Road topology representation method based on electronic information map

Abstract

The present invention discloses a road topology representation method based on an electronic information map, including: S1, image acquisition and preprocessing: S2: extracting and filtering the text information of the image through text detection and text recognition, and only retaining the road name information; synchronously applying a semantic segmentation model to perform semantic segmentation, extracting the pixel-level road information of the image, and obtaining the pixel-level road information; S3: fusing the pixel-level road information and the road name information to construct a topological map; specifically: extracting the road skeleton from the pixel-level road information, detecting the intersection through neighborhood inspection, and using its coordinates as the intersection coordinates; then obtaining the segmented roads, calculating the connectivity relationship and length between the segmented roads; and performing the topological representation of the road information after matching the text information with the segmented road information. The present invention can accurately extract effective roads, intersections, and text information from complex electronic information maps and realize the topological representation of roads, which is more convenient for computer processing.

Description

Technical Field

[0001] The invention relates to the field of machine vision and geographic information systems, and in particular to a road topology representation method based on an electronic information map. Background Art

[0002] In the digital age, map services are more than just basic navigation tools. They have become an indispensable part of many fields, such as urban planning, traffic management, and autonomous driving. In particular, the accurate representation of road networks is crucial for the efficient operation of intelligent transportation systems. The topological representation of electronic information maps can transform physical information in the real world into digital space and represent it in the form of topology, providing strong support for subsequent tasks such as traffic game models and path planning.

[0003] The traditional process of creating and updating road maps usually relies on manual data collection and manual map editing, which is time-consuming and labor-intensive, and it is difficult to keep up with the rapid changes in urban construction. With the development of technology, automated and intelligent methods for extracting road and intersection information have been rapidly developed. However, for downstream tasks such as traffic games and path planning, the only map information extraction technology currently available cannot well implement an environmental model that is easy for computers to understand. In order to solve this problem, it is particularly necessary and effective to construct a formal topological representation method for electronic information maps. However, there is currently a lack of methods that can well meet this demand. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention proposes a road topology representation method based on electronic information maps. The method aims to accurately distinguish and model different information, objects or areas in the electronic information map through semantic segmentation technology, and construct a road topology relationship based on the intersection position relationship, so as to provide more accurate and intuitive information representation for downstream applications such as navigation, location search, and travel route planning. The present invention not only provides a new, efficient and accurate electronic information map data representation method, but also provides strong technical support for efficient and convenient smart travel based on map software.

[0005] The specific technical solutions are as follows:

[0006] A road topology representation method based on an electronic information map comprises the following steps:

[0007] S1. Image acquisition and preprocessing: Use automatic acquisition software to capture the electronic information image of the electronic information map, and then preprocess the captured image to remove non-electronic map information and only retain the electronic map information;

[0008] S2: extracting text information from the electronic information map obtained in S1, which is specifically achieved through the following sub-steps:

[0009] (1) Use a text detection algorithm to perform text detection on the electronic information map and output the coordinates of the corner points of the detection box;

[0010] (2) calculating the center position of the text according to the corner point coordinates, and capturing the image in the detection frame;

[0011] (3) Use the text recognition model to perform text recognition on the captured image and convert it into text in the form of a string;

[0012] (4) Filter the extracted text information and only retain the road name information;

[0013] The electronic information map obtained by S1 is simultaneously subjected to pixel-level road information extraction, which is specifically achieved through the following sub-steps:

[0014] (1) Enhance the image of the electronic information map by using grayscale processing and histogram equalization to improve image quality;

[0015] (2) Applying the semantic segmentation model to perform semantic segmentation, separate road information and background information and label them separately to obtain pixel-level road information;

[0016] S3: Fusing the pixel-level road information and the road name information to construct a topological map, which is specifically achieved through the following sub-steps:

[0017] S3.1: Use the morphological skeletonization technique Zhang-Suen algorithm to extract the road skeleton from the pixel-level road information;

[0018] S3.2: Define intersections in the road skeleton, detect all intersections through neighborhood checking, and record the coordinates of the intersections as intersection coordinates;

[0019] S3.3: Obtaining segmented roads according to the intersection coordinates and the road skeleton, and calculating the connectivity relationship and length of each segmented road;

[0020] S3.4: Matching text information with segmented road information: Convert the center position of the text to the same coordinate system as the segmented road information, perform a neighbor search for all subsequent road skeletons within the set search area, calculate the distance between the center position of the text and the candidate road skeleton, and take the candidate road skeleton with the closest distance as the matching result of the text information;

[0021] S3.5: Topological representation of road information: The segmented roads are taken as topological nodes, and the topological nodes contain the segmented road length and road name information. The edges of the topological structure represent both the intersection information and the connection relationship between the segmented roads; the intersection information includes the intersection coordinates and the segmented road information connected by the intersection.

[0022] Furthermore, in S3.4, if there is no road skeleton in the circular search area, the radius of the search area is expanded to continue searching until a final result is matched or the radius of the search area exceeds a set threshold.

[0023] Furthermore, the electronic map information in S3.5 includes: roads, buildings, and area names.

[0024] Furthermore, in S1, Multiscreenshots software is selected as the automatic acquisition software, and the electronic information map is selected as the SUMO map.

[0025] Furthermore, in S2, the text detection algorithm uses the EAST algorithm.

[0026] Furthermore, in S2, the text recognition model uses a CRNN deep learning model.

[0027] Furthermore, in S2, the semantic segmentation model uses a VGG-encoder-decoder model, whose encoder is composed of a combination of five groups of convolutional layers and pooling layers. After each group of convolution and pooling operations, the number of channels of the feature map increases and the size decreases; the decoder is composed of a combination of four groups of upsampling layers and convolutional layers. After each group of upsampling and convolution operations, the number of channels of the feature map decreases and the size increases.

[0028] Furthermore, in S3, the Zhang-Suen algorithm refines the road image by iteratively removing pixel edges, compressing the road into a line with a pixel width to obtain a road skeleton, and the line represents the center line of the road.

[0029] The beneficial effects of the present invention are:

[0030] (1) The present invention combines a semantic segmentation model, a text detection algorithm, and a text recognition model to accurately extract effective road, intersection, and text information from complex electronic information maps. This automated information extraction method has advantages in accuracy and processing speed over existing technologies.

[0031] (2) The present invention obtains the intersection information of roads through morphological skeletonization technology and neighborhood check algorithm, effectively constructs the center line of the road network, and accurately identifies and locates intersections. This method makes the expression of road information more intuitive and structured, and more convenient for computers to understand and process.

[0032] (3) The present invention combines the extracted pixel-level road information and road name information to realize the topological representation of road information. This method improves the availability of map information and provides effective data support for downstream tasks such as navigation, positioning, and path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a flow chart of a road topology representation method based on an electronic information map in an embodiment of the present invention.

[0034] Figure 2 It is a structural diagram of the VGG-encoder-decoder model of the semantic segmentation model in an embodiment of the present invention.

[0035] Figure 3 Schematic diagram of the eight-neighborhood of intersection pixels when calculating intersection information in an embodiment of the present invention, wherein (a) is a schematic diagram of the eight-neighborhood of intersection pixels at a crossroad, and (b) is a schematic diagram of the eight-neighborhood of intersection pixels at a T-junction.

[0036] Figure 4 4 is a flow chart of matching text information with segmented road information in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The present invention will be described in detail below according to the accompanying drawings and preferred embodiments, and the purpose and effect of the present invention will become more clear. The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0038] like Figure 1 As shown, a road topology representation method based on an electronic information map specifically includes the following steps:

[0039] S1: Image acquisition and preprocessing, used to obtain and provide electronic information map data source to provide support for subsequent operations, which is implemented through the following sub-steps:

[0040] S1.1: Use automatic acquisition software to complete image capture of the electronic information map. In this embodiment, Multiscreenshots software is used to capture images of the SUMO map.

[0041] S1.2: Preprocessing the collected electronic information map image to remove irrelevant information so that the image area only contains electronic map information, the electronic map information includes: roads, buildings, area names, etc.

[0042] S2: Extraction of text information and pixel-level road information from electronic information maps, used to identify text information in the electronic information maps preprocessed by S1, and to distinguish road information and background information on the preprocessed electronic information maps. The two steps are performed simultaneously to improve overall efficiency.

[0043] The extraction of text information from the electronic information map is specifically implemented through the following sub-steps:

[0044] (1) Text detection: Use a text detection algorithm to detect the text appearing in the electronic information map. The text detection algorithm uses detection boxes for bounding, and the output is the corner coordinates of the detection boxes. The central position of the text is calculated through the corner coordinates of the detection boxes. In this embodiment, the EAST algorithm is selected as the text detection algorithm.

[0045] (2) Text recognition: Crop out the picture in the detection box and perform text recognition on the cropped picture, that is, use a text recognition model to recognize the text screenshot in picture form into text in string form. In this embodiment, the CRNN deep learning model is selected as the text recognition model.

[0046] (3) Text information screening: The outputs of text detection and text recognition are the central position of the text in the electronic information map and the corresponding text content. The processed text includes road names, building names, area names, etc. Since the present invention aims to construct a topological map of road information, therefore, text information other than road names belongs to invalid information and needs to be screened, that is, irrelevant text information is screened out, and only road name information is left to reduce the subsequent calculation amount. Since road names generally have clear semantic characteristics, such as suffixes with "road", "street", etc., a regular expression method is used to directly screen the text, and this method has high accuracy.

[0047] When performing pixel-level road information extraction of the electronic information map synchronously, to improve the accuracy of information discrimination, a deep learning semantic segmentation method is used to achieve pixel-level road information segmentation. The specific steps are as follows:

[0048] (1) Further process the electronic information map to remove the invalid background information in the preprocessed electronic information map. Specifically, a grayscale processing method is adopted. By adjusting the contrast of the image, the region of interest in the image is enhanced, and histogram equalization is applied to enhance the image contrast, making features such as roads easier to recognize.

[0049] (2) Use a semantic segmentation model to achieve the segmentation of roads and the background. In this embodiment, VGG-encoder-decoder is selected as the semantic segmentation model, and its structure is as Figure 2As shown in the figure, the encoder is composed of a combination of 5 groups of convolutional layers and pooling layers. The convolutional layers correspond to the blue areas in the figure, and the pooling layers correspond to the green areas in the figure. After each set of convolution and pooling operations, the number of channels of the feature map increases and the size decreases; the decoder is composed of a combination of 4 groups of upsampling layers and convolutional layers. The upsampling layers correspond to the yellow areas in the figure. After each set of upsampling and convolution operations, the number of channels of the feature map decreases and the size increases. In the segmentation result, the road edge and the area inside it are marked as Class 1, and the area outside the road edge (i.e., the background area) is marked as Class 0, realizing the pixel-level binary classification task and obtaining pixel-level road information. The loss function of VGG-encoder-decoder uses the cross entropy loss function.

[0050] S3: Information fusion to build a topological map, that is, to fuse the extracted pixel-level road information and the road name information in the text information to achieve topological representation of the road information, which is specifically achieved through the following sub-steps:

[0051] S3.1: Pixel-level road information processing: Morphological skeletonization technology is used to reduce wide shapes to a "road skeleton" of only one pixel width. Specifically, for images of road networks, the Zhang-Suen algorithm is used to refine the image by iteratively removing pixel edges, compressing wide roads into thin lines that can represent the center lines of the roads.

[0052] S3.2: Calculation of intersection information, which is implemented through the following sub-steps:

[0053] (S3.2.1) Intersection points are defined in the road skeleton map at the pixel level, that is, in the skeleton, if there is a pixel point with three or more other pixels in its eight-neighborhood, then the point can be defined as an intersection point.

[0054] (S3.2.2) Detect intersections through a neighborhood check process. According to the definition of intersections, for each pixel point p(x,y) in the road skeleton image, check the number of pixels representing the road in its eight neighborhoods (hereinafter referred to as domain pixels); if the number of neighborhood pixels is greater than or equal to 3, p is considered to be an intersection. Generally speaking, when the number of neighborhood pixels is 3, it indicates a T-junction intersection, and when the number of domain pixels is 4, it indicates a crossroads intersection. Figure 3 As shown in (a), the straight line and the oblique line composed of blue pixels in the schematic diagram of the intersection represent two roads respectively, the yellow pixel represents the intersection of the two roads, and the area circled by the red frame represents the eight neighborhoods of the intersection pixel. The eight neighborhoods contain 4 blue pixels, which can represent the recognition of the intersection; Figure 3As shown in (b), in the schematic diagram of the T-junction, the two straight lines composed of blue pixels represent two roads, the yellow pixels represent the intersection of the two roads, and the area circled by the red frame represents the eight-neighborhood of the intersection pixel. The eight-neighborhood area contains 3 blue pixels, which can represent the recognition of the T-junction.

[0055] (S3.2.3) Adjacent intersections can be merged to reduce false detections, and the coordinates of each final intersection can be recorded as the coordinate representation of the intersection.

[0056] S3.3: Calculation of segmented road length: Based on the intersection coordinates and the road skeleton, calculate the connectivity between the segmented roads to achieve the skeleton division of the segmented roads, and calculate the length of the segmented skeleton based on the scale information to estimate the length of the segmented road.

[0057] S3.4: The text information (specifically the road name information) is matched with the segmented road information, such as Figure 4 As shown, it is implemented through the following sub-steps:

[0058] (S3.4.1) Ensure that all data are in the same coordinate system space. In this embodiment, the initial coordinate system of the text information is the coordinate system based on the text recognition result, and the initial coordinate system of the segmented road information is the geographic coordinate system. Therefore, the center position of the text needs to be converted into geographic coordinates.

[0059] (S3.4.2) Set a search radius, i.e., the maximum distance at which the text information may match the road skeleton, and establish a circular search area based on the search radius.

[0060] (S3.4.3) For each text location, perform a neighbor search in its search area to find all road skeletons in the search area.

[0061] (S3.4.4) Using the shortest distance from a point to a line segment as the calculation principle, calculate the distance from the center of the text to each candidate road skeleton, and take the candidate road skeleton with the closest distance as the matching result of the text information; if no match is found (i.e., there is no road skeleton in the circular search area), return to step (S3.4.2), expand the search radius and continue searching until the final result is matched or the search area radius exceeds the set threshold.

[0062] S3.5: Topological representation of road information: Segmented road information and intersection information have been obtained in the previous steps, wherein the segmented road information includes: the road skeleton pixel point set corresponding to the segmented road, the segmented road length information, and the segmented road name information; the intersection information includes the coordinate information of the intersection and the information of the segmented roads connected by the intersection. Based on the segmented road information and intersection information, a road topological relationship graph is established, that is, the segmented road is used as a node of the topological structure, and the node attributes include the segmented road length and the segmented road name. The node is connected to other roads through the edge of the topological structure, and the edge represents both the intersection information and the connection relationship between roads. In the constructed topological result graph, a general trunk road will have two intersections at the beginning and the end. Assuming that all intersections are crossroads, this topological node will have an association relationship with the other 6 topological nodes, indicating that the trunk road is connected to the 6 segmented roads.

[0063] In summary, the road topology representation method based on the electronic information map proposed in the present invention performs feature mining and topological representation on the electronic map. Specifically, the preprocessed electronic information map is input into the VGG-encoder-decoder model for semantic segmentation, and the road and intersection information (i.e., pixel-level road information) of the electronic information map is extracted; at the same time, the electronic information map is used through the EAST and CRNN models to realize text detection and text recognition, and finally the text information and pixel-level road information are integrated to construct a semantic topological map between roads.

[0064] Those skilled in the art can understand that the above are only preferred examples of the invention and are not intended to limit the invention. Although the invention is described in detail with reference to the above examples, those skilled in the art can still modify the technical solutions recorded in the above examples or replace some of the technical features therein with equivalents. Any modification, equivalent replacement, etc. made within the spirit and principle of the invention shall be included in the protection scope of the invention.

Claims

1. A road topology representation method based on an electronic information map, characterized in that: The following steps are involved: S1. Image acquisition and preprocessing: Use automatic acquisition software to capture the electronic information image of the electronic information map, and then preprocess the captured image to remove non-electronic map information and only retain the electronic map information; S2: extracting text information from the electronic information map obtained in S1, which is specifically achieved through the following sub-steps: (1) Use a text detection algorithm to perform text detection on the electronic information map and output the coordinates of the corner points of the detection box; (2) calculating the center position of the text according to the corner point coordinates, and capturing the image in the detection frame; (3) Use the text recognition model to perform text recognition on the captured image and convert it into text in the form of a string; (4) Filter the extracted text information and only retain the road name information; The electronic information map obtained by S1 is simultaneously subjected to pixel-level road information extraction, which is specifically achieved through the following sub-steps: (1) Enhance the image of the electronic information map by using grayscale processing and histogram equalization to improve image quality; (2) Applying the semantic segmentation model to perform semantic segmentation, separate road information and background information and label them separately to obtain pixel-level road information; S3: Fusing the pixel-level road information and the road name information to construct a topological map, which is specifically achieved through the following sub-steps: S3.1: Use the morphological skeletonization technique Zhang-Suen algorithm to extract the road skeleton from the pixel-level road information; S3.2: Define intersections in the road skeleton, detect all intersections through neighborhood checking, and record the coordinates of the intersections as intersection coordinates; S3.3: Obtaining segmented roads according to the intersection coordinates and the road skeleton, and calculating the connectivity relationship and length of each segmented road; S3.4: Matching text information with segmented road information: Convert the center position of the text to the same coordinate system as the segmented road information, perform a neighbor search for all subsequent road skeletons within the set search area, calculate the distance between the center position of the text and the candidate road skeleton, and take the candidate road skeleton with the closest distance as the matching result of the text information; S3.5: Topological representation of road information: The segmented roads are taken as topological nodes, and the topological nodes contain the segmented road length and road name information. The edges of the topological structure represent both the intersection information and the connection relationship between the segmented roads; the intersection information includes the intersection coordinates and the segmented road information connected by the intersection.

2. The road topology representation method based on electronic information map according to claim 1 is characterized in that: In S3.4, if there is no road skeleton in the circular search area, the radius of the search area is expanded to continue searching until a final result is matched or the radius of the search area exceeds a set threshold.

3. The road topology representation method based on electronic information map according to claim 1, characterized in that: The electronic map information described in S3.5 includes: roads, buildings, and area names.

4. The road topology representation method based on electronic information map according to claim 1, characterized in that: In S1, Multiscreenshots software is selected as the automatic acquisition software, and the electronic information map is selected as the SUMO map.

5. The road topology representation method based on electronic information map according to claim 1, characterized in that: In S2, the text detection algorithm uses the EAST algorithm.

6. The road topology representation method based on electronic information map according to claim 1, characterized in that: In S2, the text recognition model uses the CRNN deep learning model.

7. The road topology representation method based on electronic information map according to claim 1, characterized in that: In S2, the semantic segmentation model uses a VGG-encoder-decoder model, whose encoder is composed of a combination of five groups of convolutional layers and pooling layers. After each group of convolution and pooling operations, the number of channels of the feature map increases and the size decreases; the decoder is composed of a combination of four groups of upsampling layers and convolutional layers. After each group of upsampling and convolution operations, the number of channels of the feature map decreases and the size increases.

8. The road topology representation method based on electronic information map according to claim 1, characterized in that: In S3, the Zhang-Suen algorithm refines the road image by iteratively removing pixel edges, compressing the road into a line with a pixel width to obtain a road skeleton, and the line represents the center line of the road.

Images