RU2803880C1 - Method and device for map generalization - Google Patents

Abstract

FIELD: map generalization.

SUBSTANCE: invention relates to a method and device for generalizing a map. The method comprises the steps of: a) obtaining cartographic information in vector form, containing at least one layer containing a set of multipolygons for objects of the same type; b) perform rasterization of the received cartographic information; c) perform filtering of rasterized cartographic information; d) perform vectorization of filtered rasterized cartographic information, and in the process of vectorization the following is carried out: finding reference points on the boundaries of adjacent objects, constructing object boundaries based on the found reference points, constructing at least one common boundary between adjacent objects by combining common vertices with the same coordinates, restoring a set of multipolygons taking into account at least one common boundary of objects; e) forming a generalized map based on the reconstructed set of multipolygons.

EFFECT: improved quality of map generalization.

10 cl, 7 dwg

Description

TECHNICAL FIELD

[0001] The claimed solution relates to the field of digital mapping, in particular to the generalization of cartographic images.

BACKGROUND OF THE ART

[0002] The prior art has traditionally used a method of manual map generalization, the effectiveness of which is based on the experience and knowledge of the cartographer.

[0003] The disadvantages of manual generalization include:

• large time costs for the work of cartographers,

• a significant difference in the quality of the final result depending on the level of skill of the performer (cartographer),

• poor scalability of the approach with an increase in the number of cartographic data,

• the impossibility of complete generalization even of the data set that existed at the time the approach was used.

[0004] With the development of scientific and technological progress, in particular in the field of digital cartography, automation of the generalization process has become possible. For example, automated generalization using geometry simplification algorithms from the spatialite library [1].

[0005] The disadvantage of this solution is that individual geometries are considered detachedly and separately, without taking into account the topological connections between various objects. In this approach, it is impossible to take into account the connection between individual, independent geometries, for example, if two geometric polygons have a common boundary, then as a result of simplification one can get a gap along the contact boundary (the so-called gap effect).

SUMMARY OF THE INVENTION

[0006] Geographic information systems provide users with the ability to view different geographic areas, including at different scales depending on their purposes. Such functions require cartographic generalization of maps, namely the selection and generalization of objects depicted on the map according to the purpose and scale, content of the map and the characteristics of the territory being mapped. An important task is to maintain the connectivity of elements at any scale.

[0007] The task of improving the map generalization mechanism was to move away from a completely manual method due to significant labor costs and poor scalability, and to solve the problems of automated simplification, which consisted of unsatisfactory processing and drawing of curves, as well as a lack of coherence in the geometry of elements.

[0008] This technical solution is aimed at eliminating the disadvantages inherent in existing solutions known from the prior art.

[0009] The claimed invention allows us to solve the technical problem of ensuring the formation of a map with high generalization quality.

[0010] The technical result is to improve the quality of map generalization.

[0011] The claimed technical result is achieved by performing a computer-implemented method for generalizing a map, performed by at least one processor, and containing the steps of:

a) receive cartographic information in vector form, containing at least one layer containing a set of multipolygons for objects of the same type;

b) perform rasterization of the received cartographic information;

c) perform filtering of rasterized cartographic information;

d) perform vectorization of the filtered rasterized cartographic information, during which a reconstructed set of multipolygons is obtained;

e) form a generalized map based on the reconstructed set of multipolygons.

[0012] In one of the particular implementation examples, the claimed method additionally contains the stage of pre-processing the received cartographic information before performing rasterization.

[0013] In another particular implementation example, the stage of pre-processing cartographic information includes:

• calculation of the required color depth,

• calculating a common bounding box for all layers containing multipolygon sets,

• determining the tile size taking into account the calculated color depth,

• based on the calculated common bounding rectangle and tile size, the processed cartographic information is divided into tiles (tiling).

[0014] In another particular implementation example, the color depth is calculated using the formula:

bits-per-pixel=ceil(log2(N+1)),

where N is the number of layers of the obtained cartographic information.

[0015] In another particular implementation, each layer has an identifier corresponding to a unique color expressed as a bit sequence, the value of the bit sequence depending on the calculated color depth.

[0016] In another particular implementation example, steps b)-d) are processed line by line, where each line representing a set of multiple tiles is processed in parallel and independently of the other lines.

[0017] In another particular example of implementation, rasterization of the received cartographic information is carried out using a line-by-line scanning algorithm (scanline algorithm).

[0018] In another particular implementation example, the step of filtering rasterized map information includes:

• applying a set of convolutional filters to rasterized map information, and/or

• calculation of the area of rasterized objects in pixels and removal of objects whose area does not meet the threshold value.

[0019] In another particular implementation example, filters used for morphological transformations are used as convolutional filters.

[0020] In another particular implementation example, the step of vectorizing filtered rasterized map information includes:

• finding reference points on the boundaries of adjacent objects,

• construction of object boundaries based on found reference points,

• construction of at least one common boundary between adjacent objects by combining common vertices with the same coordinates,

• restoration of a set of multipolygons, taking into account at least one common boundary of objects.

[0021] The claimed technical result is also achieved through the implementation of a device for generalizing a card containing

at least one processor,

at least one memory coupled to the processor and containing computer-readable instructions that, when executed by the at least one processor, perform the map generalization method.

BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In FIG. Figure 1 shows a block diagram of a computer-implemented method for generalizing a map.

[0023] In FIG. Figure 2 shows a diagram of the implementation of a computer-implemented method for generalizing a map using the example of specific objects.

[0024] In FIG. Figure 3 shows a diagram of the implementation of the process of calculating the general bounding box.

[0025] In FIG. Figure 4 shows a diagram of the implementation of the cartographic image tiling process.

[0026] In FIG. Figure 5 shows examples of the use of various filtering parameters.

[0027] In FIG. Figure 6 shows a diagram of the implementation of the vectorization process.

[0028] In FIG. Figure 7 shows a general diagram of a computing device.

IMPLEMENTATION OF THE INVENTION

[0029] Concepts and terms necessary to understand the present invention will be described below.

[0030] Cartography is a field of science, technology and production covering the study, creation and use of cartographic works [2].

[0031] A cartographic work is a work the main part of which is a cartographic image [2].

[0032] A map is a reduced, generalized image of the surface of the Earth, the surface of another celestial body or extraterrestrial space, constructed in a cartographic projection, showing the objects located on them in a certain system of conventional symbols [2].

[0033] Digital cartography is a branch of cartography that covers the theory and practice of creating and using digital cartographic products [3].

[0034] Digital cartographic products are products obtained using digital cartographic information [3].

[0035] Cartographic generalization is the selection, generalization, and identification of the main typical features of depicted objects in accordance with the purpose, scale, content of the map, features of the mapped territory and the object itself, and the degree of their knowledge [4].

[0036] Cartographic information is information expressed in the form of cartographic works, containing information about them, required for their creation and updating [2].

[0037] Digital map information is map information presented in digital form [3].

[0038] Vector form of representation (digital cartographic information) is a method of representing metric cartographic information as a sequence of vectors [3].

[0039] Raster form of representation (of digital cartographic information) is a method of presenting digital cartographic information in the form of a matrix, the elements of which are the color codes of the cartographic image [3].

[0040] Vectorization of digital cartographic information is the conversion of digital cartographic information from a raster form of representation to a vector one [3].

[0041] Geometries are special geometric objects on the map, for example, such as: points, multipoints, circles, polylines, multipolylines, polygons, multipolygons.

[0042] A polygon (polygon) is a geometry that is a closed broken line and a part of the plane limited by it.

[0043] A multipolygon is a geometry consisting of one or more polygons.

[0044] Tiles (from the English tiles - tiles) in digital cartography are rectangular fragments of equal size, created separately for each zoom level, into which the cartographic image is divided.

[0045] Tiling (in digital cartography) is the process of dividing an entire cartographic image or part thereof into rectangular fragments (tiles) in order to simplify and/or facilitate the rendering of the final cartographic image, or to implement special filtering functions.

[0046] A layer is a set of multipolygons/pixels (depending on the form of data representation - vector or raster) for objects of the same type.

[0047] A layer identifier is a bit sequence corresponding to a unique color.

[0048] Scale (units-per-pixel parameter) is a parameter that specifies the relationship between distance units at the equator and pixels in the rasterized image.

[0049] Bit depth (bits-per-pixel parameter) is a parameter that characterizes the minimum number of bits required to encode a set of unique layers.

[0050] An Axis Aligned Bounding Box (AABB) is a rectangle whose four axes are aligned with the coordinate system in which it resides.

[0051] A Bounding Box is a simple figure (usually a parallelepiped) that limits the shape of a more complex geometric model.

[0052] In FIG. 1 presents a computer-implemented method (100) for generalizing a map. The method (100) is carried out using at least one processor. In a particular implementation example, method (100) is carried out using a device for generalizing a map, which can be implemented on the basis of a computing device modified in software and hardware in such a way as to perform the functions of a device for generalizing a map. A more detailed description of the computing device is disclosed below with reference to FIG. 7.

[0053] At the first stage (101), map information is obtained in vector form, containing at least one layer containing a set of multipolygons for the same type of objects. Each layer has an identifier corresponding to a unique color expressed as a bit sequence. Receiving map information in vector form in a particular implementation example can be implemented using a network communication tool, described in detail below with reference to FIG. 7. The mentioned similar objects can be, but are not limited to, the following examples: forests, rivers, houses, roads, etc.

[0054] In one of the alternative embodiments of the claimed solution, the method (100) further comprises a step of pre-processing the received cartographic information, carried out after the step (101) and before performing the rasterization step (102).

[0055] In a particular implementation example, the stage of pre-processing of cartographic information may include the following sub-stages:

• calculation of the required color depth,

• calculating a common bounding box for all layers containing multipolygon sets,

• determining the tile size taking into account the calculated color depth,

• based on the calculated common bounding rectangle and tile size, the processed cartographic information is divided into tiles (tiling).

[0056] The purpose of the bits-per-pixel calculation substep is to find the minimum number of bits required to encode a set of layers. In a particular implementation example, the color depth parameter is calculated according to the relation N+1<=2 ^∧ (bits-per-pixel), where N is the number of layers of the obtained cartographic information, and the power of two is the minimum. For example, if the number of layers is 7, then the parameter bits-per-pixel=3 (the minimum degree is 3, and the expression takes the following form: 7<=2 ^∧ 3=8). In other words, to find the depth you need to evaluate the expression ceil(log2(N+1)).

[0057] All layers are numbered from 1 to N, where N is the number of layers participating in the generalization process. The number N is used to calculate the required depth for cartographic information (images), i.e. the number of bits needed to encode a given set of layers as colors. Thus, each layer has its own unique color, expressed as a bit sequence, and the value of the bit sequence depends on the calculated color depth.

[0058] In a particular implementation example, information about the number of layers is contained in the map information obtained in step (101), and can be extracted from the map information to calculate the color depth. In another particular implementation example, the number of layers (N) is a predetermined parameter and depends on the scale (units-per-pixel parameter); the lower the value of the units-per-pixel parameter, the more layers are needed to carry out the generalization process.

[0059] It should be noted that in the preferred embodiment of the claimed solution, the default map information contains a background layer, which is numbered as layer zero. For example, if the cartographic information contains three layers (N=3), then the following bit sequence values can be used as layer identifiers (colors):

Layer 0 (background): 0b00

1st layer: 0b01

2nd layer: 0b10

3 layer: 0b11

[0060] In the following, layer identifiers are used, for example:

• at the rasterization stage (102), to distinguish different layers in the resulting cartographic information (image);

• at the vectorization stage (104) to restore the set of multipolygons.

[0061] As an additional benefit of using an identifier corresponding to a unique color, it can be noted that it is significantly easier to locate emerging defects in the map image during generalization due to the visually distinguishable color difference between layers.

[0062] It should be noted that image processing tasks (in particular, cartographic images) are mostly memory-bounded tasks, i.e. refer to a situation in which the execution time of a given computational task is determined primarily by the amount of free memory required to store working data. In other words, the limiting factor in solving this problem is the speed of memory access. This distinguishes this type of problem from problems tied to calculations, where the number of elementary calculation steps is a decisive factor. Thus, organizing optimal access to shared memory can significantly influence the performance of the proposed solution.

[0063] Calculating the minimum required color depth allows you to optimize memory consumption, as well as speed up further processing of map information. Due to the arrangement of data in memory in a more compact manner, and, as a result, allowing for more cache-friendly read/write operations in RAM.

[0064] The process of computing a common bounding box for all layers containing multipolygon sets is illustrated in FIG. 3. The first diagram shows several sets of multipolygons (311) belonging to different layers (which is also evidenced by the unique color of each layer).

[0065] For each set of multipolygons from each individual layer, an axis-aligned bounding box (AABB) is calculated (301) based on the scale parameter (units-per-pixel). The units-per-pixel parameter specifies the relationship between distance units at the equator and pixels in the rasterized map image. For example, if the resulting cartographic information in vector form is measured in meters, then units-per-pixel=100 means that 1 pixel will draw data from a 100x100 meter square at the equator (due to the specifics of the projection used). In a particular implementation example, the scale parameter (units-per-pixel) is contained in the map information obtained in step (101) and can be extracted from the map information to calculate bounding boxes (301). In the second diagram of Fig. 3 illustrates sets of multipolygons from different layers, each of which is enclosed in a bounding box (312).

[0066] Next, using the calculated set of bounding boxes (312), a common bounding box (313) is constructed (302), which limits all sets of multipolygons from all layers. This process is carried out using the extension operation (extend).

[0067] The pseudocode of the extension operation can be expressed as follows:

[0068] Once the overall bounding box has been calculated, i.e. the size of the cartographic information required for further rasterization is known, the tiling stage is performed, i.e. dividing the processed cartographic information (in the particular case, a cartographic image) into rectangular fragments (tiles).

[0069] Before dividing into tiles, it is necessary to determine the tile size based on the calculated color depth. The tile size (tile-size parameter) allows you to divide the processed area of cartographic information (i.e., a cartographic image) into rectangular fragments (tiles) of size tile-size x tile-size pixels, and in each of the tiles it is possible to use its own color palette of size 2 ^∧ (bits-per-pixel) - 1, which allows you to increase the number of processed layers and optimize consumed computing resources by significantly saving memory used. Thus, the number of tiles into which the cartographic image will be divided depends on the tile-size parameter. The tile-size parameter is selected so that the number of layers in one tile does not exceed 2 ^∧ (bits-per-pixel) - 1, and the number of tiles is minimal.

[0070] In a particular implementation example, parameters such as tile-size and the number of tiles are selected using tools for automatic selection of parameters. Such tools may include, but are not limited to, the following examples: GridSearchCV [5], RandomizedSearchCV [6], Hyperopt [7].

[0071] In an alternative embodiment of the claimed solution, the tile-size and number of tiles are determined using automatic machine learning (AutoML). AutoML is the process of automating algorithm selection, feature generation, hyperparameter tuning, iterative modeling, and model evaluation.

[0072] The AutoML service [8] implemented on the basis of the AI Cloud ML Space platform can be used as AutoML, but is not limited to. This service allows, based on pre-prepared datasets, to determine optimal parameter values, such as tile-size and number of tiles, while optimal parameter values are understood as parameter values that satisfy the following conditions:

• in each tile the number of layers should be no more than 2 ^∧ (bits-per-pixel) - 1,

• the number of tiles should be minimal.

[0073] Next, based on the calculated common bounding box and tile size, the processed map information is divided into tiles (tiling). In a particular implementation example, all further processing is done in tiles, which allows you to encode a larger number of layers without increasing the color depth. In another particular implementation example, each tile is assigned coordinates that determine the tile's position in space. In FIG. Figure 4 shows a diagram of the implementation of the cartographic image tiling process. The illustrated cartographic image is divided into 9 tiles (401), and the maximum number of layers in each of the tiles (401) does not exceed 3, taking into account the bits-per-pixel=3 parameter, i.e. 2 ² -1=3.

[0074] In multiprocessor (multi-threaded) environments, processing at stages (102-104) is carried out line by line, where each line (411, 412, 413), representing a set of several tiles, is processed in parallel and independently of the remaining lines, which allows for greater performance as stated solutions.

[0075] At step (102), rasterization of the map information obtained at step (101) is performed. In a particular implementation example, rasterization can be carried out based on at least such parameters as scale (units-per-pixel), color depth (bits-per-pixel), number of tiles and their coordinates, and these parameters are obtained or calculated on the previous stages.

[0076] In the preferred embodiment of the claimed solution, the stage (102) of rasterization of the received cartographic information is implemented using a line-by-line scanning algorithm (scanline algorithm [9]). In a particular example of the implementation of the scanline algorithm, it is implemented taking into account the preliminary implementation of tiling of cartographic information. The essence of the implementation of the scanline algorithm taking into account tiling is that:

• each tile is processed independently and in parallel,

• within each tile, processing is carried out line by line, i.e. pixel lines are rasterized sequentially from top to bottom.

[0077] As a result of step (102), rasterized cartographic information is obtained, which in a particular embodiment can be presented as an analogue of an RGB image containing coherent (in a topological sense) and coarse geometry (due to scaling with the units-per-pixel parameter ), where each color has a unique layer identifier. It should be noted that different scales (units-per-pixel parameter) allow you to obtain geometry of different detail, i.e. The lower the units-per-pixel value, the more detailed the image we get. The rasterized cartographic information obtained in this way (in a particular case, a rasterized cartographic image) allows further processing without separate processing of the situation of poorly adjacent adjacent curves (between objects with a common boundary).

[0078] In FIG. Figure 2 shows a diagram of the implementation of a computer-implemented method for generalizing a map using the example of specific objects. Such objects in the field (221) may include, but are not limited to: a blue triangle, a red pentagon with an inscribed square, an orange square. Objects are presented in vector form and consist of multipolygons. Different colors of objects indicate that they belong to different layers, since, as noted earlier, color is a layer identifier.

[0079] After rasterization is performed (201), the objects are converted from vector form to raster form. Field (222) illustrates an example of such a transformation, where the above objects are represented as a set of pixels, and the objects have a coarsened geometry compared to the vector form of representation.

[0080] At step (103), filtering of the map information rasterized at step (102) is performed. The filtering step (103) helps eliminate gaps between objects that arise, for example, when changing the scale, as well as smoothing the boundaries of objects, thus improving the quality of map generalization.

[0081] One particular example of the implementation of filtering rasterized cartographic information includes:

• applying (202) a set of convolutional filters to the rasterized map information, and/or

• calculation of the area of rasterized objects in pixels and removal (203) of objects whose area does not meet the threshold value (filtering by area).

[0082] It should be noted that in one of the alternative embodiments of the claimed solution, filters used for morphological transformations are used as convolutional filters. By way of example, but not limitation, filters used for morphological transformations are based on dilation and/or erosion operations. Application (202) of a set of convolutional filters to rasterized cartographic information (in particular, to a rasterized cartographic image) additionally allows you to combine objects between which there are gaps (gaps), smooth out the boundaries, thereby eliminating visual flaws in the image, and as a result improve the quality of generalization cards.

[0083] An example application (202) of a convolutional filter bank is illustrated in FIG. 2 in field (223), where convolutional filters are applied to objects represented as a set of pixels.

[0084] In an alternative embodiment of the claimed solution, filtering (103) of rasterized map information further includes the step of obtaining filtering parameter values, preceding the above operations (202, 203). In this case, the application of a set of convolutional filters to rasterized cartographic information is based on the obtained values of the filtering parameters.

[0085] In one particular implementation example, the values of the filtering parameters can be calculated using a machine learning model pre-trained on a large number of map images. For example, such a machine learning model can be based on a neural network, where the hidden layer neurons are filtering parameters, and their weights are the values of the filtering parameters.

[0086] In another particular implementation example, filtering parameters can be selected based on quality metrics of the resulting generalized cartographic images. Quality metrics may include, but are not limited to, the following quality metrics.

[0087] Metric 1. Comparison of the area of objects.

[0088] We calculate the area value for the initial objects Si and for the generalized objects GSj. After this, we analyze the relationship

|SUM S _i - SUM GS _j |/SUM S _i → 0%,

where SUM is the sum of the areas of objects.

[0089] The deviation should tend to 0%, i.e. the smaller the deviation, the better the generalization was performed.

[0090] Metric 2. Number of polygons and number of points.

[0091] We calculate the number of MP polygons and the number of P points for the original objects, as well as the number of GMP polygons and GP points for the generalized objects.

[0092] An indicator of high-quality generalization is the fact that GMP<MP, i.e. the number of generalized objects is always less than or equal to the original ones, GP<P, i.e. the number of points of generalized objects is always less than that of the original objects.

[0093] As an example, but not limited to it, we correlate the initial geometry with a conditional 19th zoom (English zoom - changing the image scale). When scaling the geometry in the 13th zoom, the guideline for reducing the number of points is calculated as ~2 ^(19-13) , i.e. the number of points with the specified change in scale should decrease by 64 times compared to the original value.

[0094] Metric 3: Deviation of matched points.

[0095] We search for OP _i (x _i , y _i ) initial points for GP _j (x _j , y _j ) generalized points in the radius for maximum proximity in x and y coordinates (i.e., we solve the problem of searching for the nearest neighbor in a given radius). Based on the matched points (their number k≤j), we calculate the maximum deviation X _k and Y _k from the coordinates of the points.

[0096] We analyze MAX and MIN from deviations, as well as median and average values. This mechanism will allow not only making quality measurements, but also generally assessing the displacement of objects, which allows for alignment between the original and generalized geometries.

[0097] Metric 4. Line deviation.

[0098] For a generalized geometry, points GPj and GPj+1 are selected in pairs to form a segment, with the points being selected sequentially in a clockwise direction. A network of perpendiculars is built to the segment until it intersects with the original geometry. Perpendiculars are constructed using a grid; the option of a uniform and non-uniform grid can be selected.

[0099] The resulting perpendiculars are deviations for segments in the generalized geometry. For calculations, the average deviation along perpendiculars is used, the MAX and MIN of deviations, as well as the median and mean values are analyzed.

[0100] It should be noted that each of the presented metrics can be used independently and independently, but the best results are achieved by combining quality metrics. For example, when comparing the area of objects using metric 1, it is extremely effective to use metric 2 so that the area estimate is related to the number of polygons and points.

[0101] Filtering parameters may include, but are not limited to, the following examples: shape of the filter core, filter size (filter-radius), filter strength (object-coef), minimum allowable area (min-object-area). Examples of the use of various filtering parameters are clearly illustrated in Fig. 5.

[0102] Various geometric shapes can be used as the shape of the filter core, however, the preferred embodiment is a circle as it gives the best result (based on A/B testing).

[0103] The filter size parameter (filter-radius) affects the “roundness” of the geometry; the larger the value of the filter-radius parameter, the more “rounded” the geometry is obtained. The influence of the filter-radius parameter on the filtering result is clearly illustrated in Fig. 5 (item 501).

[0104] The value of the object-coef parameter affects the “aggressiveness” of the filter, values object-coef<1 compresses the geometry, values object-coef>1 stretches the geometry. The influence of the object-coef parameter on the filtering result is clearly illustrated in Fig. 5 (item 502).

[0105] The minimum acceptable area parameter (min-object-area) characterizes the minimum acceptable area in pixels of at least one object from the layer. This approach allows you to calculate the area of rasterized objects in pixels and remove those objects whose area is less than the minimum acceptable area parameter (min-object-area). The influence of the min-object-area parameter on the filtering result is clearly illustrated in Fig. 5 (item 503).

[0106] The min-object-area parameter is associated with a scale parameter (units-per-pixel), which specifies the relationship between equator distance units (meters) and pixels in the rasterized image. Thus, the relationship between the min-object-area and units-per-pixel parameters specifies the relationship between the area in meters and the area in pixels.

[0107] An example application (203) of area filtering is illustrated in FIG. 2 in field (224), where in the process of comparing the area of objects in pixels with the threshold value (i.e., the minimum allowable area parameter), an object was removed - an orange square, the area of which was less than the value of the min-object-area parameter.

[0108] In an alternative embodiment of the claimed solution, the step of applying a bank of convolutional filters is carried out directly at the rasterization step (102). The symmetry of the filter kernel ([10], [11]) makes it possible to apply the filter row by row rather than per pixel, similar to the optimizations for the Gaussian filter [12], which allows for more optimal memory access and, as a result, higher performance.

[0109] At step (104), vectorization of the rasterized map information filtered at step (103) is performed.

[0110] In one particular implementation example, the step of vectorizing (104) filtered rasterized cartographic information includes:

• finding reference points on the boundaries of adjacent objects,

• construction of object boundaries based on found reference points,

• construction of at least one common boundary between adjacent objects by combining common vertices with the same coordinates,

• restoration of a set of multipolygons, taking into account at least one common boundary of objects.

[0111] The implementation of the steps of the vectorization process is illustrated in FIG. 6.

[0112] The vectorization process (104) searches for reference points on the boundaries of adjacent objects. The reference points are pixels on a rasterized cartographic image that have a certain property, namely, the neighbors of such pixels are of different colors, i.e. belong to different layers. At the stage of finding reference points, the coordinates of such pixels are searched for whose neighbors have 3 or more different layer identifiers (colors). Boundary cases are processed separately, for example, points in the corners of the image (0, 0), (0, h), (w, 0), (w, h). The background color (i.e. pixels not occupied by objects) acts as a separate identifier (background layer ID) and is taken into account when searching for anchor points. The goal of the stage is to find reference points at the junctions of objects belonging to different layers (taking into account the background layer without objects), which are subsequently used to construct a graph of feasible segments, which serves to construct the boundaries of objects.

[0113] The found reference points subsequently make it possible to construct a common boundary between separate adjacent objects using common vertices with the same coordinates. The construction of a common boundary allows you to optimize the vectorization process by reusing common boundaries when converting filtered rasterized cartographic information into a vector form of representation (i.e., when converting from sets of pixels to multipolygons). Referring to FIG. 6, for example, when vectorizing a “yellow” object, characterized by a set of yellow pixels, we build the border of the “yellow” object, and if part of the border is common with another object (“red” object, characterized by a set of red pixels), then we use the values of the reference points of the common border to construct the border of the “red” object. Similarly, to construct the boundary of a “green” object, we use the values of the reference points of the common boundary between the “red” and “green” objects.

[0114] This approach allows you to avoid errors when constructing boundaries at the vectorization stage, the so-called gap effect, which can arise due to various roundings during vectorization. Thus, restoring a set of multipolygons taking into account at least one common boundary of objects allows us to eliminate the effect of gaps and thereby improve the quality of map generalization.

[0115] An example of performing vectorization (204) is illustrated in FIG. 2 in the field (225), where objects represented as a set of pixels (in raster representation form) subjected to filtering (202, 203) are converted into objects represented as a reconstructed set of multipolygons (in vector representation form). In FIG. Figure 2 clearly demonstrates the difference between the original objects consisting of multipolygons (field (221)) and the transformed objects consisting of restored multipolygons (field (225)). Objects have undergone the following changes: the blue triangle now has rounded (smoothed as a result of filtering) corners; the red pentagon also has rounded (smoothed as a result of filtering) corners, in addition, the square inscribed in it changed its shape to a circle during filtering; The orange square was removed during area filtering.

[0116] At step (105), a generalized map is formed based on the reconstructed set of multipolygons obtained at step (104) of vectorization.

[0117] In a particular implementation example, together with the generalized map, information about the values of the parameters used in steps (102-104) is additionally generated, for example, but not limited to, as metadata [13]. This information can be used to analyze and further improve the map generalization method.

[0118] In FIG. 7 shows a general view of a computing device (700), on the basis of which a device for map generalization can be implemented, providing implementation of the map generalization method.

[0119] In general, a computing device (700) contains one or more processors (701), memory devices such as RAM (702) and ROM (703), input/output interfaces (704), and input devices connected by a common information exchange bus. /output (705), and a means for network communication (706).

[0120] The processor (701) (or multiple processors, multi-core processor) may be selected from a variety of devices commonly used today, such as those from Intel™, AMD™, Apple™, Samsung Exynos™, MediaTEK™, Qualcomm Snapdragon™ and etc. A graphics processor, for example, Nvidia, AMD, Graphcore, etc., can also be used as the processor (701).

[0121] RAM (702) is a random access memory and is designed to store machine-readable instructions executable by the processor (701) to perform the necessary logical data processing operations. The RAM (702) typically contains executable operating system instructions and associated software components (applications, program modules, etc.).

[0122] A ROM (703) is one or more permanent storage devices, such as a hard disk drive (HDD), a solid state drive (SSD), flash memory (EEPROM, NAND, etc.), optical storage media ( CD-R/RW, DVD-R/RW, Blu-Ray Disc, MD), etc.

[0123] To organize the operation of device components (700) and organize the operation of external connected devices, various types of I/O interfaces (704) are used. The choice of appropriate interfaces depends on the specific design of the computing device, which can be, but is not limited to: PCI, AGP, PS/2, IrDa, FireWire, LPT, COM, SATA, IDE, Lightning, USB (2.0, 3.0, 3.1, micro, mini, type C), TRS/Audio jack (2.5, 3.5, 6.35), HDMI, DVI, VGA, Display Port, RJ45, RS232, etc.

[0124] To provide user interaction with the computing device (700), various I/O information devices (705) are used, for example, a keyboard, a display (monitor), a touch display, a touch pad, a joystick, a mouse, a light pen, a stylus, touch panel, trackball, speakers, microphone, augmented reality tools, optical sensors, tablet, light indicators, projector, camera, biometric identification tools (retina scanner, fingerprint scanner, voice recognition module), etc.

[0125] The network communication facility (706) allows the device (700) to transmit data via an internal or external computer network, such as an Intranet, Internet, LAN, or the like. One or more means (706) can be used, but not limited to: Ethernet card, GSM modem, GPRS modem, LTE modem, 5G modem, satellite communication module, NFC module, Bluetooth and/or BLE module, Wi-Fi module and etc.

[0126] Additionally, the device (700) can also use satellite navigation tools, for example, GPS, GLONASS, BeiDou, Galileo.

[0127] The submitted application materials disclose preferred examples of implementation of a technical solution and should not be interpreted as limiting other, particular examples of its implementation that do not go beyond the scope of the requested legal protection, which are obvious to specialists in the relevant field of technology.

[0128] Sources of information:

[1] SpatiaLite - spatial extensions for SQLite. https://www.gaia-gis.it/gaia-sins/spatialite-manual-2.3.1.html

[2] GOST 21667-76 Cartography. Terms and Definitions

[3] GOST 28441-99 Digital cartography. Terms and Definitions

[4] Berlyant A.M. CARTOGRAPHIC GENERALIZATION // Great Russian Encyclopedia. Volume 13. Moscow, 2009, p. 237

[5] Sklearn.mode_lselection.GridSearchCV - scikit-learn 1.2.0 documentation. https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html

[6] Sklearn.mode_lselection.RandomizedSearchCV - scikit-learn 1.2.0 documentation. https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.RandomizedSearchCV.html

[7] Hyperopt Documentation, https://hyperopt.github.io/hyperopt/

[8] AutoML - AI Cloud ML Space documentation. User guide. https://docs.sbercloud.ru/aicloud/mlspace/concepts/automl.html

[9] Scanline rendering. https://en.wikipedia.org/wiki/Scanline_rendering

[10] Separable filter. https://en.wikipedia.org/wiki/Separable_filter

[11] John Chapman. Optimizing Convolutions. https://john-chapman.github.io/2019/03/29/convolution.html

[12] Matrix filters for image processing, https://habr.com/ru/post/142818/

[13] GOST R 52573-2006. Geographic information. Metadata

Claims (29)

1. A computer-implemented method for generalizing a map, performed by at least one processor, containing the steps of: a) receive cartographic information in vector form, containing at least one layer containing a set of multipolygons for objects of the same type; b) perform rasterization of the received cartographic information; c) perform filtering of rasterized cartographic information; d) perform vectorization of filtered rasterized cartographic information, and in the process of vectorization the following is carried out: • finding reference points on the boundaries of adjacent objects, • construction of object boundaries based on found reference points, • construction of at least one common boundary between adjacent objects by combining common vertices with the same coordinates, • restoration of a set of multipolygons, taking into account at least one common boundary of objects; e) form a generalized map based on the reconstructed set of multipolygons. 2. The method according to claim 1, characterized in that it additionally contains the stage of pre-processing the received cartographic information before performing rasterization. 3. The method according to claim 2, characterized by the fact that the stage of preliminary processing of cartographic information includes: • calculation of the required color depth, • calculating a common bounding box for all layers containing multipolygon sets, • determining the tile size taking into account the calculated color depth, • based on the calculated common bounding rectangle and tile size, the processed cartographic information is divided into tiles. 4. The method according to claim 3, characterized by the fact that the color depth is calculated by the formula: bits-per-pixel=ceil(log2(N+1)), where N is the number of layers of the obtained cartographic information. 5. The method according to claim 4, characterized in that each layer has an identifier corresponding to a unique color expressed by a bit sequence, and the value of the bit sequence depends on the calculated color depth. 6. The method according to claim 3, characterized by the fact that at stages b)-d) processing is carried out line by line, where each line, representing a set of several tiles, is processed in parallel and independently of the other lines. 7. The method according to claim 1, characterized by the fact that rasterization of the received cartographic information is carried out using a line-by-line scanning algorithm. 8. The method according to claim 1, characterized in that the stage of filtering rasterized cartographic information includes: • applying a set of convolutional filters to rasterized map information, and/or • calculation of the area of rasterized objects in pixels and removal of objects whose area does not meet the threshold value. 9. The method according to claim 8, characterized by the fact that filters used for morphological transformations are used as convolutional filters. 10. Device for generalizing the map, containing: at least one processor, at least one memory associated with the processor and containing machine-readable instructions that, when executed by at least one processor, enable the method of any one of claims to be carried out. 1-9.

Images