Lane-level high-precision map construction method and system

Abstract

This invention provides a lane-level high-precision map construction method and system, relating to the field of autonomous driving technology, including: a data acquisition step S1: acquiring point cloud and image data of the map to be constructed using a data acquisition vehicle; a vector map construction step S2: constructing relevant road geometric information, including lane lines, based on the point cloud, image, and auxiliary map data; a lane centerline construction step S3: constructing lane centerlines based on the road geometric information; a topology association construction step S4: constructing lane-related topology association information by combining the road geometric information and the lane centerlines; and a map data quality inspection step S5: inspecting the relevant map data, including the road geometric information, lane centerlines, and topology association information, to ensure the accuracy and quality of the high-precision map data, thereby improving the accuracy of road surfaces and intersection surfaces in the high-precision map.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, specifically to a method and system for constructing lane-level high-precision maps. Background Technology

[0002] High-precision maps, as opposed to ordinary maps, provide more accurate and richer map information, primarily serving autonomous driving. Currently, L2+ and above autonomous driving solutions generally rely heavily on high-precision maps.

[0003] In existing technologies, high-precision sensors are generally used to collect environmental data in order to make high-precision maps more accurate. When collecting environmental data on roads, sensors are usually mounted on a data collection vehicle, which then travels on the road to collect environmental data. High-precision maps are then built based on the collected environmental data. However, this requires collecting relevant data multiple times, which is labor-intensive and cannot guarantee data consistency, thus increasing the difficulty and cost of the map construction process. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for constructing lane-level high-precision maps.

[0005] According to the present invention, a method and system for constructing lane-level high-precision maps are provided, the solution of which is as follows:

[0006] Firstly, a method for constructing lane-level high-precision maps is provided, the method comprising:

[0007] Data acquisition step S1: Collect point cloud and image data of the map to be built using a data acquisition vehicle;

[0008] Vector map construction step S2: Based on the point cloud, image and auxiliary map data, construct relevant road geometry information including lane lines;

[0009] Lane centerline construction step S3: Construct the lane centerline based on road geometry information;

[0010] Topology association construction step S4: Combine the road geometry information and lane centerline to construct lane-related topology association information;

[0011] Map data quality inspection step S5: The relevant map data, including the road geometry information, lane center lines and topological association information, are inspected to ensure the accuracy and quality of the high-precision map data.

[0012] Preferably, the data acquisition step S1 further includes: acquiring GNSS, IMU and wheel speed meter data to obtain the positioning of the acquisition vehicle.

[0013] Preferably, the vector map construction step S2 specifically includes:

[0014] Step S2.1: Extract the virtual intersection based on the intersections and turning points generated by the vehicle's driving trajectory at the intersection;

[0015] Step S2.2: Extract the road stop line;

[0016] Step S2.3: Extract road point cloud data based on the virtual intersection and road stop line;

[0017] Step S2.4: Segment the road point cloud data based on elevation information;

[0018] Step S2.5: Divide the road according to its length, ensuring that each road segment is oriented in the same direction;

[0019] Step S2.6: Extract lane line point cloud and curb point cloud data from the road point cloud data within the road, and cluster and fit them into real lane lines and road curb lines;

[0020] Step S2.7: Extract lane line point cloud and curb point cloud data from the road point cloud data within the intersection, cluster and fit them into real lane lines and road curb lines, and at the same time introduce the lane lines within the road to perform topological association on the lane lines within the intersection so that their endpoints remain consistent.

[0021] Step S2.8: Smooth the lane lines;

[0022] Step S2.9: Generate a break line perpendicular to the road based on the start and end points of the lane lines within the road;

[0023] Step S2.10: Unify the direction of the lane lines, divide the lane lines and curb lines in the road according to the break line, so that the start and end points of the lane lines of each road block fall on the break line, extract the outermost line of each road block, and construct the two edge lines of the road according to the direction, thereby constructing the road surface data.

[0024] Step S2.11: Based on the lane lines within the intersection, extract the intersection boundary lines that intersect with both sides of the road, sort the intersection boundary lines around the center point of the intersection, and connect them in sequence to construct the intersection surface data;

[0025] Step S2.12: Based on the point cloud data of the intersection boundary line and the pedestrian crossing, cluster and fit the pedestrian crossing surface data;

[0026] Step S2.13: Based on the point cloud data and type of the directional arrow, replace it with a standard arrow graphic to construct the directional arrow surface data.

[0027] Preferably, the lane centerline construction step S3 includes:

[0028] Step S3.1: Based on the road surface data and intersection surface data, divide them into lane lines and road trajectories;

[0029] Step S3.2: Based on the lane lines and road trajectory, adjust the direction of the lane lines to form two lane lines in two directions, making them consistent with the direction of the road trajectory;

[0030] Step S3.3: Based on the direction of the adjusted lane lines, divide the road into two separate directional roads;

[0031] Step S3.4: Based on the roads in different directions, generate break lines according to the locations of road changes and breaks, and in conjunction with road trend lines;

[0032] Step S3.5: Divide the road into segments based on the lane lines and their corresponding break lines, and generate lane center lines in accordance with road rules;

[0033] Step S3.6: At the intersection, generate the intersection lane center lines by combining the relationship between the lane center lines and directional arrows of the entrances and exits, as well as the road rules.

[0034] Preferably, the topology association construction step S4 includes:

[0035] Step S4.1: Associate the road surface data with the lane center line, lane lines, and road stop line;

[0036] Step S4.2: Associate the relationship between the adjacent lane lines;

[0037] Step S4.3: Associate the relationships between the upstream and downstream sides of the lane centerline;

[0038] Step S4.4: Associate the relationship between the lane centerline and the lane lines;

[0039] Step S4.5: Generate virtual lane lines, ensuring that there are lane lines on both sides of the center line of all lanes;

[0040] Step S4.6: Associate the road stop line with the lane center line;

[0041] Step S4.7: Establish the relationship between the arrow and the lane centerline, calculate the angle between the arrow and the lane centerline, and rotate the arrow accordingly;

[0042] Step S4.8: Associate the intersection surface data with the lane center line, lane lines, and road stop line;

[0043] Step S4.9: Associate the intersection surface with the lane centerline;

[0044] Step S4.10: Fill in the list of driving directions of the lane centerlines within the intersection based on the upstream and downstream relationships of the lane centerlines and the arrows;

[0045] Step S4.11: Establish the relationship between traffic lights, lane center lines, and road stop lines;

[0046] Step S4.12: Calculate the virtual and real attributes of the road based on the point cloud data and lane lines;

[0047] Step S4.13: Fill in the road elevation information and smooth it according to the road relationship.

[0048] Preferably, the map data quality inspection step S5 includes:

[0049] Step S5.1: Detect the self-intersection of lane lines and lane centerlines;

[0050] Step S5.2: Check the connection between the lane lines and the upstream and downstream of the lane center line;

[0051] Step S5.3: Inspect the junction between the road surface and the intersection surface;

[0052] Step S5.4: Detect the consistency between the attribute topology and geometric topology of the high-precision map;

[0053] Step S5.5: Detect abnormal data;

[0054] Step S5.6: Manually verify the abnormal data.

[0055] Secondly, a lane-level high-precision map construction system is provided, the system comprising:

[0056] Data acquisition module M1: Collects point cloud and image data of the map to be built using a data acquisition vehicle;

[0057] Vector map construction module M2: Constructs relevant road geometry information, including lane lines, based on the point cloud, image, and auxiliary map data;

[0058] Lane centerline construction module M3: Constructs lane centerlines based on road geometry information;

[0059] Topology association construction module M4: Combines the road geometry information and lane centerline to construct lane-related topology association information;

[0060] Map data quality inspection module M5: Inspects relevant map data, including road geometry information, lane center lines, and topological association information, to ensure the accuracy and quality of high-precision map data;

[0061] The data acquisition module M1 further includes: acquiring GNSS, IMU and wheel speed meter data to obtain the positioning of the acquisition vehicle.

[0062] Preferably, the vector map construction module M2 specifically includes:

[0063] Module M2.1: Extracts virtual intersections based on the intersections and turning points generated by the vehicle's trajectory at the intersection;

[0064] Module M2.2: Extract road stop lines;

[0065] Module M2.3: Extracts road point cloud data based on the virtual intersection and road stop line;

[0066] Module M2.4: Segment road point cloud data based on elevation information;

[0067] Module M2.5: Divides the road according to its length, ensuring that each road segment is oriented in the same direction;

[0068] Module M2.6: Extracts lane line point cloud and curb point cloud data from the road point cloud data within the road, and clusters and fits them into real lane lines and road curb lines;

[0069] Module M2.7: Extracts lane line point cloud and curb point cloud data from the road point cloud data within the intersection, clusters and fits them into real lane lines and road curb lines, and introduces lane lines within the road to perform topological association on the lane lines within the intersection to ensure that their endpoints remain consistent.

[0070] Module M2.8: Smooth lane markings;

[0071] Module M2.9: Generates a break line perpendicular to the road based on the start and end points of the lane lines within the road;

[0072] Module M2.10: Unify lane line direction, divide the lane lines and road curb lines in the road according to the break line, so that the start and end points of the lane lines of each road block fall on the break line, extract the outermost line of each road block, and construct the two edge lines of the road according to the direction, thereby constructing road surface data.

[0073] Module M2.11: Based on the lane lines within the intersection, extract the intersection boundary lines that intersect with both sides of the road, sort the intersection boundary lines around the center point of the intersection, and connect them in sequence to construct the intersection surface data;

[0074] Module M2.12: Based on the point cloud data of the intersection boundary line and the pedestrian crossing, cluster and fit the pedestrian crossing surface data;

[0075] Module M2.13: Based on the point cloud data and type of the directional arrows, replace them with standard arrow graphics to construct directional arrow surface data.

[0076] Preferably, the lane centerline construction module M3 includes:

[0077] Module M3.1: Based on the road surface data and intersection surface data, divide them into lane lines and road trajectories;

[0078] Module M3.2: Adjusts the direction of the lane lines based on the lane lines and road trajectory, dividing the lane lines into two directions to ensure they align with the direction of the road trajectory;

[0079] Module M3.3: Based on the direction of the adjusted lane lines, the road is divided into two separate directional roads;

[0080] Module M3.4: Based on the roads in the given directions, generate break lines according to the locations of road changes and breaks, and in conjunction with road trend lines;

[0081] Module M3.5: Based on the lane lines and their corresponding break lines, divide the road into segments and generate lane center lines in accordance with road rules;

[0082] Module M3.6: At intersections, the center lines of the lanes at the entrances and exits are generated by combining the relationship between the lane center lines and directional arrows and the road rules.

[0083] Preferably, the topology association construction module M4 includes:

[0084] Module M4.1: Associates road surface data with lane center lines, lane lines, and road stop lines;

[0085] Module M4.2: Associates the relationship between adjacent lane lines;

[0086] Module M4.3: Connects the upstream and downstream relationships of the lane centerline;

[0087] Module M4.4: Associates the relationship between the lane centerline and the lane lines;

[0088] Module M4.5: Generates virtual lane lines, ensuring that there are lane lines on both sides of the center line of all lanes;

[0089] Module M4.6: Associates the road stop line with the lane center line;

[0090] Module M4.7: Establishes the relationship between the arrow and the lane centerline, calculates the angle between the arrow and the lane centerline, and rotates the arrow accordingly;

[0091] Module M4.8: Associates intersection surface data with lane center lines, lane lines, and road stop lines;

[0092] Module M4.9: Associates the intersection surface with the lane centerline;

[0093] Module M4.10: Based on the upstream and downstream relationships of the lane centerlines and the arrows, populate the list of driving directions of the lane centerlines within the intersection;

[0094] Module M4.11: Establishes the relationship between traffic lights, lane center lines, and road stop lines;

[0095] Module M4.12: Calculates the virtual and real attributes of the road based on point cloud data and lane lines;

[0096] Module M4.13: Fills in road elevation information and smooths it according to road relationships;

[0097] The map data quality inspection module M5 includes:

[0098] Module M5.1: Detects the self-intersection of lane lines and lane centerlines;

[0099] Module M5.2: Detects the connection status of lane lines and lane center lines upstream and downstream;

[0100] Module M5.3: Detects the edge connection between the road surface and the intersection surface;

[0101] Module M5.4: Detects consistency between attribute topology and geometric topology in high-precision maps;

[0102] Module M5.5: Detects abnormal data;

[0103] Module M5.6: Performs manual verification of abnormal data.

[0104] Compared with the prior art, the present invention has the following beneficial effects:

[0105] 1. This invention reduces the requirements for point cloud data by collecting point cloud data and image data, and combining them with existing auxiliary map data, thereby reducing the cost of data collection;

[0106] 2. This invention ensures and improves the accuracy of road and intersection surfaces in high-precision maps through vector map construction and topological association;

[0107] 3. This invention significantly improves the accuracy of high-precision map construction by manually verifying and correcting abnormal data during the final manual verification. Attached Figure Description

[0108] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0109] Figure 1This is a flowchart of the lane-level high-precision map construction method of this application;

[0110] Figure 2 for Figure 1 A schematic diagram of a scenario from the embodiment shown in the process flow;

[0111] Figure 3 This is a rendering of the lane-level high-precision map construction method of this application. Detailed Implementation

[0112] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0113] This invention provides a method for constructing lane-level high-precision maps, such as... Figure 1 and Figure 2 As shown, it includes:

[0114] Data acquisition step S1: Collect point cloud and image data of the map to be built using a data acquisition vehicle;

[0115] In one embodiment, the method further includes collecting GNSS, IMU, and wheel speed meter data to obtain the location of the data collection vehicle.

[0116] Vector map construction step S2: Based on the point cloud, image and auxiliary map data, construct relevant road geometry information including lane lines;

[0117] In one embodiment, the vector map construction step S2 specifically includes:

[0118] Step S2.1: Extract the virtual intersection based on the intersections and turning points generated by the vehicle's driving trajectory at the intersection;

[0119] Step S2.2: Extract the road stop line;

[0120] Step S2.3: Extract road point cloud data based on the virtual intersection and road stop line;

[0121] Step S2.4: Segment the road point cloud data based on elevation information;

[0122] Step S2.5: Divide the road according to its length, ensuring that each road segment is oriented in the same direction;

[0123] Step S2.6: Extract lane line point cloud and curb point cloud data from the road point cloud data within the road, and cluster and fit them into real lane lines and road curb lines;

[0124] Step S2.7: Extract lane line point cloud and curb point cloud data from the road point cloud data within the intersection, cluster and fit them into real lane lines and road curb lines, and at the same time introduce the lane lines within the road to perform topological association on the lane lines within the intersection so that their endpoints remain consistent.

[0125] Step S2.8: Smooth the lane lines;

[0126] Step S2.9: Generate a break line perpendicular to the road based on the start and end points of the lane lines within the road;

[0127] Step S2.10: Unify the direction of the lane lines, divide the lane lines and curb lines in the road according to the break line, so that the start and end points of the lane lines of each road block fall on the break line, extract the outermost line of each road block, and construct the two edge lines of the road according to the direction, thereby constructing the road surface data.

[0128] Step S2.11: Based on the lane lines within the intersection, extract the intersection boundary lines that intersect with both sides of the road, sort the intersection boundary lines around the center point of the intersection, and connect them in sequence to construct the intersection surface data;

[0129] Step S2.12: Based on the point cloud data of the intersection boundary line and the pedestrian crossing, cluster and fit the pedestrian crossing surface data;

[0130] Step S2.13: Based on the point cloud data and type of the directional arrow, replace it with a standard arrow graphic to construct the directional arrow surface data.

[0131] Lane centerline construction step S3: Construct the lane centerline based on road geometry information;

[0132] In one embodiment, lane centerline construction step S3 includes:

[0133] Step S3.1: Based on the road surface data and intersection surface data, divide them into lane lines and road trajectories;

[0134] Step S3.2: Based on the lane lines and road trajectory, adjust the direction of the lane lines to form two lane lines in two directions, making them consistent with the direction of the road trajectory;

[0135] Step S3.3: Based on the direction of the adjusted lane lines, divide the road into two separate directional roads;

[0136] Step S3.4: Based on the roads in different directions, generate break lines according to the locations of road changes and breaks, and in conjunction with road trend lines;

[0137] Step S3.5: Divide the road into segments based on the lane lines and their corresponding break lines, and generate lane center lines in accordance with road rules;

[0138] Step S3.6: At the intersection, generate the intersection lane center lines by combining the relationship between the lane center lines and directional arrows of the entrances and exits, as well as the road rules.

[0139] Topology association construction step S4: Combine the road geometry information and lane centerline to construct lane-related topology association information;

[0140] In one embodiment, the topology association construction step S4 includes:

[0141] Step S4.1: Associate the road surface data with the lane center line, lane lines, and road stop line;

[0142] Step S4.2: Associate the relationship between the adjacent lane lines;

[0143] Step S4.3: Link the upstream and downstream relationships of the lane centerline;

[0144] Step S4.4: Associate the relationship between the lane centerline and the lane lines;

[0145] Step S4.5: Generate virtual lane lines, ensuring that there are lane lines on both sides of the center line of all lanes;

[0146] Step S4.6: Associate the road stop line with the lane center line;

[0147] Step S4.7: Establish the relationship between the arrow and the lane centerline, calculate the angle between the arrow and the lane centerline, and rotate the arrow accordingly;

[0148] Step S4.8: Associate the intersection surface data with the lane center line, lane lines, and road stop line;

[0149] Step S4.9: Associate the intersection surface with the lane centerline;

[0150] Step S4.10: Fill in the list of driving directions of the lane centerlines within the intersection based on the upstream and downstream relationships of the lane centerlines and the arrows;

[0151] Step S4.11: Establish the relationship between traffic lights, lane center lines, and road stop lines;

[0152] Step S4.12: Calculate the virtual and real attributes of the road based on the point cloud data and lane lines;

[0153] Step S4.13: Fill in the road elevation information and smooth it according to the road relationship.

[0154] Map data quality inspection step S5: The relevant map data, including the road geometry information, lane center lines and topological association information, are inspected to ensure the accuracy and quality of the high-precision map data.

[0155] In one embodiment, map data quality inspection step S5 includes:

[0156] Step S5.1: Detect the self-intersection of lane lines and lane centerlines;

[0157] Step S5.2: Check the connection between the lane lines and the upstream and downstream of the lane center line;

[0158] Step S5.3: Inspect the junction between the road surface and the intersection surface;

[0159] Step S5.4: Detect the consistency between the attribute topology and geometric topology of the high-precision map;

[0160] Step S5.5: Detect abnormal data;

[0161] Step S5.6: Manually verify the abnormal data.

[0162] The present invention also provides a lane-level high-precision map construction system, comprising:

[0163] Data acquisition module M1: Collects point cloud and image data of the map to be built using a data acquisition vehicle;

[0164] Vector map construction module M2: Constructs relevant road geometry information, including lane lines, based on the point cloud, image, and auxiliary map data;

[0165] Lane centerline construction module M3: Constructs lane centerlines based on road geometry information;

[0166] Topology association construction module M4: Combines the road geometry information and lane centerline to construct lane-related topology association information;

[0167] Map data quality inspection module M5: Inspects relevant map data, including road geometry information, lane center lines, and topological association information, to ensure the accuracy and quality of high-precision map data;

[0168] The data acquisition module M1 further includes: acquiring GNSS, IMU and wheel speed meter data to obtain the positioning of the acquisition vehicle.

[0169] Furthermore, the vector map building module M2 specifically includes:

[0170] Module M2.1: Extracts virtual intersections based on the intersections and turning points generated by the vehicle's trajectory at the intersection;

[0171] Module M2.2: Extract road stop lines;

[0172] Module M2.3: Extracts road point cloud data based on the virtual intersection and road stop line;

[0173] Module M2.4: Segment road point cloud data based on elevation information;

[0174] Module M2.5: Divides the road according to its length, ensuring that each road segment is oriented in the same direction;

[0175] Module M2.6: Extracts lane line point cloud and curb point cloud data from the road point cloud data within the road, and clusters and fits them into real lane lines and road curb lines;

[0176] Module M2.7: Extracts lane line point cloud and curb point cloud data from the road point cloud data within the intersection, clusters and fits them into real lane lines and road curb lines, and introduces lane lines within the road to perform topological association on the lane lines within the intersection to ensure that their endpoints remain consistent.

[0177] Module M2.8: Smooth lane markings;

[0178] Module M2.9: Generates a break line perpendicular to the road based on the start and end points of the lane lines within the road;

[0179] Module M2.10: Unify lane line direction, divide the lane lines and road curb lines in the road according to the break line, so that the start and end points of the lane lines of each road block fall on the break line, extract the outermost line of each road block, and construct the two edge lines of the road according to the direction, thereby constructing road surface data.

[0180] Module M2.11: Based on the lane lines within the intersection, extract the intersection boundary lines that intersect with both sides of the road, sort the intersection boundary lines around the center point of the intersection, and connect them in sequence to construct the intersection surface data;

[0181] Module M2.12: Based on the point cloud data of the intersection boundary line and the pedestrian crossing, cluster and fit the pedestrian crossing surface data;

[0182] Module M2.13: Based on the point cloud data and type of the directional arrows, replace them with standard arrow graphics to construct directional arrow surface data.

[0183] Furthermore, the lane centerline construction module M3 includes:

[0184] Module M3.1: Based on the road surface data and intersection surface data, divide them into lane lines and road trajectories;

[0185] Module M3.2: Adjusts the direction of the lane lines based on the lane lines and road trajectory, dividing the lane lines into two directions to ensure they align with the direction of the road trajectory;

[0186] Module M3.3: Based on the direction of the adjusted lane lines, the road is divided into two separate directional roads;

[0187] Module M3.4: Based on the roads in the given directions, generate break lines according to the locations of road changes and breaks, and in conjunction with road trend lines;

[0188] Module M3.5: Based on the lane lines and their corresponding break lines, divide the road into segments and generate lane center lines in accordance with road rules;

[0189] Module M3.6: At intersections, the center lines of the lanes at the entrances and exits are generated by combining the relationship between the lane center lines and directional arrows and the road rules.

[0190] The topology association building module M4 includes:

[0191] Module M4.1: Associates road surface data with lane center lines, lane lines, and road stop lines;

[0192] Module M4.2: Associates the relationship between adjacent lane lines;

[0193] Module M4.3: Connects the upstream and downstream relationships of the lane centerline;

[0194] Module M4.4: Associates the relationship between the lane centerline and the lane lines;

[0195] Module M4.5: Generates virtual lane lines, ensuring that there are lane lines on both sides of the center line of all lanes;

[0196] Module M4.6: Associates the road stop line with the lane center line;

[0197] Module M4.7: Establishes the relationship between the arrow and the lane centerline, calculates the angle between the arrow and the lane centerline, and rotates the arrow accordingly;

[0198] Module M4.8: Associates intersection surface data with lane center lines, lane lines, and road stop lines;

[0199] Module M4.9: Associates the intersection surface with the lane centerline;

[0200] Module M4.10: Based on the upstream and downstream relationships of the lane centerlines and the arrows, populate the list of driving directions of the lane centerlines within the intersection;

[0201] Module M4.11: Establishes the relationship between traffic lights, lane center lines, and road stop lines;

[0202] Module M4.12: Calculates the virtual and real attributes of the road based on point cloud data and lane lines;

[0203] Module M4.13: Fills in road elevation information and smooths it according to road relationships.

[0204] The map data quality inspection module M5 includes:

[0205] Module M5.1: Detects the self-intersection of lane lines and lane centerlines;

[0206] Module M5.2: Detects the connection status of lane lines and lane center lines upstream and downstream;

[0207] Module M5.3: Detects the edge connection between the road surface and the intersection surface;

[0208] Module M5.4: Detects consistency between attribute topology and geometric topology in high-precision maps;

[0209] Module M5.5: Detects abnormal data;

[0210] Module M5.6: Performs manual verification of abnormal data.

[0211] In one embodiment, more specifically, lane-level high-precision map construction relates to the following:

[0212] Data processing:

[0213] The data is mainly divided into two categories: point clouds and images. Due to high precision requirements, point cloud mapping is the primary method. Creating a point cloud map requires two crucial pieces of information: location information and point cloud information. While GNSS, IMU, and wheel speedometers can frequently acquire the location of the data acquisition vehicle, each sensor has inherent accuracy limitations. Therefore, the data from these sensors needs to be fused, and then the SLAM algorithm is applied to correct the position, ultimately yielding a relatively accurate location. The point cloud information, derived from LiDAR, is based on environmental scanning and includes XYZIT (X, Y, Z coordinates, light intensity, timestamp) information. Finally, the point cloud information is fused to the location information through temporal correlation matching to construct the point cloud map.

[0214] Element recognition:

[0215] Deep learning based on reflection maps can extract information such as lane lines, light poles, and traffic lights, and obtain the shape features of these road facilities.

[0216] Map building:

[0217] Based on the above data, we have the basic data and geometric features required for high-precision map construction. By using traditional geometric and GIS algorithms, machine learning and related spatial processing methods, we can construct a map model and then combine relevant intelligent algorithms and spatial topology logic to build a high-precision map.

[0218] Reference Figure 2 The main architecture shown is as follows: 1) Data service layer: mainly used to store point cloud data, image data, trajectory data, vector geometric data and related text information related to high-precision traffic maps.

[0219] 2) Data Intermediate Layer:

[0220] It primarily handles various general files, general databases, big data platforms, and spatial data services related to high-precision transportation maps, facilitating data conversion and efficient reading and writing.

[0221] 3) Model layer:

[0222] It primarily interfaces with high-precision map model structures required for intelligent driving, smart cities, and other related businesses, assisting in calculation, analysis, mapping, and storage.

[0223] 4) Algorithm Engine:

[0224] It includes various efficient spatial algorithms and data analysis, statistical algorithms, etc., and mainly serves functions such as spatial geometric modeling, automated mapping, data statistical analysis, spatial data mining, and path planning and navigation.

[0225] Next, the construction of lane-level high-precision maps according to the present invention will be described in detail.

[0226] This invention provides a method for constructing lane-level high-precision maps, such as... Figure 1 As shown, it specifically includes:

[0227] Data acquisition steps: Collect point cloud and image data of the map to be built using a data acquisition vehicle; data acquisition includes: collecting GNSS, IMU and wheel speed measurement data, and obtaining the positioning of the data acquisition vehicle.

[0228] Vector map construction steps: Based on the point cloud, image and auxiliary map data, the vectorization algorithm is used to automatically construct real road geometric information such as lane lines, curbs, guide lines, stop lines, directional arrows, ground text, intersections and road surfaces.

[0229] Specifically, the steps for constructing a vector map include:

[0230] 1) The trajectory data is used to extract virtual intersections based on intersections and turning points; specifically, the trajectory of a vehicle at an intersection will generate intersections and turning points, such as when going straight or turning.

[0231] 2) Extract the road stop line.

[0232] Supplementary map data is sufficient, such as Baidu Maps, Advanced Maps, etc. A stop line is a reference line for vehicles waiting at traffic lights, indicating where to stop. It is generally located at: 1. Traffic signal-controlled intersections. 2. Railway level crossings. 3. The front end of a left-turn waiting area.

[0233] 3) Extract road point cloud data based on the virtual intersection and road stop line.

[0234] 4) Segment the road point cloud data based on elevation information. Elevation information here refers to the information within the point cloud itself. The point cloud includes three-dimensional coordinates, and elevation is the vertical height above a given reference surface (such as the foundation, ground, or sea surface), which can be understood as the coordinate value of the Z-axis.

[0235] 5) Divide the road according to its length, ensuring that each road segment is oriented in the same direction.

[0236] 6) Extract lane line point cloud and curb point cloud data from the road point cloud data, and use a combination of machine learning and computational geometry to cluster and fit them into real lane lines and road curb lines, etc.

[0237] 7) Extract lane line point cloud and curb point cloud data from the road point cloud data within the intersection, and use a combination of machine learning and computational geometry to cluster and fit them into real lane lines and road curb lines, etc. At the same time, introduce the lane lines within the road to perform topological association on the lane lines within the intersection to ensure that their endpoints are consistent.

[0238] 8) Smooth lane markings.

[0239] 9) Generate a break line perpendicular to the road based on the start and end points of the lane lines within the road.

[0240] 10) Unify the direction of lane lines, divide the lane lines and curb lines in the road according to the break line, so that the start and end points of the lane lines of each road block fall on the break line, extract the outermost line of each road block, and construct the two edge lines of the road according to the direction, thereby constructing detailed road surface data.

[0241] 11) Based on the lane lines within the intersection, extract the intersection boundary lines that intersect with both sides of the road, sort the intersection boundary lines around the center point of the intersection, and connect them in sequence to construct the intersection surface data.

[0242] 12) Based on the point cloud data of the intersection boundary line and the pedestrian crossing, the pedestrian crossing surface data is obtained by clustering and fitting.

[0243] 13) Based on the point cloud data and type of the directional arrows, replace them with standard arrow graphics to construct directional arrow surface data, etc.

[0244] The steps for constructing a lane centerline are as follows: By combining information on changes in the road surface and lanes, the relationship between the road and the intersection, geometric reasoning, and relevant traffic rules, the lane centerline (i.e., the lane driving line) is constructed.

[0245] Specifically, the steps for constructing the lane centerline include:

[0246] (1) Based on the road surface data and intersection surface data, they are divided into lane lines and road trajectories.

[0247] (2) Adjust the direction of the lane lines according to the lane lines and the road trajectory, and divide the lane lines into two directions to make them consistent with the direction of the road trajectory.

[0248] (3) Based on the direction of the adjusted lane lines, the road is divided into two separate roads with different directions.

[0249] (4) Based on the roads in different directions, generate break lines according to the locations of road changes and breaks, and in conjunction with road trend lines.

[0250] (5) Based on the lane lines and their corresponding break lines, divide the road into segments and generate lane center lines in accordance with road rules.

[0251] (6) At the intersection, the center line of the lanes at the entrance and exit is generated by combining the relationship between the lane center lines and the directional arrows and the road rules.

[0252] Actual rules: Road width should be at least 2.5 meters on average; lane changes are prohibited over solid lines; lane changes should be made according to the nearest available lane, etc.

[0253] Topology association construction steps: Combining road geometry information and lane centerlines, construct lane-related topology association information, including lane upstream and downstream relationships, adjacency relationships, intersection association information, and elevation information.

[0254] Specifically, the steps for constructing topological associations include:

[0255] 1. Associate the road surface data with the lane center line, lane lines, and road stop line.

[0256] 2. Associativity of the relationship between adjacent lane lines.

[0257] 3. Connect the upstream and downstream relationships of the lane centerline.

[0258] 4. Establish a connection between the lane center line and the lane lines.

[0259] 5. Generate virtual lane lines to ensure that there are lane lines on both sides of the center line of all lanes.

[0260] 6. Associate the road stop line with the lane center line.

[0261] 7. Establish the relationship between the arrow and the lane centerline, calculate the angle between the arrow and the lane centerline, and rotate the arrow accordingly.

[0262] 8. Associate the intersection surface data with the lane center line, lane lines, and road stop line;

[0263] 9. Connect the intersection surface to the lane centerline.

[0264] 10. Based on the upstream and downstream relationships of the lane centerlines and the arrows, fill in the list of driving directions of the lane centerlines within the intersection.

[0265] 11. Establish the relationship between traffic lights, lane center lines, and road stop lines.

[0266] 12. Calculate the virtual and real attributes of the road based on point cloud data and lane lines.

[0267] 13. Fill in the road data elevation and smooth it according to the road relationships.

[0268] Map data quality inspection steps: The relevant map data, including road geometry information, lane center lines, and topological association information, are inspected to ensure the accuracy and overall quality of the high-precision map data.

[0269] Specifically, the map data quality inspection steps include:

[0270] 1) Detect the self-intersection of lane lines and lane center lines;

[0271] 2) Inspect the connection between the lane lines and the lane center line upstream and downstream;

[0272] 3) Inspect the junction between the road surface and the intersection surface;

[0273] 4) Detect the consistency between the attribute topology and geometric topology of the high-precision map; attribute topology is a formal structure representation method, a graph structure that represents attributes in a formal context as nodes and the degree of coupling between objects and attributes as the association between nodes. If the attributes are replaced with computers, this structure is like a computer network topology description, hence the name attribute topology. The attributes of the high-precision map itself are compared with the geometric topology here to detect the consistency between the attributes and geometric topology. Figure 1 Sexuality issues.

[0274] 5) Detect abnormal data;

[0275] 6) Manual verification.

[0276] like Figure 3 The image shown is a rendering of the lane-level high-precision map construction method of this application.

[0277] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0278] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for constructing lane-level high-precision maps, characterized in that, include: Data acquisition step S1: Collect point cloud and image data of the map to be built using a data acquisition vehicle; Vector map construction step S2: Based on the point cloud, image and auxiliary map data, construct relevant road geometry information including lane lines; Lane centerline construction step S3: Construct the lane centerline based on road geometry information; Topology association construction step S4: Combine the road geometry information and lane centerline to construct lane-related topology association information; Map data quality inspection step S5: The relevant map data, including the road geometry information, lane center lines, and topological association information, are inspected to ensure the accuracy and quality of the high-precision map data; The data acquisition step S1 further includes: acquiring GNSS, IMU and wheel speed meter data to obtain the positioning of the acquisition vehicle; The vector map construction step S2 specifically includes: Step S2.1: Extract the virtual intersection based on the intersections and turning points generated by the vehicle's driving trajectory at the intersection; Step S2.2: Extract the road stop line; Step S2.3: Extract road point cloud data based on the virtual intersection and road stop line; Step S2.4: Segment the road point cloud data based on elevation information; Step S2.5: Divide the road according to its length, ensuring that each road segment is oriented in the same direction; Step S2.6: Extract lane line point cloud and curb point cloud data from the road point cloud data within the road, and cluster and fit them into real lane lines and road curb lines; Step S2.7: Extract lane line point cloud and curb point cloud data from the road point cloud data within the intersection, cluster and fit them into real lane lines and road curb lines, and at the same time introduce the lane lines within the road to perform topological association on the lane lines within the intersection so that their endpoints remain consistent. Step S2.8: Smooth the lane lines; Step S2.9: Generate a break line perpendicular to the road based on the start and end points of the lane lines within the road; Step S2.10: Unify the direction of the lane lines, divide the lane lines and curb lines in the road according to the break line, so that the start and end points of the lane lines of each road block fall on the break line, extract the outermost line of each road block, and construct the two edge lines of the road according to the direction, thereby constructing the road surface data. Step S2.11: Based on the lane lines within the intersection, extract the intersection boundary lines that intersect with both sides of the road, sort the intersection boundary lines around the center point of the intersection, and connect them in sequence to construct the intersection surface data; Step S2.12: Based on the point cloud data of the intersection boundary line and the pedestrian crossing, cluster and fit the pedestrian crossing surface data; Step S2.13: Based on the point cloud data and type of the directional arrows, replace them with standard arrow graphics to construct directional arrow surface data; The topology association construction step S4 includes: Step S4.1: Associate the road surface data with the lane center line, lane lines, and road stop line; Step S4.2: Associate the relationship between adjacent lane lines; Step S4.3: Link the upstream and downstream relationships of the lane centerline; Step S4.4: Associate the relationship between the lane centerline and the lane lines; Step S4.5: Generate virtual lane lines, ensuring that there are lane lines on both sides of the center line of all lanes; Step S4.6: Associate the road stop line with the lane center line; Step S4.7: Establish the relationship between the arrow and the lane centerline, calculate the angle between the arrow and the lane centerline, and rotate the arrow accordingly; Step S4.8: Associate the intersection surface data with the lane center line, lane lines, and road stop line; Step S4.9: Associate the intersection surface with the lane centerline; Step S4.10: Fill in the list of driving directions of the lane centerlines within the intersection based on the upstream and downstream relationships of the lane centerlines and the arrows; Step S4.11: Establish the relationship between traffic lights, lane center lines, and road stop lines; Step S4.12: Calculate the virtual and real attributes of the road based on the point cloud data and lane lines; Step S4.13: Fill in the road elevation information and smooth it according to the road relationship.

2. The lane-level high-precision map construction method according to claim 1, characterized in that, The lane centerline construction step S3 includes: Step S3.1: Based on the road surface data and intersection surface data, divide them into lane lines and road trajectories; Step S3.2: Based on the lane lines and road trajectory, adjust the direction of the lane lines to form two lane lines in two directions, making them consistent with the direction of the road trajectory; Step S3.3: Based on the direction of the adjusted lane lines, divide the road into two separate directional roads with different directions; Step S3.4: Based on the roads in different directions, generate break lines according to the locations of road changes and breaks, and in conjunction with road trend lines; Step S3.5: Divide the road into segments based on the lane lines and their corresponding break lines, and generate lane center lines in accordance with road rules; Step S3.6: At the intersection, generate the intersection lane center lines by combining the relationship between the lane center lines and directional arrows of the entrances and exits, as well as the road rules.

3. The lane-level high-precision map construction method according to claim 1, characterized in that, The map data quality inspection step S5 includes: Step S5.1: Detect the self-intersection of lane lines and lane centerlines; Step S5.2: Check the connection between the lane lines and the upstream and downstream of the lane center line; Step S5.3: Inspect the junction between the road surface and the intersection surface; Step S5.4: Detect the consistency between the attribute topology and geometric topology of the high-precision map; Step S5.5: Detect abnormal data; Step S5.6: Manually verify the abnormal data.

4. A lane-level high-precision map construction system, characterized in that, include: Data acquisition module M1: Collects point cloud and image data of the map to be built using a data acquisition vehicle; Vector map construction module M2: Constructs relevant road geometry information, including lane lines, based on the point cloud, image, and auxiliary map data; Lane centerline construction module M3: Constructs lane centerlines based on road geometry information; Topology association construction module M4: Combines the road geometry information and lane centerline to construct lane-related topology association information; Map data quality inspection module M5: Inspects relevant map data, including road geometry information, lane center lines, and topological association information, to ensure the accuracy and quality of high-precision map data; The data acquisition module M1 further includes: acquiring GNSS, IMU and wheel speed meter data to obtain the positioning of the acquisition vehicle; The vector map construction module M2 specifically includes: Module M2.1: Extracts virtual intersections based on the intersections and turning points generated by the vehicle's trajectory at the intersection; Module M2.2: Extract road stop lines; Module M2.3: Extracts road point cloud data based on the virtual intersection and road stop line; Module M2.4: Segment road point cloud data based on elevation information; Module M2.5: Divides the road according to its length, ensuring that each road segment is oriented in the same direction; Module M2.6: Extracts lane line point cloud and curb point cloud data from the road point cloud data within the road, and clusters and fits them into real lane lines and road curb lines; Module M2.7: Extracts lane line point cloud and curb point cloud data from the road point cloud data within the intersection, clusters and fits them into real lane lines and road curb lines, and introduces lane lines within the road to perform topological association on the lane lines within the intersection to ensure that their endpoints remain consistent. Module M2.8: Smooth lane markings; Module M2.9: Generates a break line perpendicular to the road based on the start and end points of the lane lines within the road; Module M2.10: Unify lane line direction, divide the lane lines and road curb lines in the road according to the break line, so that the start and end points of the lane lines of each road block fall on the break line, extract the outermost line of each road block, and construct the two edge lines of the road according to the direction, thereby constructing road surface data. Module M2.11: Based on the lane lines within the intersection, extract the intersection boundary lines that intersect with both sides of the road, sort the intersection boundary lines around the center point of the intersection, and connect them in sequence to construct the intersection surface data; Module M2.12: Based on the point cloud data of the intersection boundary line and the pedestrian crossing, cluster and fit the pedestrian crossing surface data; Module M2.13: Based on the point cloud data and type of the directional arrows, replace them with standard arrow graphics to construct directional arrow surface data; The topology association construction module M4 includes: Module M4.1: Associates road surface data with lane center lines, lane lines, and road stop lines; Module M4.2: Associates the relationship between adjacent lane lines; Module M4.3: Connects the upstream and downstream relationships of the lane centerline; Module M4.4: Associates the relationship between the lane centerline and the lane lines; Module M4.5: Generates virtual lane lines, ensuring that there are lane lines on both sides of the center line of all lanes; Module M4.6: Associates the road stop line with the lane center line; Module M4.7: Establishes the relationship between the arrow and the lane centerline, calculates the angle between the arrow and the lane centerline, and rotates the arrow accordingly; Module M4.8: Associates intersection surface data with lane center lines, lane lines, and road stop lines; Module M4.9: Associates the intersection surface with the lane centerline; Module M4.10: Based on the upstream and downstream relationships of the lane centerlines and the arrows, populate the list of driving directions of the lane centerlines within the intersection; Module M4.11: Establishes the relationship between traffic lights, lane center lines, and road stop lines; Module M4.12: Calculates the virtual and real attributes of the road based on point cloud data and lane lines; Module M4.13: Fills in road elevation information and smooths it according to road relationships.

5. The lane-level high-precision map construction system according to claim 4, characterized in that, The lane centerline construction module M3 includes: Module M3.1: Based on the road surface data and intersection surface data, divide them into lane lines and road trajectories; Module M3.2: Adjusts the direction of the lane lines based on the lane lines and road trajectory, dividing the lane lines into two directions to ensure they align with the direction of the road trajectory; Module M3.3: Based on the direction of the adjusted lane lines, the road is divided into two separate directional roads; Module M3.4: Based on the roads in the given directions, generate break lines according to the locations of road changes and breaks, and in conjunction with road trend lines; Module M3.5: Based on the lane lines and their corresponding break lines, divide the road into segments and generate lane center lines in accordance with road rules; Module M3.6: At intersections, the center lines of the lanes at the entrances and exits are generated by combining the relationship between the lane center lines and directional arrows and the road rules.

6. The lane-level high-precision map construction system according to claim 4, characterized in that, The map data quality inspection module M5 includes: Module M5.1: Detects the self-intersection of lane lines and lane centerlines; Module M5.2: Detects the connection status of lane lines and lane center lines upstream and downstream; Module M5.3: Detects the edge connection between the road surface and the intersection surface; Module M5.4: Detects consistency between attribute topology and geometric topology in high-precision maps; Module M5.5: Detects abnormal data; Module M5.6: Performs manual verification of abnormal data.

Images