An AI recognition-based damaged road network repair and reconstruction method

Abstract

The present application relates to geographic information system technical field, specifically to a kind of broken road network repair reconstruction method based on AI identification, comprising the following steps: S1, road surface data geometry repair and road width calculation;S2, target road network topology reconstruction;S3, road attribute matching;S4, topology simplification and output.The present application, by dynamic distance threshold and topology verification, intelligent connection broken road, eliminate the phenomenon of broken connection, construct topologically connected road network, provide reliable data basis for path planning.The width is calculated by associating road surface with center line, and the output road network has key information such as road width and grade by fusing external road network attributes, to meet the trafficability evaluation needs in mechanized construction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geographic information systems, in particular to a method for repairing and reconstructing a broken road network based on AI recognition. BACKGROUND

[0002] In the mechanized construction of power transmission lines, accurate and complete digital road networks are the key data basis for construction road planning. In recent years, through artificial intelligence technology to identify ground objects from remote sensing images has become an important means of quickly obtaining road network data. However, due to the generalization ability of AI models, the quality of remote sensing images, and the complexity of ground object shapes, the road network results generated by AI recognition generally have the following technical defects:

[0003] (1) Incomplete topological structure: the road network often exists in a broken form, i.e., the disconnected phenomenon, which cannot form a connected path, directly leading to the failure of path planning algorithms.

[0004] (2) Inaccurate geometric information: the generated road network is mostly a center line without width, lacking the key physical parameter of road width, which cannot meet the evaluation needs of vehicle passability in mechanized construction.

[0005] (3) Missing semantic information: the output results usually only contain geometric information, lacking key semantic attributes such as road grades, and the application value is limited.

[0006] Currently, the processing of the above defects mainly relies on a large amount of tedious manual repair and editing by professional GIS personnel. This method not only has high labor intensity, low efficiency, and high cost, but also when facing long-distance and large-scale engineering projects, it has become a bottleneck restricting the overall progress of the project. Therefore, there is an urgent need for a technical solution that can automatically and accurately process AI recognition results to construct topologically complete and attribute-rich engineering-level road network data. SUMMARY

[0007] The technical problem to be solved by the present application is to provide a method for repairing and reconstructing a broken road network based on AI recognition, which is used to improve road information and meet the evaluation needs of vehicle passability in mechanized construction.

[0008] The technical solution adopted by the present application to solve the technical problem is a method for repairing and reconstructing a broken road network based on AI recognition, comprising the following steps:

[0009] S1, road surface data geometry repair and road width calculation, obtain road image data, AI identify road surface data on road image data, geometric repair is carried out to each road surface, the candidate center line of the road surface is screened out through spatial index, the intersection length of each road surface and each candidate center line is calculated, each road surface is uniquely assigned to the center line with the longest intersection length, and the road width is calculated according to the area of the fused road surface and the length of the center line;

[0010] S2, target road network topology reconstruction, extract the center line end point and construct the near neighbor search structure, set a dynamic distance threshold sequence from small to large, at each threshold, iterative search is carried out and the road end point pairs meeting the distance condition are connected, the newly generated road connection line in each iteration is topologically verified, the invalid connection which does not meet the cross-line crossing rule or forms a small loop is removed, and the target road network is obtained;

[0011] S3, road attribute matching, obtain an external reference road network with road grade attribute, calculate the heading similarity by spatial buffer screening and vector angle calculation, match the roads in the target road network and the external reference road network, and migrate the road grade attribute to the target road network, and obtain the final road network with road width information and road grade attribute;

[0012] S4, topology simplification and output, topologically simplify the final road network, merge redundant nodes, and output as a standard vector geographic data file.

[0013] Further, the topology verification in step S2 includes a cross-line crossing rule, that is, the newly generated connection line cannot geometrically intersect with the non-end point part of any existing road center line in the road network.

[0014] Further, the topology verification in step S2 also includes a small loop rule, that is, after the newly generated connection line is added to the road network, it is detected whether a new closed loop is formed; if a closed loop is formed, the circumference of the loop is calculated; when the circumference is less than a preset threshold, the connection line is determined to be invalid and is removed.

[0015] Further, the process of matching the roads in the target road network and the external reference road network in step S3 includes the following sub-steps:

[0016] S31, spatial constraint screening, taking the road segment to be matched in the target road network as a reference, all reference road segments falling within the buffer range are screened out from the external reference road network by creating a spatial buffer, and an initial candidate set is formed;

[0017] S32, shape constraint selection, the direction vectors of the road segment to be matched and each reference road segment in the initial candidate set are calculated, and the heading similarity between the two is quantified by calculating the included angle between the direction vectors;

[0018] S43, attribute migration, extracting the road grade attribute from the reference road segment selected through the morphological constraint and assigning it to the to-be-matched road segment in the final road network.

[0019] Further, the direction vector is the unit vector of the line connecting the head and tail endpoints of the road segment, and the judgment condition of the alignment similarity is that the included angle is less than a preset angle threshold.

[0020] Further, in step S4, the topological simplification includes merging two-degree nodes to eliminate redundant nodes, and outputting standard geographic data format.

[0021] The beneficial effects of the present application are: through dynamic distance threshold and topological verification, the broken road is intelligently connected, the broken connection phenomenon is eliminated, and a topologically connected road network is constructed, providing reliable data basis for path planning. The width is calculated by associating the road surface with the center line, and the external road network attributes are fused, so that the output road network has key information such as road width and grade, meeting the trafficability evaluation needs in mechanized construction. BRIEF DESCRIPTION OF DRAWINGS

[0022] Fig. 1 is a flowchart of the present application;

[0023] Fig. 2 is a schematic diagram of the topological reconstruction of the target road network;

[0024] Fig. 3 is a schematic diagram of road attribute matching. DETAILED DESCRIPTION

[0025] The embodiments of the present application will be described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application.

[0026] As Figs. 1-3 shown, the present application is a broken road network repair and reconstruction method based on AI recognition, comprising the following steps:

[0027] S1, road surface data geometry repair and road width calculation:

[0028] First, high-resolution road remote sensing image data is obtained, a pre-trained deep learning model (such as U-Net or MaskR-CNN) is used for road surface recognition, and road surface vector data is output. Since the AI recognition result may have geometric defects or noise, this step first performs geometric repair on the road surface to eliminate holes, burrs and deformation in the recognition result, laying a geometric foundation for subsequent generation of high-quality center line and accurate width calculation.

[0029] Subsequently, based on the spatial index (such as R-tree), each road surface is quickly retrieved for candidate centerlines; the intersection length of the road surface and each candidate centerline is calculated, and the road surface is uniquely assigned to the optimal centerline by the maximum intersection length principle. The ambiguity problem that multiple centerlines may pass through the same road surface can be solved, ensuring that each road surface is uniquely and accurately associated with a most representative centerline.

[0030] Finally, the road width calculation uses the formula: road width = fused road surface area / centerline length. The road width, a key physical geometric attribute, is given to the original road network with only centerlines, upgrading it from single linear data to enhanced linear data with width attributes that can represent the passing space, directly meeting the demand for road width in subsequent construction vehicle passing analysis.

[0031] S2, target road network topology reconstruction:

[0032] Extract the endpoints of all centerlines and construct a KD tree for nearest neighbor search. Set a dynamic distance threshold sequence (such as 10m, 20m, …, 100m), and iterate the search for endpoint pairs at each threshold. If the distance between two points is less than the current threshold and they are not connected, a connection line is generated. The nearest neighbor breakpoints with the highest possibility are connected first to avoid generating a large number of false long-distance connections by using a large threshold at once, thereby ensuring connection efficiency while improving connection accuracy.

[0033] After each iteration, immediately perform topology verification on the newly generated connection lines:

[0034] Cross-line intersection rule verification: Check if the new connection line intersects with the non-endpoint part of the existing road centerline. If it does, it is removed. This strictly adheres to the topology rules of the road network in the plane, preventing the generation of unreasonable air fly lines or cross-type intersections, ensuring that the reconstructed road network is reasonable in geometry and topology.

[0035] Micro-loop rule verification: Detect if the new connection line forms a closed loop, calculate the loop circumference, and if the circumference is less than a preset threshold (such as 50m), determine it as an invalid connection and remove it. The micro-loop rule verification can eliminate small loops that have no actual passing significance due to AI recognition errors or incorrect connections inside the road or at the intersection, maintaining the simplicity and correctness of the road network structure.

[0036] By performing topology verification, the road disconnection problem caused by incomplete AI recognition is systematically repaired, and a target road network that is completely connected in topology and meets the real-world road connection rules is constructed.

[0037] S3, road attribute matching:

[0038] An external reference road network (such as OpenStreetMap) is acquired, which contains road grade attributes. For each road segment in the target road network, a matching operation is performed:

[0039] S31, spatial constraint screening: a buffer zone (radius adjustable, such as 20 m) is created centered on the road segment to be matched, and all reference road segments falling within the buffer zone are screened from the external reference road network to form an initial candidate set. Through spatial constraint screening, the proximity of spatial positions is utilized for rapid preliminary screening, greatly reducing the calculation amount of subsequent fine matching.

[0040] S32, morphological constraint selection: the direction vectors (unit vectors) of the road segment to be matched and the candidate road segment are calculated, and the vector angle is calculated; if the angle is less than an angle threshold (such as 30°), it is considered to be similar in direction. In this step, on the basis of spatial proximity, it is further confirmed whether the candidate road segment and the road segment to be matched are the same road from the geometric shape, effectively excluding parallel roads, viaducts and ground roads, and other spatially close but directionally different interference items, to ensure the accuracy of the matching.

[0041] S33, attribute migration: the road grade attribute is extracted from the reference road segment selected by the morphological constraint selection, and is assigned to the road segment to be matched in the final road network. This step solves the core defect of the AI-recognized road network lacking semantic attributes, migrates key level information such as "expressway", "national highway", "rural road" from the authoritative reference road network to the road network generated by this method, greatly improving the application value of the output data. Through the dual constraints of space and shape, high-precision road matching is achieved, thereby injecting rich semantic attributes into the geometrically and topologically perfect road network, forming an information-complete final road network.

[0042] S4, topological simplification and output:

[0043] The final road network with enhanced attributes is topologically simplified, which mainly merges two-degree nodes (such as nodes connecting only two edges). The redundant nodes generated in the data generation and connection process are eliminated, the network structure is simplified, the data storage amount is reduced, and the operation efficiency of subsequent path planning and other analysis based on the road network is improved.

[0044] Finally, the final road network is output in a standard geographic data format (such as GeoJSON, KML). It is ensured that the results generated by the present application can be seamlessly connected to existing GIS (Geographic Information System) platforms, construction planning software or other business systems, and have good interoperability and practicality.

[0045] The embodiments of the specific implementation are preferred embodiments of the present application, and do not limit the protection scope of the present application, so that: any equivalent changes made in structure, shape, principle according to the present application should be covered within the protection scope of the present application.

Claims

1. A method for repairing and reconstructing a damaged road network based on AI recognition, characterized in that, Includes the following steps: S1. Geometric repair of road surface data and calculation of road width: acquire road image data, AI recognizes road surface data on the road image data, performs geometric repair on each road surface, selects candidate center lines of the road surface through spatial index, calculates the intersection length between the road surface and each candidate center line, uniquely assigns each road surface to the center line with the longest intersection length, and calculates the road width based on the area of ​​the fused road surface and the length of the center line. S2. Target road network topology reconstruction: extract centerline endpoints and construct a nearest neighbor search structure. Set a dynamic distance threshold sequence from small to large. Under each threshold, perform iterative search and connect road endpoint pairs that meet the distance conditions. Perform topology verification on the newly generated road connection lines in each iteration and remove invalid connections that do not conform to the cross-line intersection rules or form small loops to obtain the target road network. S3. Road attribute matching: Obtain an external reference road network with road grade attributes. Through spatial buffer filtering and vector angle calculation of direction similarity, match the roads in the target road network and the external reference road network, and transfer the road grade attributes to the target road network to obtain the final road network with road width information and road grade attributes. S4. Topology Simplification and Output: Perform topology simplification on the final road network, merge redundant nodes, and output a standard vector geographic data file.

2. The method for repairing and reconstructing a damaged road network based on AI recognition according to claim 1, characterized in that, The topology verification in step S2 includes the cross-line intersection rule, which states that the newly generated connecting line must not geometrically intersect with the non-endpoint portion of any existing road centerline in the road network.

3. The method for repairing and reconstructing a damaged road network based on AI recognition according to claim 2, characterized in that, The topology verification in step S2 also includes a micro-loop rule, that is, after the newly generated connection line is added to the road network, it is detected whether a new closed loop is formed; if a closed loop is formed, the perimeter of the loop is calculated; when the perimeter is less than a preset threshold, the connection line is determined to be invalid and is removed.

4. The method for repairing and reconstructing a damaged road network based on AI recognition according to claim 1, characterized in that, The process of matching roads in the target road network and the external reference road network in step S3 includes the following sub-steps: S31. Spatial constraint screening: Based on the road segments to be matched in the target road network, a spatial buffer is created to screen all reference road segments that fall within the range of the external reference road network, forming an initial candidate set. S32. Shape constraint selection: Calculate the direction vector of the road segment to be matched and each reference road segment in the initial candidate set, and quantify the similarity of the two directions by calculating the angle between the direction vectors. S43. Attribute transfer: Extract the road grade attribute from the reference road segments selected through morphological constraints, and assign it to the road segments to be matched in the final road network.

5. The method for repairing and reconstructing a damaged road network based on AI recognition according to claim 4, characterized in that, In step S32, the direction vector is the unit vector of the line connecting the beginning and end endpoints of the road segment, and the condition for judging the similarity of the directions is that the included angle is less than a preset angle threshold.

6. The method for repairing and reconstructing a damaged road network based on AI recognition according to claim 1, characterized in that, In step S4, topology simplification includes merging second-degree nodes to eliminate redundant nodes and outputting standard geographic data format.

Images