Method, device, equipment and medium for extracting road information from electronic maps

Abstract

The present application discloses a method, device, equipment and medium for extracting road information from an electronic map, and relates to the field of map making in the surveying and mapping and geographic information industry. The method includes adopting a progressive registration method to register and correct a target map slice set, and obtain a road raster to be processed and an annotation raster to be processed; extracting the road surface of the target area based on a road surface extraction model; inputting the annotation raster to be processed into the annotation information extraction model to obtain single-character discrete text of the target area; determining the annotation library of the target area; extracting the road elements of the target area; matching the annotation library of the target area with the road elements of the target area, and determining the road entities of the target area according to the name attributes of the road elements of the target area and the road elements of the target area. The present application realizes the accurate extraction of road entities in electronic maps.

Description

Technical Field

[0001] The present application relates to the field of map making in the surveying and mapping and geographic information industries, and in particular to a method, device, equipment and medium for extracting road information from electronic maps. Background Art

[0002] Electronic maps have obvious advantages in terms of the current situation and information richness of urban roads. They can be used as an important supplement to high-cost road collection methods such as remote sensing image interpretation and field surveying. The vigorous development of technologies such as high-precision map generation and unmanned driving has given electronic maps greater vitality. In addition, as one of the important forms of map products, electronic maps achieve visual multi-level scale-free scaling by pre-dynamically loading slices. Although they lack the beauty and rigor of professional atlas products, they have also passed the process of cartographic synthesis, element classification and grading, annotation configuration, conflict handling, etc., and can extract the experience and knowledge of map experts to assist in the rapid compilation of maps with higher requirements.

[0003] The application of electronic navigation maps has penetrated into all aspects of the transportation field. High-precision maps for unmanned driving are also booming. The continuous improvement of relevant standards has also laid the foundation for the standardized sharing of mainstream navigation data products. Therefore, it is necessary to extract electronic map road information for map Figure 2 The feasibility of scenarios such as secondary compilation and data update has been greatly enhanced, but in the extraction of road information from electronic maps, there are the following problems:

[0004] (1) Nonlinear offset of electronic map data: In response to relevant requirements, electronic maps usually need to be offset in coordinates. However, the offset is nonlinear, which leads to poor global accuracy during grid registration and the extracted data cannot meet the accuracy requirements.

[0005] (2) Overlapping of multiple elements leads to inaccurate extraction: Electronic maps display multiple elements such as roads, water systems, buildings, and annotations. In addition, similar elements are divided into multiple types. Many elements overlap and overlap each other, resulting in low accuracy of elements extracted using pure raster pixel processing methods.

[0006] Therefore, based on the above problems, there is an urgent need to provide a new method for extracting road information from electronic maps to achieve accurate extraction of road entities in electronic maps. Summary of the invention

[0007] The purpose of this application is to provide a method, device, equipment and medium for extracting road information from an electronic map, which can achieve accurate extraction of road entities in an electronic map.

[0008] To achieve the above objectives, this application provides the following solutions:

[0009] In a first aspect, the present application provides a method for extracting road information from an electronic map, comprising:

[0010] Acquire a target map slice set; the target map slice set includes map slices of a target area, and the map slices include road slices and annotation slices;

[0011] Using a progressive registration method, the target map slice set is registered and corrected to obtain a road grid to be processed and an annotation grid to be processed;

[0012] Inputting the road grid to be processed into a road surface extraction model to extract the road surface of the target area; the road surface extraction model is obtained based on a ResNet-34 model;

[0013] Inputting the annotation grid to be processed into an annotation information extraction model to obtain the single-character discrete text of the target area, and determining the center pixel coordinates of the single-character discrete text of the target area as the single-character discrete text coordinates of the target area; the annotation information extraction model is obtained based on a deep learning neural network;

[0014] According to the annotation grid to be processed and the single-character discrete text coordinates of the target area, an annotation library of the target area is obtained; the annotation library includes: semantic text information of each road name;

[0015] Using a reclassification extraction method to perform vector extraction on the road surface of the target area to obtain road elements in the target area;

[0016] Matching the annotation library of the target area with the road elements of the target area to determine the name attributes of the road elements of the target area;

[0017] The road entity of the target area is determined according to the name attribute of the road element of the target area and the road element of the target area.

[0018] Optionally, a progressive registration method is used to perform registration and deviation correction on the target map slice set to obtain a road grid to be processed and an annotation grid to be processed, specifically including:

[0019] The target map slice set is spliced ​​and merged in sequence according to the row and column numbers of the map slices of the target area to form a large grid;

[0020] Creating a grid using the boundary line of each map slice in the large grid as a grid line, and marking the coordinates of the map slice in the large grid based on the grid;

[0021] The road grid to be processed and the annotation grid to be processed in the large grid are determined by using an N-iteration method; wherein the process of the n-th iteration includes:

[0022] Selecting the registered grid range of the nth iteration from the large grid according to the maximum number of grid rows and columns of the nth iteration, the set registration accuracy and the coordinates of the map slice; 1≤n≤N;

[0023] Take out the road slices and annotation slices within the registered grid range of the nth iteration from the target map slice set;

[0024] The road slices in the range of the registered grid of the nth iteration are spliced ​​to obtain a spliced ​​road grid; the annotation slices in the range of the registered grid of the nth iteration are spliced ​​to obtain a spliced ​​annotation grid;

[0025] The coordinates of the spliced ​​road grid and the spliced ​​annotation grid are transformed by using a second-order polynomial transformation method to form an n-th iteration registered road grid and an n-th iteration registered annotation grid.

[0026] If n is equal to N, the registered road grids of the previous n iterations are spliced ​​to obtain the road grid to be processed, and the registered annotation grids of the previous n iterations are spliced ​​to obtain the annotation grid to be processed;

[0027] If n is not equal to N, the n+1th iteration is performed.

[0028] Optionally, selecting the registered grid range of the nth iteration from the large grid according to the maximum number of grid rows and columns of the nth iteration, the set registration accuracy and the coordinates of the map slice, specifically includes:

[0029] According to the maximum number of grid rows and columns of the nth iteration, the set registration accuracy and the coordinates of the map slice, the registered grid range of the nth iteration is selected from the large grid using the M-iteration method; wherein the process of the mth iteration includes:

[0030] Determine the maximum number of grid rows and columns of the mth iteration, and select the registration range of the mth iteration in the large grid according to the maximum number of grid rows and columns of the mth iteration; 1≤m≤M;

[0031] Obtaining the registration vector data within the registration range of the mth iteration;

[0032] Determine the control points of the marker within the registration range of the mth iteration and the control points of the same marker corresponding to the registration vector data as a pair of control points, and select a set number of control point pairs to form a set of control point pairs within the registration range of the mth iteration;

[0033] A second-order polynomial transformation method is used to transform the coordinates of the large-scale grid according to the control point pair set and the coordinates of the map slice within the registration range of the mth iteration, so as to obtain the large-scale grid after the mth iteration transformation; the large-scale grid after the mth iteration transformation and the registration vector data within the registration range of the mth iteration are in the same coordinate system;

[0034] According to the control point pair set within the registration range of the mth iteration and the large grid transformed after the mth iteration, the registration accuracy within the registration range of the mth iteration is calculated;

[0035] Determine whether m is equal to M;

[0036] If m is equal to M, the registration range of the mth iteration is used as the registered grid range of the nth iteration;

[0037] If m is not equal to M, then determine whether the difference between the registration accuracy within the registration range of the mth iteration and the set registration accuracy is within the set threshold range;

[0038] If the difference is within the set threshold range, the registration range of the mth iteration is used as the registered grid range of the nth iteration;

[0039] If the difference is not within the set threshold range, and the registration accuracy within the registration range of the mth iteration is less than the set registration accuracy, the maximum number of grid rows and columns of the mth iteration is increased, and the increased maximum number of grid rows and columns of the mth iteration is determined as the maximum number of grid rows and columns of the m+1th iteration, and the m+1th iteration is performed;

[0040] If the difference is not within the set threshold range, and the registration accuracy within the registration range of the mth iteration is greater than or equal to the set registration accuracy, the maximum number of grid rows and columns for the mth iteration is reduced, and the reduced maximum number of grid rows and columns for the mth iteration is determined as the maximum number of grid rows and columns for the m+1th iteration, and the m+1th iteration is performed.

[0041] Optionally, the method for determining the road surface extraction model is:

[0042] Acquire a training road slice set; the training road slice set includes road slices in a map slice of a training area;

[0043] Using a progressive registration method, the training road slice set is registered and corrected to obtain a training road grid;

[0044] Extracting road lines in the training road grid, and determining the road surface of the training area according to the road lines;

[0045] The training road grid and the road surface of the training area are used to train a ResNet-34 model to obtain the road surface extraction model.

[0046] Optionally, the method for determining the annotation information extraction model is:

[0047] Acquire a training annotation slice set; the training annotation slice set includes annotation slices in the map slices of the training area;

[0048] Using a progressive registration method, the training annotation slice set is registered and corrected to obtain a training annotation grid;

[0049] Locating the minimum area containing each annotation in the training annotation grid to obtain the minimum training rectangular selection area of ​​each annotation;

[0050] Taking pixels as basic units, segmenting the annotations in the minimum training rectangular selection area to determine single-character discrete text in the training area;

[0051] The training annotation grid and the single-character discrete text in the training area are used to train a deep learning neural network to obtain the annotation information extraction model.

[0052] Optionally, obtaining an annotation library of the target area according to the annotation grid to be processed and the single-character discrete text coordinates of the target area specifically includes:

[0053] Locating the minimum area containing each annotation in the annotation grid to be processed to obtain the minimum rectangular selection area to be processed for each annotation;

[0054] Constructing a vector surface of the minimum rectangular selection area to be processed according to the edge line coordinate string of the minimum rectangular selection area to be processed;

[0055] Performing spatial intersection on the vector surface of the minimum rectangular selection area to be processed of all annotations and the coordinates of the single-character discrete text to obtain the intersecting characters of all annotations;

[0056] According to the character segmentation order, all the intersecting characters of the annotations are combined into semantic text information that conforms to the actual attributes of the features;

[0057] The annotation library is constructed according to the semantic text information.

[0058] Optionally, a reclassification extraction method is used to perform vector extraction on the road surface of the target area to obtain road elements of the target area, specifically including:

[0059] Binarizing the pixels in the road surface of the target area to obtain a binary grid;

[0060] Processing the binary grid using an edge detection algorithm to obtain an object outline in the binary grid;

[0061] vectorizing the object outline into surface elements;

[0062] Extracting the skeleton line of the surface element to obtain a road line vector;

[0063] The road line vectors are converted to obtain road elements in the target area; the conversion process includes connecting broken lines, merging broken lines and deleting branch roads.

[0064] In a second aspect, the present application provides an electronic map road information extraction device, comprising:

[0065] A slice acquisition module, used to acquire a target map slice set; the target map slice set includes map slices of a target area, and the map slices include road slices and annotation slices;

[0066] A registration and splicing module is used to register and correct the target map slice set by using a progressive registration method to obtain a road grid to be processed and an annotation grid to be processed;

[0067] A road surface extraction module, used to input the road grid to be processed into a road surface extraction model to extract the road surface of the target area; the road surface extraction model is obtained based on a ResNet-34 model;

[0068] An annotation information extraction module is used to input the annotation grid to be processed into an annotation information extraction model to obtain the single-character discrete text of the target area, and determine the center pixel coordinates of the single-character discrete text of the target area as the single-character discrete text coordinates of the target area; the annotation information extraction model is obtained based on a deep learning neural network;

[0069] An annotation library determination module is used to obtain an annotation library of the target area according to the annotation grid to be processed and the single-character discrete text coordinates of the target area; the annotation library includes: semantic text information of each road name;

[0070] A road line extraction module, used to perform vector extraction on the road surface of the target area by using a reclassification extraction method to obtain road elements of the target area;

[0071] A matching module, used for matching the annotation library of the target area with the road elements of the target area to determine the name attributes of the road elements of the target area;

[0072] The road information integration module is used to determine the road entity of the target area according to the name attribute of the road element of the target area and the road element of the target area.

[0073] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any one of the above-mentioned methods for extracting road information from an electronic map.

[0074] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the above-mentioned methods for extracting road information from an electronic map.

[0075] According to the specific embodiments provided in this application, this application has the following technical effects:

[0076] The present application provides a method, device, equipment and medium for extracting road information from an electronic map. By adopting a progressive registration method, a target map slice set is registered and corrected so that the local accuracy of the electronic map meets the standard, which can not only improve the registration accuracy, but also significantly reduce geometric distortion. Road surface and single-character discrete text extraction are realized based on a road surface extraction model and an annotation information extraction model. The annotation library is spatially matched with road elements to complete the attributeization of road elements, which supplements the attribute information for the road and is beneficial for subsequent use. The present application realizes high-precision control and information-attributed electronic map road information extraction, forms a road entity containing attribute information, and thus realizes accurate extraction of electronic map road entities. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0078] Figure 1 A schematic diagram of a process of extracting road information from an electronic map in an embodiment of the present application;

[0079] Figure 2 A schematic diagram of a progressive six-point registration method provided in one embodiment of the present application;

[0080] Figure 3 A schematic diagram of a method for determining an annotation information extraction model provided in an embodiment of the present application;

[0081] Figure 4 A schematic diagram of a method for attribute annotation provided in an embodiment of the present application;

[0082] Figure 5 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0083] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0084] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0085] In an exemplary embodiment, Figure 1 As shown, a method for extracting road information from an electronic map is provided, and the method for extracting road information from an electronic map comprises:

[0086] Obtain a target map slice set; the target map slice set includes map slices of a target area, and the map slices include road slices and annotation slices.

[0087] The target map slice set is registered and rectified by using a progressive registration method to obtain a road grid to be processed and an annotation grid to be processed.

[0088] The road grid to be processed is input into a road surface extraction model to extract the road surface LRDARaster of the target area; the road surface extraction model is obtained based on a ResNet-34 model.

[0089] The annotation grid to be processed is input into an annotation information extraction model to obtain single-character discrete text in the target area, and the coordinates of the central pixel point of the single-character discrete text in the target area are determined as the coordinates of the single-character discrete text in the target area; the annotation information extraction model is obtained based on a deep learning neural network.

[0090] According to the annotation grid to be processed and the single-character discrete text coordinates of the target area, an annotation library of the target area is obtained; the annotation library includes: semantic text information of each road name.

[0091] A reclassification extraction method is used to perform vector extraction on the road surface of the target area to obtain the road elements of the target area.

[0092] The annotation library of the target area is matched with the road elements of the target area to determine the name attributes of the road elements of the target area.

[0093] The road entity of the target area is determined according to the name attribute of the road element of the target area and the road element of the target area.

[0094] In another exemplary embodiment of the present application, electronic map service providers usually slice maps and publish them as map slice services (WMTS). Through the API (Application Program Interface) provided by the service provider, a set of slices with a specific range, specific category, and specific level can be obtained in a targeted manner. The range is defined as MapExtent={StartX, StartY, EndX, EndY}, where StartX is the starting slice row number, StartY is the starting slice column number, EndX is the ending slice row number, and EndY is the ending slice column number; the category is defined as Type={T n}, T n It is an integer enumeration variable, corresponding to the slice category provided by the service provider, for example, {region=1, road=2, annotation=4, building=8}; the level is defined as Level={L n}, L n It is an integer enumeration variable, corresponding to the slice level provided by the service provider, usually 1 to 18. At the same time, the service provider's relevant parameters about slices should be obtained: slice service address TileUrl, slice size Size (usually 256*256), and resolution comparison table ResolutionTable (for example, the resolution of level 18 slices is 0.59 meters).

[0095] The specific steps to obtain the target map tile set Tiles are as follows:

[0096] (1) Calculation of row and column numbers.

[0097] First, obtain the bounding values ​​of the target area's outer rectangle LeftX, TopY, RightX, and BottomY in the geographic coordinate system, then convert the geographic coordinates into the projection coordinate system used by the electronic map, and determine the slice level to be obtained. Finally, convert the converted coordinates into row and column numbers according to the following formula to obtain MapExtent.

[0098] StartX=[(LeftX-CoordinateOriginX) / (Resolution*Size)].

[0099] StartY=[(CoordinateOriginY-TopY) / (Resolution*Size)].

[0100] EndX=[(RightX-CoordinateOriginX) / (Resolution*Size)].

[0101] EndY=[(CoordinateOriginY-BottomY) / (Resolution*Size)].

[0102] Where LeftX is the left boundary value of the target area outer rectangle, RightX is the right boundary value of the target area outer rectangle, TopY is the upper boundary value of the target area outer rectangle, BottomY is the lower boundary value of the target area outer rectangle, CoordinateOriginX is the X coordinate of the starting point of the projection coordinate system, CoordinateOriginY is the Y coordinate of the starting point of the projection coordinate system, Resolution is the resolution of the corresponding level of the slice, which is obtained from the Resolution Table, and Size is the size of the map slice.

[0103] (2) Slice download.

[0104] Electronic maps use layered slicing to achieve customized display of road networks, annotations, administrative divisions, etc. Annotation slices representing text information and road slices representing geographic entities belong to different map slice services. Annotation slices only contain text annotation information, such as "XX Community"; road slices contain road lines and surfaces. In order to minimize information interference, a strategy of separately obtaining and processing road slices and annotation slices is adopted, that is, the category Type takes values ​​2 and 4 respectively, and the Type, the obtained MapExtent, and the slice level are organized into TileUrl form, and the slices are downloaded cyclically and stored in a local folder in the naming method of "category-level-row number-column number" to form the target map slice collection Tiles.

[0105] In another exemplary embodiment of the present application, since the electronic map data offset is nonlinear, the geo-registration parameters of different regions and ranges are inconsistent, and multiple attempts to correct the offset are required. The present application adopts a progressive registration method, which gradually expands the range to obtain the maximum number of grid rows and columns MaxSizeX, MaxSizeY and the set of control point pairs under the accuracy constraint, and splices the slices to obtain the grid to be processed. The specific steps are as follows:

[0106] The target map slice set is registered and rectified by using a progressive registration method to obtain a road grid to be processed and an annotation grid to be processed, which specifically includes:

[0107] The target map slice set Tiles is stitched and merged in sequence according to the row and column numbers of the map slices of the target area to form a large grid TotalTile.

[0108] For the convenience of positioning, a grid is created with the boundary line of each map slice in the large grid as the grid line, the grid spacing is the size of each map slice, and the coordinates of the map slice in the large grid are marked based on the grid.

[0109] The road grid to be processed and the annotation grid to be processed in the large grid are determined by using an N-iteration method; wherein the n-th iteration process includes:

[0110] The registered grid range of the nth iteration is selected from the large grid according to the maximum number of grid rows and columns of the nth iteration, the set registration accuracy and the coordinates of the map slice; 1≤n≤N.

[0111] The road slices and annotation slices within the aligned grid range of the nth iteration are taken from the target map slice set.

[0112] The road slices within the registered grid range of the nth iteration are spliced ​​to obtain a spliced ​​road raster; the annotation slices within the registered grid range of the nth iteration are spliced ​​to obtain a spliced ​​annotation raster.

[0113] The coordinates of the spliced ​​road raster and the spliced ​​annotation raster are transformed by using the second-order polynomial transformation method to form the n-th iteration registered road raster and the n-th iteration registered annotation raster.

[0114] If n is equal to N, the registered road grids of the previous n iterations are spliced ​​to obtain the road grid to be processed, and the registered annotation grids of the previous n iterations are spliced ​​to obtain the annotation grid to be processed.

[0115] If n is not equal to N, the n+1th iteration is performed.

[0116] In another exemplary embodiment of the present application, selecting the registered grid range of the nth iteration from the large grid according to the maximum number of grid rows and columns of the nth iteration, the set registration accuracy and the coordinates of the map slice specifically includes:

[0117] According to the maximum number of grid rows and columns of the nth iteration, the set registration accuracy and the coordinates of the map slice, the registered grid range of the nth iteration is selected from the large grid using the M-iteration method; wherein the process of the mth iteration includes:

[0118] Determine the maximum number of grid rows and columns for the mth iteration, and select the alignment range for the mth iteration in the large grid according to the maximum number of grid rows and columns for the mth iteration; 1≤m≤M; in the first iteration, the maximum number of grid rows and columns for the first iteration can be set according to empirical values.

[0119] Obtain the registration vector data within the registration range of the mth iteration; according to relevant regulations, electronic maps are usually processed with deviation, so it is necessary to obtain the registration vector data of the target area for registration correction. In addition, it is necessary to determine the set registration accuracy, that is, the data accuracy of the project design file. If the project design file has clear mean error requirements, then the set registration accuracy Accuracy is the mean error value. If there is no clear requirement, refer to the principle of "the shortest distance on the map that the human eye can distinguish is 0.1 mm" and calculate the indicator Accuracy of the set registration accuracy according to the following formula:

[0120] Accuracy=0.1 / 1000*Scale*Factor.

[0121] In the formula, Scale is the data scale, and Factor is the coefficient (which can be adjusted according to actual conditions).

[0122] The control points of the marker within the registration range of the mth iteration and the control points of the same marker corresponding to the registration vector data are determined as a pair of control point pairs, and a set number of control point pairs are selected to form a control point pair set within the registration range of the mth iteration.

[0123] The second-order polynomial transformation method is used to transform the coordinates of the large grid according to the control point pair set and the coordinates of the map slice in the registration range of the mth iteration, and the large grid after the mth iteration transformation is obtained; the large grid after the mth iteration transformation and the registration vector data in the registration range of the mth iteration are in the same coordinate system. The second-order polynomial transformation method is a commonly used method in geometric transformation. By selecting enough control points before and after the transformation, the second-order polynomial equation X=a is solved. 0+ a1x + a2x 2 ,Y=b 0+ b1x + b2x 2 The transformation coefficient a0, etc. is then resampled for the large grid before transformation, and the sampling points are calculated point by point according to the second-order polynomial equation to generate the large grid after transformation.

[0124] According to the control point pair set within the registration range of the mth iteration and the large grid after the mth iteration transformation, the registration accuracy within the registration range of the mth iteration is calculated; specifically, the registration accuracy CurAccuracy within the registration range of the mth iteration is calculated according to the following formula:

[0125] CurAccuracy= .

[0126] Where ConPts i For the icontrol point pairs, and distance() is the Euclidean distance between the control point pairs.

[0127] Determine whether m is equal to M.

[0128] If m is equal to M, the registration range of the mth iteration is used as the registered grid range of the nth iteration.

[0129] If m is not equal to M, it is determined whether the difference between the registration accuracy within the registration range of the mth iteration and the set registration accuracy is within a set threshold range.

[0130] If the difference is within the set threshold, the registration range of the mth iteration is used as the registered grid range of the nth iteration. The control point pair set within the registration range of the mth iteration is stored as a TXT format text.

[0131] If the difference is not within the set threshold range, and the registration accuracy within the registration range of the mth iteration is less than the set registration accuracy, the maximum number of grid rows and columns for the mth iteration is increased by the step size Step (which can be set according to the empirical value), and the increased maximum number of grid rows and columns for the mth iteration is determined as the maximum number of grid rows and columns for the m+1th iteration, and the m+1th iteration is performed.

[0132] If the difference is not within the set threshold range, and the registration accuracy within the registration range of the mth iteration is greater than or equal to the set registration accuracy, the maximum number of grid rows and columns of the mth iteration is reduced by the step size Step, and the reduced maximum number of grid rows and columns of the mth iteration is determined as the maximum number of grid rows and columns of the m+1th iteration, and the m+1th iteration is performed. Specifically, the set number of control point pairs is set to 6, such as Figure 2 As shown in the figure, a 6*6 slice grid is displayed. The progressive six-point registration method is used to select six control point pairs to form a control point pair set ConPts={ConPts1, ConPts2, ConPts3, ConPts4, ConPts5, ConPts6}, where ConPts1, ConPts2, ConPts3, and ConPts4 are corner control points. Since there may be no obvious matching markers at the corner points, obvious marker points near the corner points can be selected. ConPts5 and ConPts6 are internal control points, and obvious marker points can be selected within the registration range of the mth iteration. Figure 2 The enlargement and reduction effects of the mth iteration with a registration range of 5*5 and a step size of 1 are shown in FIG.

[0133] In another exemplary embodiment of the present application, the method for determining the road surface extraction model is:

[0134] A training road slice set is obtained; the training road slice set includes road slices in the map slices of the training area.

[0135] A progressive registration method is used to perform registration and deviation correction on the training road slice set to obtain a training road grid.

[0136] Road lines in the training road grid are extracted, and a road surface of the training area is determined according to the road lines.

[0137] The training road grid and the road surface of the training area are used to train a ResNet-34 model to obtain the road surface extraction model.

[0138] In electronic maps of different levels, the road morphology, details, and features vary greatly. The sample library should be constructed by comprehensively considering the application scenarios to avoid extracting roads that do not match the scale. Therefore, a hierarchical classification strategy is used to design samples. The sample design is shown in Table 1. Dark solid line roads (usually highways, expressways, national roads and other higher-level roads) and light solid line roads (usually main roads, secondary roads, branches, county roads, and township roads) are extracted from the 15-level slices of the registration vector data, and dark solid line roads (ramps) and light dotted line roads (internal roads) are extracted from the 18-level slices. When constructing the sample, firstly, a buffer analysis is performed on the extracted road lines to obtain the road surface, and then the road surface is manually trimmed to coincide with the road boundary of the training road grid, and the classification code is annotated to form a standard sample library.

[0139] Table 1 Sample design table

[0140]

[0141] Construct a road surface extraction model. In the process of deep learning, it is often better to use a specific neural network model for a specific type of ground object. Based on the constructed standard sample library, the transfer learning method is used to train on the ResNet-34 model. By setting parameters such as iteration epochs, batch size, and learning rate, continuous iterations are performed to finally obtain the model that best meets actual requirements.

[0142] In another exemplary embodiment of the present application, the extraction of electronic map annotation information should solve two problems: one is the accurate extraction of surface information, that is, accurate text recognition and information semanticization; the other is the extraction of spatial information, that is, the spatial location record of semantic text information. This embodiment proposes a set of electronic map annotation extraction and vectorization methods consisting of automatic annotation positioning and segmentation, semantic text information recognition sample library and prediction model, discrete character semanticization, and annotation vectorization. Figure 3 As shown, the method for determining the annotation information extraction model is:

[0143] A training annotation slice set is obtained; the training annotation slice set includes annotation slices in map slices of a training area.

[0144] The progressive registration method is used to perform registration and deviation correction on the training annotation slice set to obtain a training annotation grid.

[0145] The minimum area containing each annotation in the training annotation grid is located to obtain the minimum training rectangular selection area of ​​each annotation.

[0146] Taking pixels as basic units, the annotations in the minimum training rectangular selection area are segmented to determine the single-character discrete text in the training area.

[0147] The training annotation grid and the single-character discrete text in the training area are used to train a deep learning neural network to obtain the annotation information extraction model.

[0148] Specifically, in the automatic annotation positioning and segmentation, the image enhancement, edge detection and other algorithms in traditional image processing are used to realize the positioning of the minimum training rectangular selection MinEnvelope containing each annotation in the training annotation grid. The coordinates of the central pixel point of MinEnvelope are recorded (the grid points after georeferencing can directly obtain the projection coordinates) as the positioning coordinates of MinEnvelope. Then, MinEnvelope is statistically processed with pixels as the basic unit. Combined with the geometric characteristics of the text, the text in MinEnvelope is segmented into single characters to obtain the single-character discrete text in the training area. At the same time, the central pixel coordinates of each single-character discrete text are recorded as its positioning coordinates. In constructing the annotation information extraction model, according to the graphic characteristics of the map annotation text, a representative text recognition sample library is established by the single-character discrete text of the training annotation grid and the training area, and then a deep learning neural network is trained to form an annotation information extraction model.

[0149] In another exemplary embodiment of the present application, obtaining the annotation library of the target area according to the annotation grid to be processed and the single-character discrete text coordinates of the target area specifically includes:

[0150] The minimum area containing each annotation in the annotation grid to be processed is located to obtain the minimum rectangular selection area to be processed for each annotation.

[0151] A vector surface of the minimum rectangular selection area to be processed is constructed according to the edge line coordinate string of the minimum rectangular selection area to be processed.

[0152] The vector surface of the minimum rectangular selection area to be processed of all annotations and the coordinates of the single-character discrete text are spatially intersected to obtain the intersecting characters of all annotations.

[0153] According to the character segmentation order, all the intersecting characters of the annotations are combined into semantic text information that conforms to the actual attributes of the features.

[0154] The annotation library is constructed according to the semantic text information.

[0155] As an optional execution method, the annotation library is constructed according to the semantic text information. Specifically, the positioning coordinates of MinEnvelope are used as the positioning coordinates of the semantic text information, and then the annotation is vectorized. First, a vector point feature class and a surface feature class are created in the blank annotation library AnnoData. The attribute field design is shown in Table 2, wherein the "associated single character" field records which elements in the point feature class the "name" of the semantic text information is composed of by storing the unique identification ID of the point feature, and the "character size" field records the character size of the semantic text information; then, the point feature is created using the positioning coordinates of the obtained single-character discrete text, and the surface feature is created using the positioning coordinates of the semantic text information, and the relevant attribute assignment is completed.

[0156] Table 2 Annotation attribute field design table

[0157]

[0158] In another exemplary embodiment of the present application, a reclassification extraction method is used to perform vector extraction on the road surface of the target area to obtain the road elements of the target area, which specifically includes:

[0159] The road surface pixels extracted in the above embodiment take values ​​of 0, 1, 2 or 0, 3, 4, and the pixels in the road surface of the target area are binarized to obtain a binary raster LRDARasterReclass. The binarization includes setting the pixel value of 1 to 1, and setting the pixel value of not 1 to 0, so as to separate the object to be extracted from others.

[0160] The binary grid is processed by an edge detection algorithm to obtain the object contour in the binary grid.

[0161] The object outline is vectorized into surface elements.

[0162] The skeleton line of the surface element is extracted by the skeleton line method to obtain the road line vector, and the road width is recorded.

[0163] The extracted road line vectors have many topological problems, and human-computer interactive post-processing is performed to convert the vectorization results into practical vector data. Specifically, the road line vectors are converted to obtain road elements in the target area; the conversion process includes operations such as connecting broken lines, breaking T-shaped lines, merging broken lines, and deleting branch roads.

[0164] In another exemplary embodiment of the present application, the processed road line elements and the obtained annotation library AnnoData are two independent elements of surface and point, which are not convenient for application. The semantic text information is converted into the name attribute of the road element through annotation attributeization, such as Figure 4 As shown, the annotation library of the target area is matched with the road elements of the target area to determine the name attributes of the road elements of the target area; the road entities of the target area are determined according to the name attributes of the road elements of the target area and the road elements of the target area, and the specific steps are as follows:

[0165] The road elements of the target area are stored in the road element set LRDL.

[0166] Take a road feature in the road feature set as the target road.

[0167] Build buffer polygons for target roads.

[0168] The semantic text information of the spatial intersection of the annotation library AnnoData of the target area and the buffer surface of the target road is obtained as the candidate annotation target set Annos.

[0169] The envelope between each semantic text information in the candidate annotation target set and the target road is calculated, and the semantic text information corresponding to the maximum envelope value is taken as the optimal target.

[0170] The envelope calculation formula is:

[0171] .

[0172] .

[0173] In the formula, is the envelope, t is the number of characters of semantic text information in the candidate annotation target set, The first i The coordinates of the characters, The first i The size of each character can be obtained based on the value of the "Character Size" field in the candidate annotation target set. is the width of the target road; For the target road, The buffer polygon for the target road.

[0174] The optimal target is assigned to the name field of the target road, the assigned road element is obtained, and the optimal target unique identification ID is recorded.

[0175] Return and execute the step of taking a road element in the road element set as the target road until all road elements in the road element set are processed and a preliminary road entity is obtained. The preliminary road entity contains some unassigned elements that need to be further processed. The specific steps are as follows:

[0176] An unassigned road feature in the preliminary road entity is used as a seed road segment.

[0177] Search the preliminary road entities for a connecting road segment connected to the seed road segment.

[0178] Calculate the angles between all connected segments and the seed segment, and take the connected segment with the maximum angle as the segment to be merged.

[0179] Calculate the width difference between the segment to be merged and the seed segment.

[0180] If the width difference is within the set threshold range, the road segment to be merged and the seed road segment are merged into a new seed road segment, and the step of searching for a connecting road segment connected to the seed road segment from the preliminary road entity is returned and executed until there is no connecting road segment connected to the seed road segment in the preliminary road entity, and a seed road segment is obtained.

[0181] For the remaining unvalued road elements in the preliminary road entity, return and execute the step of taking an unvalued road element in the preliminary road entity as a seed road segment, until all unvalued road elements in the preliminary road entity are processed, and all the obtained seed road segments constitute a road stroke set. The road stroke set is a set of line segments connected by a group of naturally extended and unbroken roads with "good continuity" in direction. It is actually a series of road segment sets connected by specific rules in the road network, which can be used to extract road entities with the same name.

[0182] According to the map annotation related principles, the annotations of some unassigned road sections are inferred to form the annotation attribute results. The inference rules are as follows:

[0183] (1) If there are sections with the same name in the road stroke set, the section names between the sections with the same name are assigned the same name as the sections with the same name.

[0184] (2) If there is only one assigned road element in the road stroke set, all road segment names in the road stroke set are assigned the same name as the assigned road element.

[0185] A road entity in a target area is obtained according to the preliminary road entity and the road stroke set.

[0186] In another exemplary embodiment of the present application, cartographic experience extraction is performed. According to the display color of electronic map elements and annotation information, cartographic experience such as element classification and annotation configuration is extracted and stored in the attributes. The specific steps are as follows:

[0187] Define cartographic experience storage fields and create related fields for the road feature set LRDL. The field design is shown in Table 3, where the "display level" field records the source slice level of the road feature, the "element display level" field records the display priority of the road feature, the "annotation number" field records the number of times the road feature is annotated in sections at the corresponding slice level, and the "annotation position" field records the relative position of each annotation of the road feature.

[0188] Assign values ​​to cartographic experience extraction, specifically:

[0189] (1) The "Display Level" field is directly assigned the slice level.

[0190] (2) The “Feature Display Level” field is assigned the classification code of the extracted road feature.

[0191] (3) Calculate the number of unique identifiers of the semantic text information Anno corresponding to the road element and assign them to the "Number of Annotations" field; obtain the geometry of the annotation surface element from the annotation surface element based on the ID, take the geometric center point of the surface element, calculate the point on the road element that is closest to the center point as the adjacent point, calculate the distance from the starting point of the road element to the adjacent point, and the ratio of this distance to the length of the road element is the value of the "Annotation Position".

[0192] Table 3 LRDL mapping experience related attribute design table

[0193]

[0194] Traditional electronic map road extraction technology usually relies on manual collection or adopts raster extraction methods based on map color features. Compared with traditional methods, this application has the following significant advantages:

[0195] (1) Limit the single grid processing range and significantly reduce geometric distortion.

[0196] Traditional methods usually perform geometric correction on vectors after extraction. Due to the nonlinear offset characteristics of electronic maps, mathematical transformation methods cannot achieve good correction results for the overall data. This application uses pre-defined accuracy indicators and progressive registration methods to segment large-area grids so that local accuracy meets the standards, which can not only improve registration accuracy but also significantly reduce geometric distortion.

[0197] (2) Integrated extraction of vectors and annotations supported by deep learning.

[0198] Traditional methods often use conventional grid pixel segmentation, growth, expansion and other algorithms, which are limited by grid accuracy and will extract too many details, increasing the workload of post-processing. This application uses deep learning methods and adopts the idea of ​​constructing sample extraction by layering and object, extracting the most obvious objects from different levels and types of grids, which can significantly reduce the workload of post-processing and realize the integrated extraction of roads and characters.

[0199] (3) Attribute semantic text information and supplement road attribute information.

[0200] Traditional methods do not involve the extraction of map annotations, resulting in missing attribute information. This application determines semantic text information based on characters extracted by deep learning, spatially matches them with road elements, and sets reasoning assignment rules to complete the attributeization of semantic text information, supplementing the attribute information for road elements, which is beneficial for subsequent use.

[0201] (4) Extract mapping experience and improve secondary application efficiency.

[0202] Electronic maps are also a kind of map product. However, the cartographic experience of element classification and annotation configuration is relatively abstract and cannot be directly obtained. It is necessary to summarize and extract from the information such as element display symbols and display locations, which results in the fact that the experience information is usually ignored. Traditional methods do not involve summarizing and refining the cartographic experience contained in electronic maps. This application mainly focuses on element classification and annotation configuration, extracts and records relevant information, and provides a map based on this data. Figure 2 Applications such as sub-compilation and data updating provide experience knowledge and improve efficiency.

[0203] In response to the above problems, this application constructs an "acquisition-extraction-integration" processing closed loop by limiting the slicing range, integrated extraction of roads and annotations, and extraction of mapping experience, thereby realizing an electronic map road information extraction method with high-precision control, information attribution, and quantification of institutional experience.

[0204] This application first obtains the target map slice set, registration vector data and sets the registration accuracy requirements to complete data and parameter preparation; then, the target map slice set is geo-referenced with the registration vector data, and the maximum number of raster rows and columns is determined in combination with the raster information and data accuracy requirements, and they are spliced ​​into a raster to be processed to solve the problems of electronic map coordinate offset and geometric deformation; secondly, deep learning samples are created and a road surface extraction model and an annotation information extraction model are trained, and the model is used to extract road surfaces and single-character discrete text. The model is a pre-trained model and supports transfer learning; finally, the road surface and single-character discrete text extracted by the road surface extraction model and the annotation information extraction model are processed, including road line extraction, road line post-processing, annotation attributeization, and cartographic experience extraction operations, and finally a road entity containing attribute information is formed.

[0205] This application enriches the road update data source and improves the update efficiency by extracting road information from electronic maps, integrating road lines, road elements, names, road categories and other attributes, annotating points, road classification and other cartographic experience and knowledge.

[0206] Based on the same inventive concept, the embodiment of the present application also provides an electronic map road information extraction device for implementing the electronic map road information extraction method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more electronic map road information extraction device embodiments provided below can refer to the limitations of the electronic map road information extraction method above, and will not be repeated here.

[0207] In an exemplary embodiment, a device for extracting road information from an electronic map is provided, comprising:

[0208] The slice acquisition module is used to acquire a target map slice set; the target map slice set includes map slices of a target area, and the map slices include road slices and annotation slices.

[0209] The registration and splicing module is used to use a progressive registration method to perform registration and deviation correction on the target map slice set to obtain a road grid to be processed and an annotation grid to be processed.

[0210] The road surface extraction module is used to input the road grid to be processed into a road surface extraction model to extract the road surface of the target area; the road surface extraction model is obtained based on the ResNet-34 model.

[0211] An annotation information extraction module is used to input the annotation grid to be processed into an annotation information extraction model to obtain the single-character discrete text of the target area, and determine the center pixel point coordinates of the single-character discrete text of the target area as the single-character discrete text coordinates of the target area; the annotation information extraction model is obtained based on a deep learning neural network.

[0212] The annotation library determination module is used to obtain the annotation library of the target area according to the annotation grid to be processed and the single-character discrete text coordinates of the target area; the annotation library includes: semantic text information of each road name.

[0213] The road line extraction module is used to perform vector extraction on the road surface of the target area by using a reclassification extraction method to obtain the road elements of the target area.

[0214] The matching module is used to match the annotation library of the target area with the road elements of the target area to determine the name attributes of the road elements of the target area.

[0215] The road information integration module is used to determine the road entity of the target area according to the name attribute of the road element of the target area and the road element of the target area.

[0216] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store electronic map road information extraction data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for extracting road information from an electronic map is implemented.

[0217] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0218] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0219] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0220] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0221] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0222] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0223] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for extracting road information from an electronic map, characterized in that: The electronic map road information extraction method comprises: Acquire a target map slice set; the target map slice set includes map slices of a target area, and the map slices include road slices and annotation slices; Using a progressive registration method, the target map slice set is registered and corrected to obtain a road grid to be processed and an annotation grid to be processed; Inputting the road grid to be processed into a road surface extraction model to extract the road surface of the target area; the road surface extraction model is obtained based on a ResNet-34 model; Inputting the annotation grid to be processed into an annotation information extraction model to obtain the single-character discrete text of the target area, and determining the center pixel coordinates of the single-character discrete text of the target area as the single-character discrete text coordinates of the target area; the annotation information extraction model is obtained based on a deep learning neural network; According to the annotation grid to be processed and the single-character discrete text coordinates of the target area, an annotation library of the target area is obtained; the annotation library includes: semantic text information of each road name; Using a reclassification extraction method to perform vector extraction on the road surface of the target area to obtain road elements in the target area; Matching the annotation library of the target area with the road elements of the target area to determine the name attributes of the road elements of the target area; The road entity of the target area is determined according to the name attribute of the road element of the target area and the road element of the target area.

2. The method for extracting road information from an electronic map according to claim 1, characterized in that: The target map slice set is registered and rectified by using a progressive registration method to obtain a road grid to be processed and an annotation grid to be processed, which specifically includes: The target map slice set is spliced ​​and merged in sequence according to the row and column numbers of the map slices of the target area to form a large grid; Creating a grid using the boundary line of each map slice in the large grid as a grid line, and marking the coordinates of the map slice in the large grid based on the grid; The road grid to be processed and the annotation grid to be processed in the large grid are determined by using an N-iteration method; wherein the process of the n-th iteration includes: Selecting the registered grid range of the nth iteration from the large grid according to the maximum number of grid rows and columns of the nth iteration, the set registration accuracy and the coordinates of the map slice; 1≤n≤N; Take out the road slices and annotation slices within the registered grid range of the nth iteration from the target map slice set; The road slices in the range of the registered grid of the nth iteration are spliced ​​to obtain a spliced ​​road grid; the annotation slices in the range of the registered grid of the nth iteration are spliced ​​to obtain a spliced ​​annotation grid; The coordinates of the spliced ​​road grid and the spliced ​​annotation grid are transformed by using a second-order polynomial transformation method to form an n-th iteration registered road grid and an n-th iteration registered annotation grid. If n is equal to N, the registered road grids of the previous n iterations are spliced ​​to obtain the road grid to be processed, and the registered annotation grids of the previous n iterations are spliced ​​to obtain the annotation grid to be processed; If n is not equal to N, the n+1th iteration is performed.

3. The method for extracting road information from an electronic map according to claim 2, characterized in that: Selecting the registered grid range of the nth iteration from the large grid according to the maximum number of grid rows and columns of the nth iteration, the set registration accuracy and the coordinates of the map slice, specifically including: According to the maximum number of grid rows and columns of the nth iteration, the set registration accuracy and the coordinates of the map slice, the registered grid range of the nth iteration is selected from the large grid using the M-iteration method; wherein the process of the mth iteration includes: Determine the maximum number of grid rows and columns of the mth iteration, and select the registration range of the mth iteration in the large grid according to the maximum number of grid rows and columns of the mth iteration; 1≤m≤M; Obtaining the registration vector data within the registration range of the mth iteration; Determine the control points of the marker within the registration range of the mth iteration and the control points of the same marker corresponding to the registration vector data as a pair of control points, and select a set number of control point pairs to form a set of control point pairs within the registration range of the mth iteration; A second-order polynomial transformation method is used to transform the coordinates of the large-scale grid according to the control point pair set and the coordinates of the map slice within the registration range of the mth iteration, so as to obtain the large-scale grid after the mth iteration transformation; the large-scale grid after the mth iteration transformation and the registration vector data within the registration range of the mth iteration are in the same coordinate system; According to the control point pair set within the registration range of the mth iteration and the large grid transformed after the mth iteration, the registration accuracy within the registration range of the mth iteration is calculated; Determine whether m is equal to M; If m is equal to M, the registration range of the mth iteration is used as the registered grid range of the nth iteration; If m is not equal to M, then determine whether the difference between the registration accuracy within the registration range of the mth iteration and the set registration accuracy is within the set threshold range; If the difference is within the set threshold range, the registration range of the mth iteration is used as the registered grid range of the nth iteration; If the difference is not within the set threshold range, and the registration accuracy within the registration range of the mth iteration is less than the set registration accuracy, the maximum number of grid rows and columns of the mth iteration is increased, and the increased maximum number of grid rows and columns of the mth iteration is determined as the maximum number of grid rows and columns of the m+1th iteration, and the m+1th iteration is performed; If the difference is not within the set threshold range, and the registration accuracy within the registration range of the mth iteration is greater than or equal to the set registration accuracy, the maximum number of grid rows and columns for the mth iteration is reduced, and the reduced maximum number of grid rows and columns for the mth iteration is determined as the maximum number of grid rows and columns for the m+1th iteration, and the m+1th iteration is performed.

4. The method for extracting road information from an electronic map according to claim 1, characterized in that: The method for determining the road surface extraction model is: Acquire a training road slice set; the training road slice set includes road slices in a map slice of a training area; Using a progressive registration method, the training road slice set is registered and corrected to obtain a training road grid; Extracting road lines in the training road grid, and determining the road surface of the training area according to the road lines; The training road grid and the road surface of the training area are used to train a ResNet-34 model to obtain the road surface extraction model.

5. The method for extracting road information from an electronic map according to claim 1, characterized in that: The method for determining the annotation information extraction model is: Acquire a training annotation slice set; the training annotation slice set includes annotation slices in the map slices of the training area; Using a progressive registration method, the training annotation slice set is registered and corrected to obtain a training annotation grid; Locating the minimum area containing each annotation in the training annotation grid to obtain the minimum training rectangular selection area of ​​each annotation; Taking pixels as basic units, segmenting the annotations in the minimum training rectangular selection area to determine single-character discrete text in the training area; The training annotation grid and the single-character discrete text in the training area are used to train a deep learning neural network to obtain the annotation information extraction model.

6. The method for extracting road information from an electronic map according to claim 1, characterized in that: According to the annotation grid to be processed and the single-character discrete text coordinates of the target area, an annotation library of the target area is obtained, which specifically includes: Locating the minimum area containing each annotation in the annotation grid to be processed to obtain the minimum rectangular selection area to be processed for each annotation; Constructing a vector surface of the minimum rectangular selection area to be processed according to the edge line coordinate string of the minimum rectangular selection area to be processed; Performing spatial intersection on the vector surface of the minimum rectangular selection area to be processed of all annotations and the coordinates of the single-character discrete text to obtain the intersecting characters of all annotations; According to the character segmentation order, all the intersecting characters of the annotations are combined into semantic text information that conforms to the actual attributes of the features; The annotation library is constructed according to the semantic text information.

7. The method for extracting road information from an electronic map according to claim 1, characterized in that: The reclassification extraction method is used to extract vectors from the road surface of the target area to obtain the road elements of the target area, specifically including: Binarizing the pixels in the road surface of the target area to obtain a binary grid; Processing the binary grid using an edge detection algorithm to obtain an object outline in the binary grid; vectorizing the object outline into surface elements; Extracting the skeleton line of the surface element to obtain a road line vector; The road line vectors are converted to obtain road elements in the target area; the conversion process includes connecting broken lines, merging broken lines and deleting branch roads.

8. An electronic map road information extraction device, characterized in that: The electronic map road information extraction device comprises: A slice acquisition module, used to acquire a target map slice set; the target map slice set includes map slices of a target area, and the map slices include road slices and annotation slices; A registration and splicing module is used to register and correct the target map slice set by using a progressive registration method to obtain a road grid to be processed and an annotation grid to be processed; A road surface extraction module, used to input the road grid to be processed into a road surface extraction model to extract the road surface of the target area; the road surface extraction model is obtained based on a ResNet-34 model; An annotation information extraction module is used to input the annotation grid to be processed into an annotation information extraction model to obtain the single-character discrete text of the target area, and determine the center pixel coordinates of the single-character discrete text of the target area as the single-character discrete text coordinates of the target area; the annotation information extraction model is obtained based on a deep learning neural network; An annotation library determination module is used to obtain an annotation library of the target area according to the annotation grid to be processed and the single-character discrete text coordinates of the target area; the annotation library includes: semantic text information of each road name; A road line extraction module, used to perform vector extraction on the road surface of the target area by using a reclassification extraction method to obtain road elements of the target area; A matching module, used for matching the annotation library of the target area with the road elements of the target area to determine the name attributes of the road elements of the target area; The road information integration module is used to determine the road entity of the target area according to the name attribute of the road element of the target area and the road element of the target area.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for extracting road information from an electronic map according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for extracting electronic map road information according to any one of claims 1 to 7 is implemented.

Images