all kilometer stakes on the road reference line. j , j∈m.

[0062] 2) Adjust the position of the projected kilometer piles along the road reference line so that the distance error between the adjusted adjacent projected kilometer piles in each direction on each road is globally minimized, and use the adjusted kilometer piles as the reference points of the measurable pile number reference system.

[0063] This step can be solved using a heuristic algorithm to find a good enough solution within a reasonable time. There are many heuristic algorithms. In implementation, you can choose an existing method to calculate according to the actual situation. This application will not describe it in detail. The following only uses the annealing algorithm as an example to illustrate the idea:

[0064] Assume D j is S'(x', y', z') j The length from the start of the road (along the road reference line), then d j is S'(x', y', z') j with S'(x', y', z') j-1 Distance on the road reference line (unit: m) d j =D j -D j-1 The objective function is to minimize the global error of adjacent kilometer piles on the route, which is defined as:

[0065]

[0066] For each kilometer post, only its position adjustment along the road reference line is considered to minimize the objective function. The adjusted positions of each kilometer post must be arranged in order, that is: D j-1 ≤D j ≤D j+1 .

[0067] Suppose the projected and interpolated kilometer post is the initial solution, the above objective function is the energy function, and the adjustment of the kilometer post position is the state transfer. In each iteration, a kilometer post can be randomly selected and randomly displaced a small distance forward and backward along the road reference line. The global minimum is taken as the goal. Through iterative adjustment, the energy function is gradually reduced, so as to obtain the final adjusted kilometer post position, that is, the benchmark kilometer post.

[0068] 3) Check the generated benchmark kilometer piles. If a large error is found, j , you can re-solve or check whether there is a large error in the original high-precision map collection kilometer stakes. For the ramp part along the road line, the kilometer stakes are calculated and placed along the road reference line of the ramp according to the rules of S1A.

[0069] 4) Repeat the above steps (i.e., steps 1)-3) in S1B) for all directions of all roads in the road network, and finally form a kilometer pile reference network covering all road reference lines;

[0070] 5) Using the generated kilometer-level milepost reference network, average interpolation is performed along the road reference line in the longitudinal direction at the meter level to obtain the meter-level milepost reference network, so that the distance between each meter-level milepost is as close to 1m as possible to achieve the spatial position definition of the road-level meter-level milepost.

[0071] 6) Based on each meter-level milepost, the center lines of each lane of the road are projected again in the transverse direction of the road to obtain a lane-divided meter-level milepost benchmark network.

[0072] 7) All the mileage posts currently obtained are encoded according to the S1A rule to obtain lane-divided meter-level mileage posts with mileage codes and spatial coordinates.

[0073] S1C Conversion between mileage post system and geographic spatial position reference system

[0074] 1) Convert mileage piles to spatial coordinates. Since all mileage piles in the mileage pile benchmark network have CGCS2000 coordinates, the geometric coordinates of the mileage piles can be directly queried through the mileage pile codes and further converted to any geographic coordinate system.

[0075] 2) Spatial coordinates to mileage posts. For any geographic spatial coordinates, they must first be converted to CGCS2000 coordinates, and then the nearest mileage post is found in the mileage post benchmark network through a heuristic algorithm, and its mileage value is calculated to achieve the conversion of spatial coordinates to mileage posts. Secondly, the buffer zone is checked to determine whether the coordinates are within the road range. If not, a null value is returned. Finally, if it is within the road range, the nearest neighboring point found is used as the returned mileage post value.

[0076] Based on the generated mileage post reference system, S2 parameterizes the mileage post of various elements collected by high-precision maps, so that they have both spatial coordinate values ​​(longitude, latitude, elevation) and mileage post coding values ​​to achieve the association between spatial coordinate elements and mileage post elements.

[0077] According to geometric characteristics, the elements of high-precision maps can be divided into four categories: points, lines, surfaces, and volumes. For each type of element, a different parameterization method is used to achieve the associative mapping between spatial coordinate elements and mileage post elements.

[0078] 1) For high-precision map features with point geometry, the milepost expression of the point feature is obtained based on the results of S1C. Usually, the milepost closest to the feature is used, and the geometric distance from the feature to the milepost is recorded.

[0079] 2) For high-precision map elements whose geometric shape is a line (or line string), first extract their key points and key geometric attributes. Key points can include start and end points, center points, etc., and key geometric attributes can include length, start and end point distance, etc. For the extracted key point elements, obtain the mileage expression of the point elements based on the results of S1C, and record the relevant geometric attributes according to business needs.

[0080] 3) For high-precision map elements with a surface geometry, first extract their key points and key geometric attributes. Key points may include outer contour points, geometric centers, and centroids, and key geometric attributes may include areas, perimeters, etc. For extracted key point elements, obtain the mileage expression of the point elements based on the results of S1C, and record the relevant geometric attributes.

[0081] 4) For high-precision map elements with a volumetric geometric shape, use their projection surface on the XY plane and process them in the same way as the above-mentioned surface elements to achieve the associative mapping between spatial coordinate elements and milepost elements.

[0082] Step S2 realizes the expression of various elements and data collected by high-precision maps in the form of mileage piles to facilitate comparison with business data.

[0083] S3 queries and calculates the appropriate high-precision map data elements for various types of data in the business data set, and sorts them by relevance to achieve the association between business data and high-precision map data.

[0084] The purpose of this process is to unify the business data to be matched with the high-precision map data, and perform data processing to prepare for subsequent fuzzy retrieval and matching.

[0085] 1) Build a unified data classification and coding standard to achieve data classification and unified coding of data resources. This process can refer to the actual situation and combine and integrate various current national standards, industry standards, local standards, etc. to achieve unified data processing. The standards used must be in the order of standard levels (for example, national standards > industry standards > local standards, and national standards are preferred when there are national standards).

[0086] 2) Based on the constructed data classification and coding standards, for calibration business, the business data to be calibrated and the high-precision map data used for calibration reference are unified and merged with the table as the dimension to obtain a complete data list. For the same type of data tables in the list, a unified description method is used.

[0087] 3) Data merging: Merge the two types of data (a. All business data to be spatially calibrated and b. Map data with high spatial position accuracy) into the same data set to facilitate further calculation of feature vectors in step 4). The detailed execution process is as follows:

[0088] ① Calculate the UUID of all feature data records; existing algorithms can be used here, such as random UUID or hash UUID-5 based on specific feature values, or geohash value encoding based on the longitude and latitude data of the data;

[0089] ② Based on the data classification rules, classify and classify all factor data and include the classification codes in the data set;

[0090] ③ Based on the mileage reference system, all element data are parameterized by mileage and included in the data set; as supplementary information for spatial positioning, geographic coordinate system records can be added for applicable data;

[0091] ④ For the common information of the two types of data, further field-level merging and data integration are performed;

[0092] ⑤For non-common information, record them separately to achieve data merging.

[0093] 4) For each record in the merged data set, select the feature fields that are meaningful for fusion, and calculate the feature vector values ​​respectively to realize the similarity analysis and similarity distance calculation between the element data. The "meaningful for fusion" refers to the information field that can objectively describe the individual ontology of the element and is conducive to determining the size of the similarity difference between the records to be fused. The more types of information contained in the feature field (the richer the amount of information), the higher the accuracy of the fusion match. The feature field should at least contain fields with business meaning and fields containing positioning information. The fields with business meaning can be names, numbers, classifications and codes, and the fields containing positioning information can be road information, mileage location, geographic location, etc. For example, for a certain element: data source "A", facility "gantry", name "Liuji to Chenji", number "XXXX", facility classification code "YYYY", mileage code "S0049M01K0268+262.15|0", longitude and latitude (E119.2087349, N32.4956651), GEOHASH code "WTUBBETHUU8G", located on the S49 downlink route, heading angle 310.61... The name "gantry" can be used as a field with business meaning, and the longitude and latitude (E119.2087349, N32.4956651) can be combined as the characteristic field.

[0094] Furthermore, the selected feature fields are converted into text and then the low-dimensional feature vector values ​​are calculated using the embedding technology / model. This technology is a technology that maps high-dimensional data (such as feature information, description definitions, field vocabulary, etc.) to a low-dimensional vector space. Its advantage is that it can significantly reduce the dimension of the data while retaining the important characteristics of the original data. The feature values ​​after dimensionality reduction are easier to be processed by other models, and can also improve computing performance. It is widely used in the fields of recommendation systems and natural language processing.

[0095] In this case, the purpose of using embedding technology is to convert the element information that is helpful for fusion in the merged data set into text and then convert it into a low-dimensional feature vector, so as to facilitate similarity analysis and distance calculation between data, thereby realizing the effective fusion of business data and high-precision map data under the consideration of the complex dimensions of business data and high-precision map data. The specific implementation of this process can be considered in the following two ways according to the actual situation: ① Online API service call, ② Local model processing. At present, both methods have mature models / solutions, and this case will not be described in detail.

[0096] 5) Calculate the distance. For each business data to be calibrated in the merged data set, use a distance function to calculate the first k high-precision map reference feature data closest to it, and sort the records in ascending order of distance. The distance function uses the vector data (feature vector value) of two data records as input, and can use the Euclidean distance or cosine distance function.

[0097] For the business data to be calibrated, the high-precision map data with the shortest distance is the most matching data; at this point, data fusion at the algorithm level is completed.

[0098] S4 evaluates the utility of the matched data. For low-reliability matching results, manual verification and round-robin voting are used to obtain the final result.

[0099] In order to meet the production system's requirements for fused data quality, the results of the algorithm fusion need to be manually checked. The manual check method is to display the two types of fused data one by one in the general GIS software or custom-developed WebGIS system based on the coordinates of the same geographic coordinate system or the pile numbers of the road mileage pile system, and provide their various attributes for manual review of whether the algorithm matching is accurate. Taking the check of a single calibration result pair as an example, the checker first checks the first adaptation item (shortest distance). If it fails, the next adaptation item is checked one by one until all the first k adaptation items are checked, and the qualified or unqualified conclusion is fed back to the system.

[0100] Depending on the system's requirements for the quality of fused data, spot checks or full checks can be used. Spot checks are to randomly select a certain proportion of data from the fused results for verification to verify the accuracy of the algorithm fusion; full checks are to check all the data in the fused data set to verify the comprehensiveness of the algorithm fusion.

[0101] According to the requirements for the quality of fusion verification, it can be divided into single-person verification and multi-round verification. Single-person verification means that a verification staff who is familiar with the business and maps will verify the fused data, which is more suitable for random inspection. Multi-person verification means that multiple verification staff who are familiar with the business and maps will conduct multiple rounds of verification on the fused data, and use a circular voting method to confirm the final result. It is more suitable for business systems with higher requirements for data quality, and it is often implemented through a customized WebGIS system.

[0102] During cyclic voting, two parameters are used to control the data qualification threshold: 1) the minimum number of times a single data record is jointly checked by different people, n, and the maximum number of times it is jointly checked, m; 2) the minimum proportion of a single data record that passes the joint inspection by different people, p%. After the inspection system operation is started, all data in 5) of S3) are stored in the pool to be inspected, randomly assigned to different verification personnel and inspected. When a data record meets the requirements of more than n joint inspections and a pass ratio greater than p%, it is placed in the verified pool, otherwise it is returned to the pool to be inspected; when a record in the pool to be inspected reaches the maximum number of inspections, m, but fails, it is placed in the exception pool and handled by a dedicated person; until the pool to be inspected is empty, the inspection is completed, and the passed data records are placed in the final data pool for use by the business system.

[0103] The embodiment realizes the fusion of business data and high-precision map data through the S1-S4 core process, and provides fused data with high spatial precision definition and business attributes based on business data.

Claims

1. A method for fusing multi-source multi-component spatial data, wherein the multi-source multi-component spatial data includes business data with spatial attribute information and map data with high-precision spatial location information, characterized in that: The method comprises: According to the spatial position of the physical kilometer piles in the high-precision map data and the road reference line, the projection kilometer piles on the road reference line are obtained to form a kilometer pile benchmark network; The kilometer pile benchmark network is interpolated at meter level along the road reference line in the longitudinal direction of the road to obtain the meter-level mileage pile benchmark network. Each meter-level mileage pile is projected along the center line of each lane in the transverse direction of the road to obtain the lane-by-lane meter-level mileage pile benchmark network as the mileage pile reference system. Merge the business data and high-precision map data into the same data set, select feature fields for each element record in the merged data set, calculate feature vector values ​​using embedding technology, and use them for similarity analysis and similarity distance calculation between element data, and match high-precision map data elements and business data elements according to the shortest distance to achieve the association and fusion of business data and high-precision map data; the feature fields at least include fields containing business meanings and fields containing positioning information; Among them, each mileage post in the meter-level mileage post benchmark network is assigned a code, and the code composition includes: a route code describing the route of the road where the mileage post is located, a road code describing the type and direction of the road where the mileage post is located, a mileage code describing the mileage where the mileage post is located, and a lane code describing the lane where the mileage post is located; the data set includes mileage post parameterized data obtained by parameterizing the elements in the high-precision map data and business data based on the mileage post reference system.

2. The method according to claim 1, characterized in that If there are one or more missing points between two adjacent kilometer posts, use the kilometer post points projected before and after the missing points and perform average interpolation according to the number of missing points along the road reference line.

3. The method according to claim 1, characterized in that: It also includes adjusting the positions of the projected kilometer piles along the road reference line to minimize the distance error between adjacent projected kilometer piles globally, and constructing a kilometer pile reference network using the adjusted projected kilometer piles as reference points of the pile number reference system.

4. The method according to claim 1, characterized in that: The route code includes a character bit representing the road grade and a character bit representing the road number; The road code includes a character bit indicating the road type and a character bit indicating the driving direction; the road type includes a main road and a ramp, which are distinguished by different letters, and the driving direction includes forward and reverse, which are distinguished by different numbers; The mileage code includes a character position indicating kilometers, a character position indicating a specific kilometer value, a character position indicating meters, a character position indicating a specific meter value and / or a character position indicating a decimal point and a character position indicating a centimeter value; The lane code includes a separator character bit and a character bit representing a lane number. The lane number is: corresponding to the lane longitudinally, numbered incrementally from the inside to the outside by integers; when the value is 0, there is no lane information.

5. The method according to claim 4, characterized in that The road code also includes character bits for road section disambiguation and expansion, where numbers are used when the road section where the milepost is located is the main line, and letters are used for ramps.

6. The method according to claim 1, characterized in that It also includes establishing a conversion model between the mileage stake reference network and the geographic spatial position reference system; For the conversion of mileage posts to spatial coordinates, the geometric coordinates are queried through the mileage post codes and converted to any geographic coordinate system; For the conversion of geospatial coordinates to mileage posts, the geospatial coordinates are converted to the coordinate system corresponding to the geometric coordinates of the mileage post, and then the nearest mileage post is found in the mileage post reference network to convert the spatial coordinates to the mileage post.

7. The method according to claim 1, characterized in that The milepost parameterization of the elements is to associate and map the point, line, surface and body elements to the mileposts, so that the elements have milepost parameterization data consisting of their own spatial coordinate values ​​and milepost coding values.

8. The method according to claim 1, characterized in that The merging of business data and high-precision map data into the same data set includes: UUID is compiled for all feature data records in business data and high-precision map data; Based on the unified data classification and coding standards, all element data are classified and coded, and based on the mileage reference system, all element data are parameterized by mileage; the classification coding and mileage parameterization data are included in the data set; For the common information in the business data and high-precision map data, further field-level merging is performed; for the non-common information, they are recorded separately.

9. The method according to claim 1, characterized in that: It also includes, after the data are matched, verifying the matched data by manual verification and / or circular voting to complete the data fusion.

10. A fusion system for multi-source and multi-component spatial data, characterized in that: The system comprises: A data acquisition module is used to acquire business data with spatial attribute information and map data with high-precision spatial location information; The mileage pile reference system construction module obtains the projected mileage piles on the road reference line according to the spatial position of the physical mileage piles in the high-precision map data and the road reference line, forming a mileage pile reference network; the mileage pile reference network is interpolated at the meter level along the road reference line in the longitudinal direction of the road to obtain a meter-level mileage pile reference network, and each meter-level mileage pile is projected along the center line of each lane in the transverse direction of the road to obtain a lane-by-lane meter-level mileage pile reference network as a mileage pile reference system; Each milepost in the meter-level milepost reference network is assigned a code, the code comprising: a route code describing the route of the road where the milepost is located, a road code describing the type and direction of the road where the milepost is located, a mileage code describing the mileage where the milepost is located, and a lane code describing the lane where the milepost is located; A data merging module, merging business data and high-precision map data into the same data set; the data set includes mileage post parameterized data obtained by parameterizing the elements in the high-precision map data and business data based on the mileage post reference system; The data fusion module selects characteristic fields for each element record in the merged data set, calculates the characteristic vector value using embedding technology, and uses it for similarity analysis and similarity distance calculation between element data, and matches the high-precision map data elements and business data elements according to the shortest distance to achieve the association and fusion of business data and high-precision map data; the characteristic fields include at least a field containing business meaning and a field containing positioning information.

Images