A static lightweight high-precision map making method based on a highway marking design file

Abstract

This invention discloses a static, lightweight, high-precision map production method based on highway marking design files. First, the centerline and two-dimensional marking layers are extracted. Then, based on road network topology, reference line direction change points, and the lateral position of the markings, the two-dimensional markings are hierarchically divided into road-level, lane-segment-level, and lane-level layers. Next, by constructing a three-dimensional centerline point set and utilizing nearest-point projection, combined with cross-section normal vectors and road cross slope, the two-dimensional marking points are serialized into three-dimensional coordinates. Finally, data storage units are constructed using the marking boundaries, lane objects are directly defined, and a standardized LD Map file conforming to the OpenDRIVE specification is compiled and generated. This invention improves the efficiency of high-precision map production, reduces the production cost and threshold, and can be used for the industrialization of autonomous driving maps.

Description

Technical Field

[0001] This invention relates to the fields of highway design and high-precision mapping, specifically to a method for creating static, lightweight, high-precision maps based on highway marking design files. Background Technology

[0002] Autonomous driving using vehicle-to-everything (V2X) technology relies heavily on high-definition maps (HD maps), which can achieve centimeter-level accuracy. However, the high production cost and slow update speed of HD maps currently hinder the widespread adoption of autonomous driving. Standard-defined maps, on the other hand, only provide meter-level accuracy and are only suitable for general navigation.

[0003] To accelerate the deployment of autonomous driving, map providers have proposed a lightweight, high-precision map, known as LD Map, which falls between HD Map and HD Map. LD Map removes non-driving elements such as streetlights and road signs, focusing only on essential map elements like lane lines and stop lines, achieving decimeter-level accuracy at a significantly lower cost than HD Map.

[0004] The core technology of traditional high-precision map production mainly lies in data collection. Currently, there are two data collection methods: professional collection, which uses specialized collection vehicles and equipment to collect road data multiple times to ensure accuracy; and crowdsourced collection, which involves users driving vehicles to collect road data and transmit it to the cloud.

[0005] On the compilation and production side, post-processing software (such as Inertial Explorer) is first needed to parse and solve the massive amount of data. This step requires a lot of computing power. Then, road elements need to be extracted from the parsed and solved data. This step usually uses manual or machine learning methods, which also has high computing power and labor costs.

[0006] Regardless of the method used to create high-precision maps, all have drawbacks such as high production and updating costs and low efficiency. This is because the original data contains abundant three-dimensional spatial data, making it extremely difficult to extract road data from this overly redundant data.

[0007] Therefore, research has emerged on high-precision map creation methods based on road design data. This method utilizes the Python library ElementTree (a module for processing XML data) to programmatically construct road, lane, lane marking, and safety facility layers from Opendrive data (a standardized XML file format for describing road networks). This method requires users to define explicit mapping rules for attributes and labels. Since the structure and topology of different roads vary, rewriting the code is necessary for creating each unique road. There is still room for improvement in terms of efficiency and complexity. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention proposes a static, lightweight, high-precision map production method based on highway marking design files. The core improvements cover the entire process from data acquisition to compilation and production, aiming to optimize the high-precision map production steps and reduce production costs.

[0009] Among them, highway marking design is mostly completed through various non-standardized custom development software based on CAD. The output results are mostly two-dimensional polyline objects or custom non-standard objects. LD Map, as a high-precision map, is a standardized object containing three-dimensional information. Therefore, the conversion of non-standard design objects into standardized LD Map format objects has become a key prerequisite for realizing LD Map production.

[0010] To solve this core conversion problem, this invention develops a dedicated tool in two steps: First, a processing tool based on highway design data is developed, whose core function is to convert two-dimensional marking objects into three-dimensional polyline objects, and integrates the function of further converting three-dimensional polylines into a serialized intermediate data format; Second, a conversion tool from this intermediate data format to LDMap is developed. Through the collaboration of these two tools, the complete conversion from highway marking design files to LDMap is finally achieved, completing the LDMap production.

[0011] To achieve the above objectives, this invention provides a method for creating static, lightweight, high-precision maps based on highway marking design files, comprising the following steps: S1: Obtain highway road design documents, extract the road design centerline and two-dimensional marking layers, and based on the road network topology connectivity, the topology change points in the direction of the road reference line, and the lateral position order of the two-dimensional marking objects, divide the two-dimensional marking objects into road-level, lane-segment-level, and lane-level hierarchical divisions respectively to obtain a set of classified and organized two-dimensional marking objects. S2: Obtain the plane coordinates and corresponding elevation data of the road design centerline, construct a three-dimensional point set of the road design centerline, and determine the projection point Q (Xq, Yq, Zq) corresponding to the mark point P (Xp, Yp) to be processed in the three-dimensional point set using the nearest point projection method. Calculate the normal vector of the road cross section at projection point Q. , Based on the difference in planar coordinates between the marking point P and the projection point Q, the lateral offset distance d of the marking point P relative to the centerline is calculated. Combined with the road cross slope i, the elevation of the marking point P is then calculated. Obtain the three-dimensional coordinates of the marking points: ; That is, three-dimensional datum data; S3: Serialize each set of two-dimensional marking objects into three-dimensional coordinates of marking points according to step S2, construct independent data storage units according to the marking boundaries, and the names are directly related to the hierarchical division in step S1; S4: Parse the data storage unit, use the lane marking boundary as the lane left and right boundary definition function to directly construct the lane object, and compile to generate a standardized LD Map file that conforms to the Opendrive specification.

[0012] Further, step S1, before extracting the road design centerline and two-dimensional marking layers, includes: The system performs structured analysis on various CAD design files, identifies and extracts graphic objects belonging to the marking category, and classifies and filters them according to layer name, object type, color code or line type attribute. It filters out auxiliary annotations, dimension lines and drawing frames that are not related to markings, and generates a clean two-dimensional marking layer.

[0013] Furthermore, obtaining the plane coordinates and corresponding elevation data of the road design centerline includes: Obtain the plane coordinate data and corresponding elevation data of the station number on the road centerline, perform interpolation processing on the centerline, and generate a centerline point set containing multiple continuous three-dimensional discrete points.

[0014] Furthermore, the steps for dividing the two-dimensional marking object into road-level, lane-segment-level, and lane-level hierarchical classifications are as follows: Road object division: Based on the topological connectivity of the road network and the continuity of the road design centerline, the entire road network is divided into several independent road objects. Among them, the main line and each ramp constitute independent road objects. The bifurcation and merging positions between road objects are marked by intersection objects. Lane segment object division: Along the direction of the road reference line, identify road topology change points, and divide the area between adjacent change points in the direction of the road reference line into several lane segment objects. The marking segments in each lane segment object are assigned to the corresponding lane segment set. The topology change points include: the appearance of variable speed lanes, the intervention of auxiliary lanes, the entry or exit of ramps, and lane merging or forking.

[0015] Furthermore, in step S4, the specific steps for directly constructing the lane object using the lane marking boundaries as the lane left and right boundary definition functions are as follows: Data parsing and index construction: Based on the name encoding format of the data storage unit, the road sign, lane segment sign, and marking sign corresponding to each data storage unit are parsed out to form a structured data set with a three-level index of road-lane segment-marking. Boundary pair identification: Based on the identification order in the structured data set, identify two spatially adjacent data storage units and define them as the left boundary marking and right boundary marking of the lane, respectively; Lane geometry generation: Using a curve fitting algorithm, the coordinates of the left and right boundary lines of the lanes are converted into mathematical boundary definition functions. Based on the mathematical boundary definition functions, a lane geometry model is generated, and the topological connection relationship between lanes is established to form a lane object.

[0016] Furthermore, at the road level, a corresponding road object is created for each road (main line and each ramp) in the road network and assigned a unique identifier. The reference line geometry of the road object is automatically fitted and generated by the map building tool based on the input lane boundary point sequence, and described in segments of straight lines and arcs. At the lane segment level, a corresponding lane segment object is created on each road object according to the points where the number of lanes changes. The lane segment objects are connected end to end in the road direction, covering the entire length of the road. At the lane level, each lane is constructed using the left and right boundary definition method. For N+1 three-dimensional markings arranged in a horizontal order within the same lane segment object, two adjacent markings constitute the left and right boundaries of a lane: the k-th marking serves as the left boundary of the k-th lane, and the (k+1)-th marking serves as the right boundary of the k-th lane. In this way, all N lanes within the lane segment are constructed one by one.

[0017] Furthermore, establishing the topological connection relationship between lanes includes: constructing a longitudinal topological connection between adjacent lane segments within the same road object, mapping lane identifiers based on the sequential relationship of lane segments to achieve the continuity of lanes in the road extension direction; and constructing a lateral topological connection between different road objects via intersection objects, based on the road connection relationship defined in the intersection objects to achieve connectivity at the road network level.

[0018] Furthermore, the compilation and generation of standardized LD Map files in step S4 specifically includes: mapping and encoding the constructed road objects, lane segment objects, lane objects, and intersection objects according to the data architecture specified by the Opendrive standard; serializing the encoded data into an XML format file with the extension .xodr to generate the final high-precision map data file.

[0019] The resulting .xodr file is the final LD ​​Map product delivered by this invention, which can be directly read and used by mainstream autonomous driving simulation platforms, perception algorithm testing systems, and map publishing processes.

[0020] This invention also provides a system for creating static, lightweight, high-precision maps based on highway marking design files, comprising: Geometric Spatialization Module: Used to execute steps S1~S2, using the road design centerline as the elevation calculation benchmark, to convert the two-dimensional marking layer data into three-dimensional marking data with three-dimensional spatial coordinates; Data serialization module: used to execute step S3, serialize each two-dimensional datum object set into the three-dimensional coordinates of datum points, construct independent data storage units according to the datum boundaries, and the names are directly related to the hierarchical division in step S1; Map compilation module: used to execute step S4, parse the data storage unit, directly construct lane objects using lane marking boundaries as lane left and right boundary definition functions, and compile to generate a standardized LDMap file conforming to the Opendrive specification.

[0021] In the method provided by this invention, road marking data serves as both the original output of the design and the direct geometric source of high-precision maps. There is no intermediate manual 3D modeling step between the two, achieving the shortest path conversion from design data to map data. Compared with existing high-precision map production methods that rely on field laser scanning or manual 3D modeling, this method significantly shortens the production cycle and reduces production costs, demonstrating outstanding engineering practical value. Furthermore, it breaks through the bottlenecks of traditional high-precision map production, delivering standardized products in a low-cost, high-efficiency, and high-precision manner. This provides strong technical support for the industrialization of autonomous driving maps and has significant economic value and promising prospects for promotion.

[0022] This invention achieves fully automated conversion from highway marking CAD design files to Opendrive format high-precision maps (LD Maps). Compared with existing technologies, the advantages of this invention are as follows: (1) Efficiency improvement: Traditional methods rely on field laser scanning or manual 3D modeling, which takes several months; this invention directly reuses existing Excel coordinate data in the design phase, and the three steps (3D modeling, structuring, and standardized compilation) are seamlessly connected, reducing the modeling time of a single interchange to several hours, and improving overall efficiency by 5-10 times.

[0023] (2) Cost reduction: No additional on-site data collection equipment and manpower are required, saving high costs such as lidar and GNSS post-processing, reducing production costs by more than 70%, making it particularly suitable for rapid iteration of large-scale highway networks.

[0024] (3) Accuracy and quality assurance: The parametric interval labeling mechanism is adopted to accurately assign RoadMark semantics (including color, line type, line width, and sOffset interval) such as white solid line, yellow solid line, and white dashed line to the lane boundaries, strictly complying with the ASAMOpendrive 1.4~1.7 specifications, avoiding manual labeling errors, and ensuring that the map geometric accuracy and topological connectivity are 100% consistent.

[0025] (4) Standardization and compatibility: The output .xodr file can be directly read by mainstream autonomous driving simulation platforms and perception algorithm testing systems such as CARLA, SUMO, IPG CarMaker, Apollo, MATLAB / Simulink, etc. It supports complex intersection merging and diverging topologies, greatly expanding application scenarios.

[0026] (5) Engineering practicality: It supports batch processing of multiple files, automatic identification and deletion of edge lanes, and batch connection of front and rear drive lanes. It is suitable for any complex interchange and main line scenario and has strong maintainability and scalability. Attached Figure Description

[0027] Figure 1 This is a geometric representation of the normal vector of a road cross-section; Figure 2 Geometric diagram for calculating the lateral offset and elevation of a cross section; Figure 3 Example of an intermediate data file (A1 coordinate data); Figure 4 This is a schematic diagram of the intersection topology. Figure 5 Example of configuring lane connection relationships; Figure 6 This is a diagram illustrating the input of 3D calibration lines and the internal fitting of the tool. Figure 7 This represents the hierarchical structure of XML nodes in OpenDrive files. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0029] This invention assigns lane boundary objects semantic attributes based on their marking types. Supported marking types include, but are not limited to, white solid lines, yellow solid lines, and white dashed lines. The assignment method employs a parameterized interval labeling mechanism, applying marking attributes to the entire length or a specified mileage interval of the lane boundary object. These attributes include descriptive parameters such as marking color, line type (solid or dashed), and line width. In Opendrive, these marking semantic attributes are attached to the corresponding Lane object as RoadMark objects, described through fields such as type (line type), color (color), and width (marking width). The three-dimensional coordinates recorded in the intermediate data file represent the actual physical geometry of the road markings. This geometry naturally corresponds to the spatial position of the road markings. The semantic assignment further supplements this with color, line type, and other attribute information, unifying the physical geometry and semantic description of the boundary object. RoadMark is an object in Opendrive used to describe road markings, specifically the visual features of lane edges—the lines drawn on the road surface. Lane is the core object in Opendrive describing lanes, representing a specific driving lane on the road.

[0030] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. Example 1

[0031] This embodiment provides a method for creating a static, lightweight, high-precision map based on highway marking design files. The specific steps are as follows: Step S1: Obtain the highway road design documents and extract the road design centerline and two-dimensional marking layers.

[0032] Since CAD design drawings are typically two-dimensional planar data (X, Y) lacking precise elevation information (Z), the information extraction process involves non-standard, unstructured data. Therefore, information filtering, mapping, and transformation must be performed simultaneously. The specific steps are as follows: (1) Extraction and cleaning of gradation elements: The CAD-based design file is structured and parsed to identify and extract graphic objects belonging to the gradation category. The objects are classified and filtered according to layer name, object type, color code or line type attribute. Auxiliary annotations, dimension lines and drawing frames unrelated to gradation are filtered out to generate a clean set of two-dimensional gradation objects. (2) Complete the hierarchical division of road objects, lane segments and lanes, and assign each marking object to the corresponding road set, lane segment set and lane set to obtain a set of two-dimensional marking objects after classification and sorting; The road set division involves dividing the entire road network into several independent road objects based on the network's topological connectivity. Continuous lane boundary markings within each road object are assigned to a corresponding main road set. Each road object corresponds to a set, and each main road is labeled with a different letter, such as A, B, C, and D. The bifurcations or merging points of different road objects are marked using intersection objects. Markings within the intersection area are assigned to the corresponding inbound or outbound road objects based on their connectivity. Lane segment set division: Along the road reference line direction, identify road topology change points, and divide the area between adjacent change points in the road reference line direction into several lane segment objects. The marking segments in each lane segment object are assigned to the corresponding lane segment set. The topology change points include: the appearance of variable speed lanes, the intervention of auxiliary lanes, the entry or exit of ramps, and lane merging or branching. Taking main road A as an example: starting from the road's starting point until the first topological change point (such as the starting point of the auxiliary lane before the exit ramp), the markings within this interval are assigned to lane segment A1; when an auxiliary lane appears (such as the speed change lane before the ramp exit), the number of lanes increases, and starting from the starting point of the auxiliary lane until the bifurcation nose (the geometric separation point between the main road and the ramp), the markings within this interval are assigned to lane segment A2; after passing the bifurcation nose, the auxiliary lane ends, the number of main road lanes is restored, and the markings after this point are assigned to lane segment A3; for independent road B (ramp), it is also divided into lane segments such as B1 and B2 according to its own widening, narrowing, or merging.

[0033] Lane boundary and object construction: Within each lane segment, the road surface markings are numbered sequentially according to their lateral positions and used as lane boundary lines to be assigned to the corresponding lane set; lane objects are generated based on the enclosed space of two adjacent lane boundary lines, where N lane objects are defined by N+1 boundary lines, and the boundary line identifier is generated by combining the lane segment identifier and the boundary number.

[0034] The lane markings within the A1 lane section are numbered A11, A12, A13, A14, and A15 (5 lanes correspond to 5 boundary lines), and the lane markings within the A2 lane section are numbered A21, A22, A23, A24, A25, and A26 (6 lanes correspond to 6 boundary lines).

[0035] After all roads in the entire road network are divided, each road marking object is assigned to the corresponding lane segment and lane, resulting in a set of classified and organized two-dimensional road marking objects. This forms a complete three-level hierarchical structure of roads, lane segments, and lanes, allowing the road assignment, lane segment assignment, and lateral position of each marking to be read directly from the number. This facilitates the subsequent use of two adjacent markings as the left and right boundaries of a lane.

[0036] Step S2: Using the road design centerline as the elevation calculation benchmark and combining the longitudinal profile design parameters, the two-dimensional marking is converted into a three-dimensional marking with three-dimensional spatial coordinates, thus completing the geometric spatialization.

[0037] Three-dimensional coordinate precise mapping: Obtain the plane coordinate data and corresponding elevation data of the station number on the road centerline, perform interpolation processing on the centerline, and generate a centerline point set containing multiple continuous three-dimensional discrete points; For the marking point P to be calculated, the projection point Q corresponding to the marking point P is determined from the set of centerline points using the nearest point projection method; Calculate the normal vector of the road cross section at projection point Q, which is the unit direction vector in the horizontal plane that is perpendicular to the road's forward direction and points to the right side of the road. Based on the difference in planar coordinates between the marking point P and the projection point Q, calculate the lateral offset distance d of the marking point relative to the centerline; Obtain the elevation value Zq of the projection point Q, and calculate the elevation Zp of the marking point P based on the preset road cross slope i and the lateral offset distance d. Output the three-dimensional coordinates (Xp, Yp, Zp) of the marking point to obtain the three-dimensional spatial position of the road marking.

[0038] The spatial data of the road design centerline serves as a reference for elevation calculation. It is a three-dimensional alignment with known elevations provided by the highway route specialist. Its actual cross-sectional position is not necessarily located at the geometric center of the road cross-section. Depending on the design scheme, it may be located at the center of the road cross-section, the left edge of the hard shoulder, the right edge of the hard shoulder, or other designated locations. Its function is limited to calculating the Z coordinates of each two-dimensional marking node and does not participate in the subsequent construction of the LD Map.

[0039] The elevation data of the centerline station is retrieved from the road longitudinal profile design data, which comes from the longitudinal profile design results file, including one or more of the following: longitudinal profile design table, centerline three-dimensional coordinate results table, or centerline elevation data exported from design software.

[0040] like Figure 1 As shown, for a gradation node P (with plane coordinates Xp and Yp), the nearest point projection method is used to determine its corresponding projection point Q (with coordinates Xq, Yq, and Zq) in the three-dimensional discrete point sequence of the centerline. That is, in the set of line segments formed by each adjacent node of the centerline, the line segment with the closest plane distance to point P is searched, and point P is projected onto the line segment. The resulting projection point is Q.

[0041] Let n be the normal vector of the road cross section at projection point Q, and let its components in the X and Y directions be denoted as . and , , L is the vector length. . Let be the slope of the tangent line to the centerline at the projection point Q.

[0042] The lateral offset d of point P relative to the centerline is defined as the projection of the difference vector between the plane coordinates of point P and the plane coordinates of point Q onto the horizontal component of the normal vector n, and the calculation formula is as follows: When d is positive, point P is located to the right of the centerline; when d is negative, point P is located to the left of the centerline. Considering that roads in the physical world have cross slopes for drainage, the cross slope ratio i is usually taken as 2% (i.e., i = 0.02). Figure 2 As shown, the slope decreases from the road centerline towards both sides. Based on this, the three-dimensional elevation of the road marking node P is... It can be deduced from the following formula: in, Let |d| be the elevation of the projection point Q, |d| be the absolute value of the lateral offset, and i be the cross slope ratio. The above formula takes a minus sign to reflect the physical fact that the road surface decreases from the centerline to both sides. When the road is in a curve superelevation section, the cross slope ratio i should be replaced according to the actual superelevation design value. For other road sections, i = 0.02 is uniformly used. Since the lateral offset calculation takes the actual location of the centerline as the zero point, the form of the above formula remains unchanged regardless of the location of the centerline in the road cross section, and it has universality. After completing the above calculation for all nodes on the two-dimensional marking, the complete three-dimensional coordinates (Xp, Yp, Zp) of each node are determined, thereby converting the two-dimensional marking object into a three-dimensional polyline object.

[0043] In this step, the two-dimensional markings are assigned three-dimensional coordinates that conform to the design intent, which solves the problem of map elevation errors in scenarios such as suspended bridges or deep road cuts.

[0044] S3: Data Serialization: After serializing the coordinate and attribute information of the 3D markings, the output is a structured intermediate data file. This file stores not only coordinate points (x, y, z) but also semantic information, such as marking type (solid line, dashed line), color, width, etc. This step serves as an intermediate state, allowing automated quality checks (such as checking for marking breakpoints and overlaps) before generating the final map, without modifying the core conversion logic.

[0045] The structured intermediate data file adopts a tabular data storage format (preferably an Excel workbook .xlsx in this embodiment, but CSV or database tables can also be used), and each three-dimensional marking line (i.e., a lane boundary line) corresponds to an independent data storage unit (such as an independent worksheet) in the file.

[0046] The names of data storage units strictly follow the encoding format of "road sign - lane segment sign - lane marking sign". For example, AA1A11 represents the first lane marking boundary within lane segment A1 of road A, AA2A26 represents the outermost lane marking boundary within lane segment A2 of road A, and BB1B11 represents the first lane marking boundary within lane segment B1 of road B (ramp). This naming rule directly associates the data unit name with the LD Map hierarchy. Each data storage unit contains column field headers and coordinate sequence data.

[0047] like Figure 3 As shown, the table headers are in the following order: serial number, X coordinate, Y coordinate, and Z coordinate. The coordinate sequence data is arranged row by row according to the order of the nodes in the direction of the lane marking. Based on the identifier order in the naming rules, the lane marking data of two adjacent data storage units logically constitute a pair of left and right boundaries of a lane, facilitating subsequent program reading by pair.

[0048] S4: Map Compilation: Parse the structured intermediate data file, using the 3D lane marking data as the sole input source for constructing the lane geometry model. Directly construct lane objects through left and right boundary definition functions, and compile to generate a standardized LD Map file conforming to the OpenDrive specification, as detailed below: (1) Based on the name encoding format of the data storage unit, the road sign, lane segment sign and marking sign corresponding to each data storage unit are parsed out to form a structured data set indexed by the three levels of road-lane segment-marking; At the road level, based on the division results of step S1, the map building tool automatically fits and generates reference lines for road objects according to the input lane boundary point sequence, describing them in segments of straight lines and arcs, without the need to input road centerline data separately; at the lane segment level, corresponding lane segment objects are created on each road object according to the lane number variation segment, and each lane segment object is connected end to end in the road direction, together covering the entire length of the road; at the lane level, each lane is constructed using the left and right boundary definition method. For N+1 three-dimensional markings arranged in a horizontal order within the same lane segment object, two adjacent markings constitute the left and right boundaries of a lane: the k-th marking serves as the left boundary of the k-th lane, and the (k+1)-th marking serves as the right boundary of the k-th lane, and so on, constructing all N lanes within the lane segment; (2) Based on the identification order in the structured data set, identify two spatially adjacent data storage units and define them as lane left boundary markings and lane right boundary markings respectively. Convert the coordinate points of the lane left boundary markings and lane right boundary markings into mathematical boundary definition functions. Generate lane geometric models based on the mathematical boundary definition functions and establish topological connection relationships between lanes to form lane objects.

[0049] In the data analysis process, the engineering coordinates are uniformly translated from the origin, converting the absolute coordinates in the engineering geodetic coordinate system into relative local coordinates with the map reference point as the origin, in order to meet the requirements of the map format for coordinate magnitude, while retaining the coordinate precision of no less than three decimal places. The translation amount is determined by the reference point coordinates of the engineering project, and is uniformly applied to all marking data to ensure that the relative positional relationship between each marking remains unchanged after the transformation.

[0050] like Figure 4 and 5 As shown, for lane connections between adjacent lane segments within the same road object, taking the connection between lane segments A1 and A2 as an example: A1 includes lanes A11, A12, A13, and A14, and A2 includes lanes A21, A22, A23, A24, and A25, where A25 is a newly added auxiliary lane. The preceding and succeeding relationships of each lane are configured according to their corresponding lateral positions. The preceding lane of A21 is A11, the preceding lane of A22 is A12, the preceding lane of A23 is A13, and the preceding lanes of A24 and A25 are both A14. This reflects the physical reality that when an auxiliary lane is added to the main line, the original outermost lane simultaneously connects to two subsequent lanes. This connection relationship is recorded in Opendrive in the form of references to the preceding and succeeding segments of the lane object.

[0051] For connections between different road objects, at the nose of the physical bifurcation point where the mainline and the ramp diverge, mainline road A and ramp road B are connected through an intersection object. The intersection object contains several connection records, and each connection record specifies the incoming road, the connecting road, and the specific lane pair.

[0052] Taking the connection between the end of segment A2 and segments A3 and B1 as an example: A25 serves as an auxiliary lane, and its successor lane is lane B11 in lane segment B1 of ramp road B. The successor lanes from A21 to A24 are A31 to A34, respectively. This connection relationship is recorded through the mapping between connection entries and lane pairs in the intersection object, realizing a complete topological representation of the mainline and ramps at the bifurcation node.

[0053] (3) After receiving the above boundary point sequence, the map building tool automatically fits it into a reference line geometry and lane width polynomial, generating an internal expression that conforms to the Opendrive specification. This fitting process is completed internally by the tool, as detailed below. Figure 6 As shown.

[0054] Specifically, the compilation and generation of standardized LD Map files includes: mapping and encoding the constructed road objects, lane segment objects, lane objects, and intersection objects according to the data architecture specified by the Opendrive standard; serializing the encoded data into an XML format file with the extension .xodr to generate the final high-precision map data file.

[0055] It should be noted that: Figure 7 As shown, during the XML format file export process, the sequence of 3D line boundary points input by this invention is automatically fitted into the reference line (reference line, expressed as a segment of straight line segments and arcs) and lane width polynomial (described as a cubic polynomial describing the lateral offset of the lane relative to the reference line) required by Opendrive. The header node (recording map metadata and reference point coordinates), road node set (each road object corresponds to a road node, which contains plan view, lanes and lane section sub-nodes, each lane contains line width polynomial and RoadMark line attributes), and intersection node set (recording the topological connections of all branching and merging points) are output sequentially. The above reference line fitting and width calculation are all completed automatically by the tool and are not within the processing scope of this invention.

[0056] After exporting, the resulting .xodr file is the final LD ​​Map product delivered by this invention, which can be directly read and used by mainstream autonomous driving simulation platforms, perception algorithm testing systems, and map publishing processes.

[0057] The processing boundary of this invention, which takes the three-dimensional marking point sequence as input, ends here. The entire lane object construction process uses the three-dimensional marking coordinate data as the sole geometric source, without introducing any other geometric input, and without needing to manually construct reference lines or calculate width parameters.

[0058] A system for creating static, lightweight, high-precision maps based on highway marking design files, mainly comprising: (1) Geometric spatialization module, configured to acquire highway road design documents, extract road design centerline and two-dimensional marking layer data, use the road design centerline as the elevation calculation benchmark, and combine the road longitudinal section design parameters to convert the two-dimensional marking layer data into three-dimensional marking data with three-dimensional spatial coordinates; (2) Data serialization module, configured to serialize the coordinate information and attribute information of the three-dimensional marking data to generate a structured intermediate data file; (3) Map compilation module, configured to parse the structured intermediate data file, using the three-dimensional line data as the sole input source for constructing lane geometry, directly constructing lane objects through left and right boundary definition functions, and compiling to generate a standardized LD Map file conforming to the Opendrive specification.

[0059] This invention drives the topology creation process by reading a pre-compiled lane connection configuration table. The configuration table is stored in a structured table format, with each row recording a unique identifier for a lane and its corresponding list of preceding and succeeding lane identifiers. It supports one-to-many branching and merging relationships, with multiple identifiers separated by delimiters. The algorithm sequentially reads each row of the configuration table, locates the target lane in the constructed LD Map object set based on the lane identifier, clears the existing preceding and succeeding references for that lane, and then reconstructs the reference set based on the identifier list in the configuration table, completing the topology binding. By using a configuration table-driven approach, the lane connection relationships in complex interchange areas are decoupled from the geometric construction logic, allowing the topology configuration to be maintained and modified independently without re-executing the geometric construction process. This significantly improves the flexibility and maintainability of creating large-scale complex road network maps.

[0060] Although the above embodiments have described the present invention and its implementation in detail, it should be noted that for those skilled in the art, any changes, modifications, substitutions, combinations, simplifications, etc., made to the corresponding conditions without departing from the technical principles of the present invention should be considered as equivalent substitutions, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A method for creating static, lightweight, high-precision maps based on highway marking design files, characterized in that, Includes the following steps: S1: Obtain highway road design documents, extract the road design centerline and two-dimensional marking layers, and based on the road network topology connectivity, the topology change points in the direction of the road reference line, and the lateral position order of the two-dimensional marking objects, divide the two-dimensional marking objects into road-level, lane-segment-level, and lane-level hierarchical divisions respectively to obtain a set of classified and organized two-dimensional marking objects. S2: Obtain the plane coordinates and corresponding elevation data of the road design centerline, construct a three-dimensional point set of the road design centerline, and determine the projection point Q (Xq, Yq, Zq) corresponding to the mark point P (Xp, Yp) to be processed in the three-dimensional point set using the nearest point projection method. Calculate the normal vector of the road cross section at projection point Q. , Based on the difference in planar coordinates between the marking point P and the projection point Q, the lateral offset distance d of the marking point P relative to the centerline is calculated. Combined with the road cross slope i, the elevation of the marking point P is then calculated. Obtain the three-dimensional coordinates of the marking points: ; That is, three-dimensional datum data; S3: Serialize each set of two-dimensional marking objects into three-dimensional coordinates of marking points according to step S2, construct independent data storage units according to the marking boundaries, and the names are directly related to the hierarchical division in step S1; S4: Parse the data storage unit, use the lane marking boundary as the lane left and right boundary definition function to directly construct the lane object, and compile to generate a standardized LD Map file that conforms to the Opendrive specification.

2. The static lightweight high-precision map production method according to claim 1, characterized in that, Step S1, before extracting the road design centerline and 2D marking layers, includes: The system performs structured analysis on various CAD design files, identifies and extracts graphic objects belonging to the marking category, and classifies and filters them according to layer name, object type, color code or line type attribute. It filters out auxiliary annotations, dimension lines and drawing frames that are not related to markings, and generates a two-dimensional marking layer.

3. The static lightweight high-precision map production method according to claim 1, characterized in that, Obtaining the plane coordinates and corresponding elevation data of the road design centerline includes: Obtain the plane coordinate data and corresponding elevation data of the station number on the road centerline, perform interpolation processing on the centerline, and generate a centerline point set containing multiple continuous three-dimensional discrete points.

4. The static lightweight high-precision map production method according to claim 1, characterized in that, The steps for dividing the two-dimensional marking objects into road-level, lane-segment-level, and lane-level hierarchical classifications are as follows: Road object division: Based on the topological connectivity of the road network and the continuity of the road design centerline, the entire road network is divided into several independent road objects. Among them, the main line and each ramp constitute independent road objects. The bifurcation and merging positions between road objects are marked by intersection objects. Lane segment object division: Along the road reference line direction, identify road topology change points, and divide the area between adjacent change points in the road reference line direction into several lane segment objects. The marking segments in each lane segment object are assigned to the corresponding lane segment set. The topology change points include: the appearance of variable speed lanes, the intervention of auxiliary lanes, the entry or exit of ramps, and lane merging or branching. Lane boundary and object construction: Within each lane segment, the three-dimensional marking data is mapped to lane boundary lines in lateral position order; lane objects are generated based on the enclosed space of two adjacent lane boundary lines, where N lane objects are defined by N+1 boundary lines, and the boundary line identifier is generated by combining the lane segment identifier and the boundary number.

5. The static lightweight high-precision map production method according to claim 1, characterized in that, In step S4, the specific steps for directly constructing the lane object using the lane marking boundaries as the lane left and right boundary definition functions are as follows: Data parsing and index construction: Based on the name encoding format of the data storage unit, the road sign, lane segment sign, and marking sign corresponding to each data storage unit are parsed out to form a structured data set with a three-level index of road-lane segment-marking. Boundary pair identification: Based on the identification order in the structured data set, identify two spatially adjacent data storage units and define them as the left boundary marking and right boundary marking of the lane, respectively; Lane geometry generation: Using a curve fitting algorithm, the coordinates of the left and right boundary lines of the lanes are converted into mathematical boundary definition functions. Based on the mathematical boundary definition functions, a lane geometry model is generated, and the topological connection relationship between lanes is established to form a lane object.

6. The static lightweight high-precision map production method according to claim 5, characterized in that, The establishment of topological connections between lanes includes: constructing longitudinal topological connections between adjacent lane segments within the same road object, mapping lane identifiers based on the sequential relationship of lane segments to achieve continuity of lanes in the road extension direction; and constructing lateral topological connections between different road objects via intersection objects, based on the road connection relationships defined in the intersection objects to achieve connectivity at the road network level.

7. The static lightweight high-precision map production method according to claim 1, characterized in that, The compilation and generation of standardized LD Map files in step S4 specifically includes: mapping and encoding the fields of the constructed road objects, lane segment objects, lane objects, and intersection objects according to the data architecture specified by the Opendrive standard; serializing the encoded data into an XML format file with the extension .xodr to generate the final high-precision map data file.

8. A system for creating static, lightweight, high-precision maps based on highway marking design files, characterized in that, include: Geometric Spatialization Module: Used to perform steps S1~S2 as described in claim 1, using the road design centerline as the elevation calculation benchmark, to convert the two-dimensional marking layer data into three-dimensional marking data with three-dimensional spatial coordinates; Data serialization module: used to execute step S3 as described in claim 1, serialize each two-dimensional datum object set into three-dimensional coordinates of datum points, construct independent data storage units according to the datum boundaries, and the names are directly related to the hierarchical division in step S1; Map compilation module: used to execute step S4 as described in claim 1, parse the data storage unit, directly construct lane objects using lane marking boundaries as lane left and right boundary definition functions, and compile to generate a standardized LD Map file conforming to the Opendrive specification.

Images