Basement scene automatic vector map construction method

Abstract

The invention provides a basement scene automatic vector map construction method and system, and the method comprises the steps: carrying out the IPM conversion of a plurality of ground-oriented top view images, carrying out the splicing, obtaining a bird's-eye view, carrying out the semantic segmentation of the bird's-eye view, extracting semantic elements, and converting the semantic elements into point cloud data; providing IMU and vehicle wheel speed data, combining with the point cloud data to construct a local map, and correcting to obtain a semantic point cloud map; preprocessing the semantic point cloud map, and clustering the point cloud data in the semantic point cloud map according to the similarity semantic point cloud to obtain clustered point cloud data; and based on the semantic point cloud map, combining the clustering point cloud data, constructing boundaries on two sides of a road, generating a lane center line and a boundary line, obtaining a vector map, embedding parking space information, and performing curve optimization on the center line and the boundary line, so as to obtain a basement scene automatic vector map. Vector representation of the complex environment of the underground garage is achieved, and high-quality basic map data support is provided for applications such as vehicle navigation and path planning.

Description

Technical Field

[0001] The present invention relates to the technical fields of vector map construction and Automated Valet Parking (AVP), and specifically, to an automated vector map construction method for basement scenarios. Background Art

[0002] As a key technology in the field of Automated Valet Parking (AVP), the automated vector map construction technology for basement scenarios has received extensive attention from the academic and industrial communities in recent years. As a highly enclosed and complex environment space, the unique building structure and the existence of interference information in the underground garage pose significant challenges to the accurate mapping of the basement map.

[0003] Existing vector map construction technologies are generally divided into the following two types: one is manual surveying and mapping, which has the disadvantages of long time consumption and easy errors, and is difficult to adapt to the frequent changes in the basement environment; the other is to use low-precision sensor data, but the generated map details are blurred, unable to accurately reflect the complex structure in the basement, and lack a dynamic update mechanism, resulting in a lag in response to newly added or removed obstacles, affecting the accuracy and safety of navigation. Summary of the Invention

[0004] In view of the above deficiencies in the prior art, the present invention provides an automated vector map construction method for basement scenarios, aiming to achieve the vector representation of the complex environment of the underground garage through high-precision image data acquisition, intelligent algorithm processing, and automated vector map construction technology, and provide high-quality basic map data support for applications such as vehicle navigation and path planning.

[0005] According to one aspect of the present invention, an automated vector map construction method for basement scenarios is provided, including:

[0006] Performing Inverse Perspective Mapping (IPM) transformation on multiple top-down images facing the ground and then stitching them to obtain an aerial view;

[0007] Performing semantic segmentation on the aerial view, extracting semantic elements therein, and converting the semantic elements into point cloud data;

[0008] Providing Inertial Measurement Unit (IMU) and vehicle wheel speedometer data, combining the IMU (Inertial Measurement Unit) and vehicle wheel speedometer data with the point cloud data to construct a local map, and using loop detection to correct the cumulative error of the local map to obtain a semantic point cloud map;

[0009] Performing preprocessing on the semantic point cloud map, and then clustering the point cloud data in the preprocessed semantic point cloud map according to the set similarity semantic points to obtain clustered point cloud data;

[0010] Based on the semantic point cloud map and in combination with the clustered point cloud data, construct the boundaries on both sides of the road, generate the lane center line and the boundary line, and obtain a vector map;

[0011] Embed parking space information in the vector map, and perform curve optimization on the lane center line and the boundary line in the vector map to construct an automated vector map for the basement scenario.

[0012] According to another aspect of the present invention, there is provided a system for constructing an automated vector map for a basement scenario, including:

[0013] A semantic point cloud map construction module, which is used to perform IPM transformation on multiple ground-facing overhead images and then splice them to obtain an aerial view; perform semantic segmentation on the aerial view, extract the semantic elements therein, and convert the semantic elements into point cloud data; provide IMU (Inertial Measurement Unit) and vehicle wheel speedometer data, combine the IMU (Inertial Measurement Unit) and vehicle wheel speedometer data with the point cloud data to construct a local map, and use loop detection to correct the cumulative error of the local map to obtain a semantic point cloud map;

[0014] A vector map construction module, which is used to preprocess the semantic point cloud map, and then cluster the point cloud data in the preprocessed semantic point cloud map according to the set similarity semantic point cloud to obtain clustered point cloud data; based on the semantic point cloud map, construct the boundaries on both sides of the road, and generate the lane center line and the boundary line in the clustered point cloud data to obtain a vector map; embed parking space information in the vector map, and perform curve optimization on the lane center line and the boundary line in the vector map to construct an automated vector map for the basement scenario.

[0015] According to a third aspect of the present invention, there is provided a computer terminal, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it can be used to execute the method described in any one of the above of the present invention, or, run the system described in the above of the present invention.

[0016] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it can be used to execute the method described in any one of the above of the present invention, or, run the system described in the above of the present invention.

[0017] Due to the adoption of the above technical solutions, compared with the prior art, the present invention has at least one of the following beneficial effects:

[0018] The present invention collects basement environment data through a fish-eye camera, and uses IPM image processing technology to accurately stitch the collected image data into a panoramic bird's-eye view. Further, semantic information such as lane lines and arrows is extracted from the bird's-eye view, and a semantic point cloud map is constructed based on this semantic information, realizing high-precision scanning and detail restoration of the basement environment, ensuring the accuracy of map construction, and improving the precision of map construction.

[0019] The present invention simplifies the map construction process, reduces the need for human intervention through technologies such as automatically clustering obstacles and generating side boundaries, and improves the construction efficiency and reliability.

[0020] The present invention integrates a pure vision semantic mapping technology, combines data fusion of sensors such as a wheel speedometer and an IMU, quickly and accurately analyzes the collected images, extracts semantic information such as road signs and arrows, and constructs a high-precision and high-coverage automated semantic point cloud map, solving the technical problem of low precision in the traditional basement map construction process.

[0021] The present invention realizes efficient and accurate vector map construction in a complex environment through the fusion of point cloud filtering, point cloud clustering, and curve optimization strategies, providing a more reliable and efficient map supply solution for fields such as intelligent driving and robot navigation. Brief Description of the Drawings

[0022] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objectives, and advantages of the present invention will become more apparent:

[0023] Figure 1 It is a working flow chart of the method for constructing an automated vector map of a basement scene in an embodiment of the present invention.

[0024] Figure 2 It is a working schematic diagram of constructing a semantic point cloud map in a preferred embodiment of the present invention.

[0025] Figure 3 It is a working schematic diagram of constructing a vector map in a preferred embodiment of the present invention.

[0026] Figure 4 It is a schematic diagram of the composition modules of the system for constructing an automated vector map of a basement scene in an embodiment of the present invention. Detailed Embodiment

[0027] The following is a detailed description of the embodiments of the present invention: This embodiment is implemented on the premise of the technical solution of the present invention, and gives detailed implementation manners and specific operation processes. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several deformations and improvements can be made, and these all belong to the protection scope of the present invention.

[0028] In view of the defects existing in the map drawing method based on measurement means or low-precision sensor data in the prior art, an embodiment of the present invention provides a method for automatically constructing a vector map of a basement scenario. This method collects basement environment data through a fish-eye camera, combines the Simultaneous Localization and Mapping (SLAM) algorithm and the topological map generation technology to achieve the precise construction of a vector map of an underground garage, providing map data support for vehicle navigation and path planning, and having important practical application value for improving the quality of vehicle safety driving and efficient parking and other operation tasks, significantly improving the efficiency, accuracy and intelligent level of basement map drawing.

[0029] Specifically, as Figure 1 shown, the method for automatically constructing a vector map of a basement scenario provided by this embodiment may include the following operations:

[0030] S1, perform IPM transformation on multiple top-down images facing the ground and then splice them to obtain an aerial view;

[0031] S2, perform semantic segmentation on the aerial view, extract the semantic elements therein, and convert the semantic elements into point cloud data;

[0032] S3, provide IMU (Inertial Measurement Unit) and vehicle speedometer data, combine the IMU (Inertial Measurement Unit) and vehicle speedometer data with the point cloud data to construct a local map, and use loop detection to correct the cumulative error of the local map to obtain a semantic point cloud map;

[0033] S4, preprocess the semantic point cloud map, and then cluster the point cloud data in the preprocessed semantic point cloud map according to the set similar semantic point clouds to obtain clustered point cloud data;

[0034] S5, based on the semantic point cloud map, and in combination with the clustered point cloud data, construct the boundaries on both sides of the road, and generate lane centerlines and boundary lines to obtain a vector map;

[0035] S6, embed parking space information in the vector map, and perform curve optimization on the lane centerlines and boundary lines in the vector map to construct an automatically generated vector map of the basement scenario.

[0036] The following further details the technical solutions provided by the embodiments of the present invention in conjunction with the preferred embodiments.

[0037] The method for automatically constructing a vector map of a basement scenario provided by the embodiments of the present invention mainly includes the following two parts:

[0038] The part of constructing the semantic point cloud map, whose working architecture is asFigure 2 As shown, including steps S1 to S3 described above;

[0039] A vector map construction part, whose working architecture is as Figure 3 shown, including steps S4 to S6 described above.

[0040] In some preferred embodiments, for S1 above, after performing IPM transformation on multiple overhead images and then stitching them, an aerial view is obtained, and the following operations can further be included:

[0041] Obtain overhead RGB image data facing the ground through multiple fisheye cameras;

[0042] Using IPM technology, reverse-map the RGB image data into the ground space, and obtain the projection transformation matrix from the side view plane to the overhead view plane through calibration to obtain a perspective-transformed picture;

[0043] According to the relative positional relationship between the fisheye cameras, stitch the pictures after inverse perspective transformation into an aerial view.

[0044] In some preferred embodiments, for S2 above, perform semantic segmentation on the aerial view, extract the semantic elements therein, and convert the semantic elements into point cloud data, and the following operations can further be included:

[0045] Pass the aerial view through a pre-trained neural network semantic segmentation model to perform semantic segmentation on the aerial view, extract semantic features, obtain semantic elements, and convert the semantic elements into point cloud data; among them, the semantic elements include: lane lines, arrows, parking space lines, etc.

[0046] In some preferred embodiments, for S3 above, provide IMU (Inertial Measurement Unit) and vehicle wheel speedometer data, combine the IMU (Inertial Measurement Unit) and vehicle wheel speedometer data with the point cloud data to construct a local map, and use loop detection to correct the cumulative error of the local map to obtain a semantic point cloud map, and the following operations can further be included:

[0047] Take the measurement values of the IMU (Inertial Measurement Unit) and the vehicle wheel speedometer as odometry data;

[0048] According to the odometry data, obtain the real-time global coordinates of the vehicle, continuously transform the point cloud data to the global coordinates, and construct a local map;

[0049] Perform loop detection on the local map to correct the cumulative error of the local map to obtain a semantic point cloud map.

[0050] In some preferred embodiments, in step S4, the semantic point cloud map is preprocessed, and then the point cloud data in the preprocessed semantic point cloud map is clustered according to the similarity semantic point cloud to obtain the clustered point cloud data. It may further include the following operations:

[0051] Filter the noise, outliers, and regions with irregular smooth data density in the semantic point cloud map according to the coordinate values of the points in the point cloud, remove the points within a specific range, and obtain the point cloud space;

[0052] Divide the point cloud space into multiple three-dimensional voxels, where all the points within each voxel are approximated by the centroid or center point of all the points within the voxel to complete downsampling and obtain the preprocessed semantic point cloud map;

[0053] Estimate the normal vector of each point in the point cloud data in the preprocessed semantic point cloud map to obtain the surface orientation and shape features; further, based on the angle between the spatial distance and the normal vector between points, perform point cloud clustering to obtain the clustered point cloud data.

[0054] In some preferred embodiments, in step S5, based on the semantic point cloud map and combined with the clustered point cloud data, construct the boundaries on both sides of the road, generate the lane centerline and the boundary line, and obtain the vector map. It may further include the following operations:

[0055] Based on the semantic point cloud map, equally divide the centerline therein according to the number of points, and then, taking each section of the centerline as a reference, calculate the distance from each point on the centerline to the clustered obstacles on both sides in the clustered point cloud data to construct the boundaries on both sides of the road;

[0056] The human driving trajectory line is generated based on the actually collected real underground garage driving data, and the lane centerline is regenerated taking the segmented human driving trajectory line as a reference; taking each section of the lane centerline as a reference, generate the lane boundary line with a fixed lane width to obtain the vector map.

[0057] In some preferred embodiments, in step S6, perform curve optimization on the lane centerline and the boundary line in the vector map. It may further include the following operations:

[0058] Adopt the Bezier curve technique, and by adjusting the positions of the control points, perform fine curve fitting on the initially generated lane centerline and boundary line to generate a curve that not only fits the original data points but also has high smoothness, improving the accuracy and smoothness of the lane centerline and the boundary line;

[0059] Based on the curve fitting result, introduce the quadratic programming (QP) method to perform curve optimization on the obtained lane centerline, improving the smoothness of the lane centerline near the right-angle bend.

[0060] Based on the same inventive concept, an embodiment of the present invention further provides a basement scenario automated vector map construction system.

[0061] Specifically, as Figure 4 shown, the basement scenario automated vector map construction system provided by this embodiment may include the following modules:

[0062] Semantic point cloud map construction module, which is used to perform IPM transformation on multiple ground-facing overhead images and then splice them to obtain an aerial view; perform semantic segmentation on the aerial view, extract the semantic elements therein, and convert the semantic elements into point cloud data; provide IMU (Inertial Measurement Unit) and vehicle wheel speedometer data, combine the IMU (Inertial Measurement Unit) and vehicle wheel speedometer data with the point cloud data to construct a local map, and use loop detection to correct the cumulative error of the local map to obtain a semantic point cloud map;

[0063] Vector map construction module, which is used to preprocess the semantic point cloud map, and then cluster the point cloud data in the preprocessed semantic point cloud map according to the set similarity semantic point cloud to obtain clustered point cloud data; based on the semantic point cloud map, construct the boundaries on both sides of the road, and generate lane centerlines and boundary lines in the clustered point cloud data to obtain a vector map; embed parking space information in the vector map, and perform curve optimization on the lane centerlines and boundary lines in the vector map to construct the basement scenario automated vector map.

[0064] In some preferred embodiments, the above-mentioned semantic point cloud map construction module may further include the following sub-modules:

[0065] IPM (Inverse Perspective Mapping) module: Its function is to generate four top views through IPM transformation by four fisheye cameras, and further splice them into a whole picture according to the relative position relationship between the four cameras as the image input of the entire system. Further, the work content of this module includes:

[0066] a) Receive camera data: Use four fisheye cameras to take pictures directly downward at the ground and transmit the RGB image data in real time.

[0067] b) Image mapping: Use IPM technology to reverse-map the RGB image into the ground space, and obtain the projection transformation matrix from the side view plane to the top view plane through calibration.

[0068] c) Image stitching: According to the installation positions of the four fisheye cameras, obtain the relative installation positions of the four cameras, and splice the four pictures after inverse perspective transformation into an aerial view and transmit it to the semantic segmentation module. In some preferred embodiments, the fisheye cameras can be installed on the vehicle body.

[0069] Semantic Segmentation Module: Its function is to receive the IPM bird's-eye view, segment semantic information from the image data using a neural network, extract semantic elements such as lane lines and arrows, and convert the semantic elements into point cloud data.

[0070] Local Mapping Module: Its function is to use the measurement values of the IMU (Inertial Measurement Unit) and the vehicle wheel speedometer as odometry data, obtain the real-time global coordinates of the vehicle based on the odometry data, and construct a local map by continuously transforming the point cloud data features learned from the neural network into the global coordinates.

[0071] Loop Detection Module: Its function is to correct the cumulative error of the local map.

[0072] In some preferred embodiments, the above vector map construction module may further include the following sub-modules:

[0073] Point Cloud Filtering Module: Its function is to remove noise, outliers, smooth irregular regions of data density, and perform downsampling. Further, the work content of this module includes:

[0074] a) Pass-through Filtering: Filter according to the coordinate values of the points in the point cloud to remove points within a specific range.

[0075] b) Voxel Grid Filtering: Divide the point cloud space into multiple three-dimensional voxels, and approximate all points within each voxel with the centroid or center point of all points within the voxel to achieve downsampling.

[0076] Point Cloud Clustering Module: Its function is to divide the point cloud data into multiple clusters according to a certain similarity semantic point cloud to achieve the purpose of clustering. Further, the work content of this module includes:

[0077] a) Euclidean Clustering Based on Normal Vectors: Estimate the normal vector of each point in the point cloud data in the preprocessed semantic point cloud map to obtain the surface orientation and shape features; further, based on the spatial distance and the angle between the normal vectors of points, use the difference between the Euclidean distance and the normal vector to achieve point cloud clustering.

[0078] Boundary Construction Module: Its function is to extract and construct the boundaries on both sides of the road from the original data. Further, the work content of this module includes:

[0079] a) Centerline Segmentation: Segment the centerline at equal distances according to the number of points.

[0080] b) Generate the Distance from the Centerline to Both Sides: Based on each segment of the centerline, calculate the distance from each point on the centerline to the clustering obstacles on both sides to construct the boundary.

[0081] Centerline and boundary line generation module: Its function is to generate the lane centerline and the lane boundaries on both sides. Further, the work content of this module includes:

[0082] a) Generate the centerline: The human-driven trajectory line is generated based on the actual collected real underground garage driving data, and the boundary centerline is regenerated with the segmented human-driven trajectory line as the benchmark.

[0083] b) Generate the boundaries on both sides: With each segment of the centerline as the benchmark, the boundaries on both sides are generated with a fixed lane width length.

[0084] Parking space embedding module: Its function is to embed parking space information in the vector map.

[0085] Curve optimization module: Its function is to improve the accuracy and smoothness of the centerline and boundary lines. Further, the work content of this module includes:

[0086] a) Curve fitting: Using the Bezier curve technology, the initially generated centerline and boundary lines are finely fitted. By adjusting the positions of the control points, curves that fit the original data points and have high smoothness are generated, improving the accuracy and smoothness of the centerline and boundary lines.

[0087] b) Curve optimization: On the basis of curve fitting, the quadratic programming (QP) method is further introduced to optimize the centerline curve, improving the smoothness of the centerline near right-angle turns.

[0088] An embodiment of the present invention also provides a computer terminal, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it can be used to execute the method of any one of the above embodiments of the present invention, or, run the system of any one of the above embodiments of the present invention.

[0089] Optionally, a memory for storing programs; the memory may include volatile memory (e.g., random-access memory, such as static random-access memory (SRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDR SDRAM), etc.); the memory may also include non-volatile memory, such as flash memory. The memory is used to store computer programs (such as application programs and functional modules for implementing the above method), computer instructions, etc. The above computer programs, computer instructions, etc. may be partitioned and stored in one or more memories. And the above computer programs, computer instructions, data, etc. may be called by the processor.

[0090] The above computer programs, computer instructions, etc. may be partitioned and stored in one or more memories. And the above computer programs, computer instructions, data, etc. may be called by the processor.

[0091] A processor for executing the computer programs stored in the memory to implement each step in the method or each module in the system according to the above embodiments. For specific details, please refer to the relevant descriptions in the foregoing method and system embodiments.

[0092] The processor and the memory may be of an independent structure or an integrated structure integrated together. When the processor and the memory are of an independent structure, the memory and the processor may be coupled and connected through a bus.

[0093] An embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it can be used to execute the method according to any one of the above embodiments of the present invention, or to run the system according to any one of the above embodiments of the present invention.

[0094] Among them, the computer-readable medium includes computer storage media and communication media, where the communication media includes any medium facilitating the transmission of a computer program from one place to another. The storage media can be any available medium accessible by a general-purpose or special-purpose computer. An exemplary storage media is coupled to the processor so that the processor can read information from and write information to the storage media. Of course, the storage media can also be a component of the processor. The processor and the storage media can be located in the ASIC. Additionally, the ASIC can be located in the user equipment. Of course, the processor and the storage media can also exist as discrete components in the communication device.

[0095] The basement scenario automated vector map construction method, system, terminal, and medium provided by the above embodiments of the present invention integrate the pure vision semantic mapping technology and simultaneously adopt the fusion of point cloud filtering, point cloud clustering, and curve optimization strategies, achieving fast and accurate parsing of the collected images, extracting rich semantic information, and constructing a high-precision and high-coverage automated semantic point cloud map. This comprehensive technical solution not only solves the technical problem of low precision in the traditional basement map construction process but also can efficiently and accurately complete the construction of the vector map in a complex environment, providing a more reliable and efficient map supply solution for cutting-edge fields such as intelligent driving and robot navigation, and promoting the further development of related technologies.

[0096] Matters not described in detail in the above embodiments of the present invention are all well-known technologies in the art.

[0097] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various deformations or modifications within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. An automated vector map construction method for basement scenarios, characterized in that Including: Performing IPM transformation on multiple ground-facing overhead images and then stitching them to obtain an aerial view; Performing semantic segmentation on the aerial view, extracting semantic elements therein, and converting the semantic elements into point cloud data; Providing inertial measurement unit and vehicle wheel speedometer data, combining the inertial measurement unit and vehicle wheel speedometer data with the point cloud data to construct a local map, and using loop detection to correct the cumulative error of the local map to obtain a semantic point cloud map; Preprocessing the semantic point cloud map, and then clustering the point cloud data in the preprocessed semantic point cloud map according to the set similarity semantic point cloud to obtain clustered point cloud data; Based on the semantic point cloud map and combining with the clustered point cloud data, constructing the boundaries on both sides of the road, and generating lane centerlines and boundary lines to obtain a vector map; Embedding parking space information in the vector map, and performing curve optimization on the lane centerlines and boundary lines in the vector map to construct an automated vector map of the basement scenario.

2. The method for automatically constructing a vector map of a basement scenario according to claim 1, wherein The performing IPM transformation on multiple overhead images and then stitching them to obtain an aerial view includes: Obtaining overhead RGB image data facing the ground through multiple fisheye cameras; Using IPM technology to reversely map the RGB image data into the ground space, and obtaining the projection transformation matrix from the side view plane to the overhead view plane through calibration to obtain the image after perspective transformation; According to the relative position relationship between the fisheye cameras, stitching the images after inverse perspective transformation into an aerial view.

3. The method for automatically constructing a vector map of a basement scenario according to claim 1, wherein The performing semantic segmentation on the aerial view, extracting semantic elements therein, and converting the semantic elements into point cloud data includes: Performing semantic segmentation on the aerial view through a pre-trained neural network semantic segmentation model, extracting semantic features in the aerial view to obtain semantic elements, and converting the semantic elements into point cloud data; wherein, the semantic elements include: lane lines, arrows, and parking space lines.

4. The method for constructing an automated vector map of a basement scenario according to claim 1, wherein The providing inertial measurement unit and vehicle wheel speedometer data, combining the inertial measurement unit and vehicle wheel speedometer data with the point cloud data to construct a local map, and using loop detection to correct the cumulative error of the local map to obtain a semantic point cloud map includes: Taking the measurement values of the inertial measurement unit and the vehicle wheel speedometer as mileage data, obtaining the real-time global coordinates of the vehicle according to the mileage data, continuously transforming the point cloud data into the global coordinates, and constructing a local map; Performing loop detection on the local map to correct the cumulative error of the local map to obtain a semantic point cloud map.

5. The method for automatically constructing a vector map of a basement scenario according to claim 1, wherein The preprocessing the semantic point cloud map, and then clustering the point cloud data in the preprocessed semantic point cloud map according to the similarity semantic point cloud to obtain clustered point cloud data includes: Filtering the noise, outliers, and regions with irregular smooth data density in the semantic point cloud map according to the coordinate values of the points in the point cloud, removing points within a specific range to obtain a point cloud space; Divide the point cloud space into multiple three-dimensional voxels, where all points within each voxel are approximately represented by the centroid or center point of all points within the voxel, complete downsampling, and obtain a preprocessed semantic point cloud map; Estimate the normal vector of each point in the point cloud data of the preprocessed semantic point cloud map to obtain the surface orientation and shape features; based on the spatial distance and the angle between the normal vectors of points, achieve point cloud clustering to obtain clustered point cloud data.

6. The method for constructing an automated vector map of a basement scenario according to claim 1, wherein, Based on the semantic point cloud map and in combination with the clustered point cloud data, construct the boundaries on both sides of the road, and generate the lane center line and boundary lines to obtain a vector map, including: Based on the semantic point cloud map, equally divide the center line therein according to the number of points, and then, taking each section of the center line as a reference, calculate the distance from each point on the center line to the clustered obstacles on both sides in the clustered point cloud data, and construct the boundaries on both sides of the road; The human-driven trajectory line is generated based on the actually collected real underground garage driving data, and the lane center line is regenerated taking the segmented human-driven trajectory line as a reference; taking each section of the lane center line as a reference, generate lane boundary lines with a fixed lane width to obtain a vector map.

7. The method for constructing an automated vector map for a basement scenario according to claim 1, wherein The curve optimization of the lane center line and boundary lines in the vector map may further include the following operations: Adopt the Bezier curve technique to perform curve fitting on the initially generated lane center line and boundary lines by adjusting the positions of the control points; On the basis of the result of the curve fitting, adopt the quadratic programming method to perform curve optimization on the obtained lane center line.

8. An automated vector map construction system for basement scenarios, characterized in that Including: A semantic point cloud map construction module, which is used to perform IPM transformation on multiple ground-facing overhead images and then splice them to obtain an aerial view; perform semantic segmentation on the aerial view, extract the semantic elements therein, and convert the semantic elements into point cloud data; provide inertial measurement unit and vehicle wheel speedometer data, combine the inertial measurement unit and vehicle wheel speedometer data with the point cloud data to construct a local map, and use loop detection to correct the cumulative error of the local map to obtain a semantic point cloud map; A vector map construction module, which is used to preprocess the semantic point cloud map, and then cluster the point cloud data in the preprocessed semantic point cloud map according to the set similarity semantic point cloud to obtain clustered point cloud data; Based on the semantic point cloud map, construct the boundaries on both sides of the road, and generate the lane center line and boundary lines in the clustered point cloud data to obtain a vector map; embed parking space information in the vector map, and perform curve optimization on the lane center line and boundary lines in the vector map to construct an automated vector map of the basement scenario.

9. A computer terminal, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that When the processor executes the computer program, it can be used to execute the method described in any one of claims 1-7, or, run the system described in claim 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it can be used to execute the method described in any one of claims 1-7, or, run the system described in claim 8.

Images