Graphics-free lane level positioning method based on vehicle and road cloud cross-domain complex system

Abstract

The invention relates to the technical field of intelligent traffic, in particular to a non-graphical lane level positioning method based on a vehicle and road cloud cross-domain complex system, and the method comprises the steps: configuring vehicle end equipment, a road side RSU and a vehicle and road cloud cooperative interaction module; acquiring a road image based on vehicle end equipment, and extracting lane features; acquiring a lane information table according to the roadside RSU, uploading the lane information table to the vehicle-road cloud collaboration platform to verify the lane information table, and broadcasting the verified lane information table to the vehicle end; matching the extracted lane features with a lane information table received by the vehicle end to position the lane of the vehicle; and outputting a positioning result of the lane of the vehicle. According to the method, the core pain points of high cost, large scene limitation and low positioning success rate in the prior art are effectively solved, and key technical support is provided for large-scale landing of high-grade intelligent driving.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent transportation, in particular to a lane-level positioning method without map based on a vehicle-road cloud cross-domain complex system. BACKGROUND

[0002] In current intelligent driving technology, lane-level positioning is one of the core technologies to ensure vehicle driving safety and path planning. Existing lane-level positioning schemes are mainly divided into three categories:

[0003] A positioning scheme based on a high-precision map: road features are collected by a vehicle-mounted sensor (camera, laser radar), and are matched with lane lines, kerbs and other static features in a pre-stored high-precision map to achieve positioning. This scheme needs to rely on a high-precision map with an accuracy of within 10 cm, and the map needs to cover the entire operating section.

[0004] A positioning scheme based on GPS: satellite signals are received by GPS, and the positioning drift is corrected by an inertial navigation module. This scheme is suitable for open road scenarios, but its performance decreases in shaded areas.

[0005] A positioning scheme based on traditional V2X: lane number information is broadcast by a roadside unit (RSU), and the vehicle end only determines the lane by "lane number matching + simple distance calculation", without involving line type sequence comparison.

[0006] In the above schemes, the scheme based on a high-precision map is currently mainstream, but the map needs to be updated regularly to adapt to road construction, marking replacement and other dynamic changes. The scheme based on GPS is significantly affected by environmental shading. The traditional V2X scheme has insufficient matching reliability in multi-lane and complex line type scenarios due to the simple comparison logic. SUMMARY

[0007] The purpose of the present application is to provide a lane-level positioning method without map based on a vehicle-road cloud cross-domain complex system to solve the problems of high-precision map dependence, GPS failure scenario positioning and low success rate of traditional V2X positioning.

[0008] The present application provides the following scheme:

[0009] A lane-level positioning method without map based on a vehicle-road cloud cross-domain complex system, comprising:

[0010] configuring a vehicle end device, a roadside RSU and a vehicle-road cloud cooperative interaction module;

[0011] collecting road images based on the vehicle end device and extracting lane features;

[0012] obtaining a lane information table based on the roadside RSU, uploading the lane information table to a vehicle-road cloud cooperative platform, checking the lane information table, and broadcasting the lane information table that passes the check to the vehicle end.

[0013] Matching the extracted lane features with the lane information table received by the vehicle end to locate the lane of the vehicle;

[0014] Outputting the locating result of the lane of the vehicle.

[0015] Further, the vehicle end device configuration process is as follows: a high-definition camera is installed on the inner side of the front windshield of the vehicle, the installation height h should be between 1.5 meters and 2.0 meters, and the pitch angle θ should be between 15° and 20°, after installation, using Zhang Zhengyou's calibration method, using a chessboard calibration board with known size, by collecting multiple images of the calibration board in different poses, the internal parameter matrix K of the camera is calculated, at the same time, according to the installation height and angle, the external parameter matrix of the camera relative to the vehicle coordinate system is determined, the external parameter matrix includes rotation matrix R and translation vector t.

[0016] Further, the road side RSU configuration process is as follows:

[0017] At the critical position where the GPS signal is about to fail, deploy RSU with a spacing not exceeding 500 meters, each RSU should integrate laser radar, high-definition camera and V2X communication module, all RSUs need to access the unified vehicle-road cloud collaborative management platform through the network, and register their geographic location ID and coverage range in the platform, forming a collaborative sensing network;

[0018] The configuration process of the vehicle-road cloud collaborative interaction module is as follows:

[0019] Deploy the collaborative interaction module on the cloud server, the module needs to preset the lane information verification rule library, the rules include "lane number mutation detection" and "line type sequence logic conflict detection", at the same time, the platform needs to establish synchronization with the high-precision clock server to ensure that all devices are time-synchronized, and to issue communication certificates for vehicle end and roadside devices.

[0020] Further, the vehicle end camera continuously collects RGB road images at a rate of 30 frames per second, then, each frame of image is preprocessed, the preprocessing process is as follows:

[0021] Using weighted average method for gray scale: Gray=0.299R+0.587G+0.114B, and using 5x5 Gaussian kernel with standard deviation σ=1.2 to filter the gray scale image;

[0022] Based on the internal and external parameter matrix, calculate the inverse perspective transformation matrix H from the image plane to the ground plane, multiply the preprocessed image by the matrix H to transform the image from the front view to the bird's eye view with the vehicle rear axle center as the origin.

[0023] Furthermore, edge detection is performed on the overhead view using the Canny operator, and probabilistic Hough transform is used to detect straight line segments. The parameters are set as follows: distance resolution = 1 pixel, angle resolution = 1 degree, accumulator threshold = 150. The detected line segments are clustered and extended to fit complete left and right lane lines. For each fitted line, the line type is identified by analyzing its pixel continuity: continuous pixel segments with a length of 5-6 meters are identified as solid lines, and those with a gap of 3-4 meters are identified as dashed lines. Finally, the equations and line type sequences of the left and right lane lines of the lane where the vehicle is located are output.

[0024] Furthermore, the RSU's lidar and camera synchronously collect data. The lidar point cloud is used to accurately calculate the three-dimensional spatial position of each lane line, and the camera image is used to assist in identifying the lane line color and type. The lane color is yellow and white, and the lane type is solid and dashed. Through spatiotemporal synchronization and coordinate transformation, the data from the two sensors are fused to generate a lane information table containing the number of lanes, the geometric coordinates of each lane line, and the line type order.

[0025] Furthermore, the RSU will upload the generated lane information table to the vehicle-road-cloud collaborative platform and verify the generated lane information table;

[0026] For lane information that passes verification, the RSU encapsulates it into a standardized V2X message, and the broadcast frequency is dynamically adjusted according to real-time traffic flow density.

[0027] Furthermore, the vehicle parses the received roadside broadcast message, extracts the "number of lanes" field, and compares it with the number of lanes fitted by the vehicle itself. If the numbers match, the next step of fine comparison is performed; if they do not match, the vehicle image processing flow is triggered to be re-executed. The maximum number of re-executions is 3. If they still do not match, the positioning fails and a "perception anomaly" is reported.

[0028] Furthermore, in the vehicle-view coordinate system, with the center projection point of the rear axle of the vehicle as the origin O and the vehicle's forward direction as the X-axis, points PL1, PL2 on the left lane line and points PR1, PR2 on the right lane line are taken at distances of D1 = 10 meters and D2 = 30 meters from the origin, respectively. Then the left lane vector... PL2−PL1, Right Lane Vector PR2−PR1, the roadside broadcast information directly includes the standard vectors of each lane calculated in its coordinate system.

[0029] Furthermore, the vehicle calculates the cosine similarity between its own left and right lane vectors and the standard vectors of each lane broadcast on the roadside, and sets a similarity threshold. If the cosine similarity between the standard vector of the lane and the lane vector of the vehicle is greater than the similarity threshold, the vehicle is determined to be in that lane, and the lane ID is output.

[0030] The above solution achieves the following beneficial technical effects:

[0031] This invention constructs a mapless positioning system that integrates vehicle, road, and cloud technologies, completely eliminating reliance on high-precision maps and GPS signals, significantly reducing system deployment and maintenance costs. The method utilizes vehicle-side visual perception to extract lane features and combines them with authoritative lane information provided by roadside equipment and verified in the cloud. Through innovative multi-level comparison logic, it achieves stable and accurate lane-level positioning in complex scenarios such as tunnels and overpasses. The system possesses high reliability, strong adaptability, and excellent real-time performance, effectively addressing the core pain points of existing technologies, such as high cost, limited application scenarios, and low positioning success rate, providing crucial technical support for the large-scale deployment of advanced autonomous driving. Attached Figure Description

[0032] Figure 1 This is a flowchart of a graphless lane-level localization method based on a complex cross-domain vehicle-road-cloud system provided by one or more embodiments of the present invention.

[0033] Figure 2 This is a flowchart of the lane positioning method of the present invention. Detailed Implementation

[0034] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Specifically, the mapless lane-level positioning method based on the vehicle-road-cloud cross-domain complex system described in this embodiment constructs a high-precision, high-reliability lane-level positioning system that does not rely on high-precision maps and GPS signals through a three-level architecture of vehicle-side visual perception, roadside information provision, and cloud-based collaborative verification. This method is particularly suitable for scenarios where traditional positioning methods fail, such as tunnels, overpasses, and urban canyons.

[0036] Please see Figure 1 The diagram shown is a flowchart of a graphless lane-level localization method based on a complex cross-domain vehicle-road-cloud system according to this embodiment. The method includes:

[0037] Step S101: Configure the vehicle-side equipment, roadside RSU and vehicle-road-cloud collaborative interaction module. The vehicle-road-cloud collaborative interaction module is a software platform deployed on the cloud or edge server.

[0038] Specifically, the configuration process for the vehicle-side equipment is as follows:

[0039] A high-definition camera is installed on the inside of the vehicle's windshield at a height h between 1.5 and 2.0 meters, with a pitch angle θ between 15° and 20°. After installation, the camera is calibrated using the Zhang Zhengyou calibration method with a known-sized checkerboard calibration board. Multiple images of the calibration board in different poses are acquired, and the camera's internal parameter matrix K and distortion coefficients are calculated. The internal parameter matrix K is in the following form:

[0040] ;

[0041] Where fx and fy are the focal length pixel values, (cx and cy) are the principal point coordinates, and the external parameter matrix of the camera relative to the vehicle coordinate system is determined according to the installation height and angle. The external parameter matrix includes the rotation matrix R and the translation vector t.

[0042] The roadside RSU configuration process is as follows:

[0043] In critical locations such as tunnels and overpass approach bridges where GPS signals are about to fail, RSUs should be deployed at intervals of no more than 500 meters. Each RSU should integrate a lidar, a high-definition camera, and a V2X communication module (supporting IEEE 802.11p or C-V2X protocol). All RSUs must be connected to a unified vehicle-road-cloud collaborative management platform through the network and register their geographical location ID and coverage area in the platform to form a collaborative perception network.

[0044] The configuration process of the vehicle-road-cloud collaborative interaction module is as follows:

[0045] Deploy a collaborative interaction module on a cloud server. This module needs to pre-set a lane information verification rule base. The rules include, but are not limited to, "lane number change detection" and "line order logic conflict detection". At the same time, the platform needs to establish synchronization with a high-precision clock server to ensure that the time of all devices is synchronized and to issue communication certificates to vehicle-side and roadside devices.

[0046] Specifically, the process of "determining the extrinsic parameter matrix of the camera relative to the vehicle coordinate system based on the installation height and angle" is as follows:

[0047] First, a calibration plate of known size is precisely placed in front of the vehicle, ensuring that it is simultaneously within the field of view of the vehicle-mounted camera and another reference sensor that is precisely aligned with the vehicle coordinate system. After synchronously acquiring camera images and reference sensor data, the pixel coordinates of the calibration plate corner points in the camera image coordinate system are obtained through image recognition. At the same time, the three-dimensional coordinates of these corner points in the vehicle coordinate system are obtained through the reference sensor. Using these two precisely corresponding sets of coordinate points, the cv2.solvePnP() function in the OpenCV library is directly called to solve for the rotation matrix R and translation vector t required to transform the points from the vehicle coordinate system to the camera coordinate system, thus obtaining the complete external parameter matrix [R|t].

[0048] Specifically, through standardized equipment installation, calibration, and networking processes, a unified spatial benchmark and communication foundation were established for the system, ensuring accurate alignment and reliable interaction of data from vehicle, road, and cloud units, thus providing the necessary prerequisites for high-precision positioning.

[0049] Please continue reading. Figure 1 As shown, the graph-free lane-level localization method based on the vehicle-road-cloud cross-domain complex system further includes:

[0050] Step S102: Collect road images based on vehicle-mounted devices and extract lane features.

[0051] Specifically, the vehicle-mounted camera continuously acquires RGB road images at a rate of 30 frames per second. Subsequently, each frame of the image undergoes preprocessing, and the preprocessing process is as follows:

[0052] The grayscale image was converted using a weighted average method: Gray = 0.299R + 0.587G + 0.114B, and the grayscale image was filtered using a 5×5 Gaussian kernel with a standard deviation σ = 1.2.

[0053] The inverse perspective transformation matrix H from the image plane to the ground plane is calculated based on the internal and external parameter matrices. The preprocessed image is multiplied by this matrix H to transform the image from a frontal view to a top-down view with the center of the vehicle's rear axle as the origin.

[0054] Edge detection is performed using the Canny operator on the overhead view image, and probabilistic Hough transform is used to detect straight line segments. The parameters are set as follows: distance resolution = 1 pixel, angle resolution = 1 degree, accumulator threshold = 150. The detected line segments are clustered and extended to fit complete left and right lane lines. For each fitted line, the line type is identified by analyzing its pixel continuity: continuous pixel segments with a length of 5-6 meters are identified as solid lines, and those with a gap of 3-4 meters are identified as dashed lines. Finally, the equations and line type sequences of the left and right lane lines of the lane where the vehicle is located are output.

[0055] Specifically, the process of "calculating the inverse perspective transformation matrix H from the image plane to the ground plane based on the internal and external parameter matrices" is as follows: Based on the internal parameter matrix K and the external parameter matrix, the ground plane constraint, i.e., the height coordinate Z=0 of the ground point, is substituted into the standard camera imaging model, i.e., s[u, v, 1]T= K[R|t] [X, Y, 0, 1]T). By eliminating the scale factor s and performing matrix operations, the 3x3 homography matrix H connecting the image pixel coordinates (u,v) and the ground point coordinates (X,Y) can be directly derived.

[0056] Specifically, by using image processing and inverse perspective transformation technology, the tilted view images captured by the vehicle-mounted camera are converted into standard overhead views, effectively eliminating perspective distortion. This allows for the accurate and stable extraction of key features such as the number, location, and line type of lane lines.

[0057] Please continue reading. Figure 1 As shown, the graph-free lane-level localization method based on the vehicle-road-cloud cross-domain complex system further includes:

[0058] Step S103: Obtain the lane information table based on the roadside RSU and upload it to the vehicle-road-cloud collaborative platform to verify the lane information table, and broadcast the verified lane information table to the vehicle.

[0059] Specifically, the RSU's lidar and camera synchronously collect data. The lidar point cloud is used to accurately calculate the three-dimensional spatial position of each lane line, and the camera image is used to assist in identifying the lane line color and type. The lane color is yellow and white, and the lane type is solid and dashed. Through spatiotemporal synchronization and coordinate transformation, the data from the two sensors are fused to generate a lane information table containing the number of lanes, the geometric coordinates of each lane line, and the line type order.

[0060] The RSU will upload the generated lane information table to the vehicle-road-cloud collaborative platform and verify the generated lane information table.

[0061] For lane information that passes verification, the RSU encapsulates it into a standardized V2X message, and the broadcast frequency is dynamically adjusted according to real-time traffic flow density.

[0062] Specifically, in this embodiment, the verification process for the lane information table can be as follows: compare the current RSU information with the information of its upstream RSU to ensure that the change in the number of lanes conforms to the road design. For example, if the current RSU covers an area with 3 lanes, while its directly upstream RSU has 2 lanes and no ramp information, it is determined to be an abnormal change in the number of lanes. Then, check whether the alignment sequence conforms to traffic rules, such as the logic of combining solid and dashed lines, and compare it with historical data of the same period in the past 24 hours. If there is an inconsistency and no road construction report is found, it is determined to be abnormal. After all verification rules are passed, the information is marked as "verified". Otherwise, it will be discarded and the RSU device self-test will be triggered. This embodiment does not specifically limit the above process. Those skilled in the art can set it freely according to their needs.

[0063] Specifically, the process of dynamically adjusting the broadcast frequency according to the real-time traffic flow density can be as follows: when the platform detects that the vehicle density of a road segment is higher than 30 vehicles / km, a high-frequency broadcast of 10Hz is used; when the density is lower than 10 vehicles / km, a low-frequency broadcast of 1Hz is used. In this embodiment, the above settings are not specifically limited, and those skilled in the art can set them freely according to their needs.

[0064] Specifically, through the fusion perception of roadside equipment and the collaborative verification mechanism of the cloud platform, the vehicle-side is provided with authoritative lane information that has undergone multiple verifications. This avoids the spread of erroneous data caused by a single device failure or misidentification from the source, thus ensuring the reliability of the baseline information.

[0065] Please continue reading. Figure 1 As shown, the graph-free lane-level localization method based on the vehicle-road-cloud cross-domain complex system further includes:

[0066] Step S104: Match the extracted lane features with the lane information table received by the vehicle to locate the vehicle's lane.

[0067] Please see Figure 2 As shown, the lane positioning method includes:

[0068] Step S201: Perform initial matching of the number of lanes.

[0069] Specifically, the vehicle parses the roadside broadcast message it receives, extracts the "number of lanes" field, and compares it with the number of lanes fitted by the vehicle itself. If the numbers match, the next step of fine comparison is performed; if they do not match, the vehicle's image processing flow is triggered to be re-executed. The maximum number of re-executions is 3. If they still do not match, the positioning fails and a "perception anomaly" is reported.

[0070] Please continue reading. Figure 2 As shown, the lane positioning method further includes:

[0071] Step S202: Construct the projection vector.

[0072] Specifically, in the vehicle-view coordinate system, with the center projection point of the rear axle of the vehicle as the origin O and the vehicle's forward direction as the X-axis, points PL1, PL2 on the left lane line and points PR1, PR2 on the right lane line are taken at distances of D1=10 meters and D2=30 meters from the origin, respectively. Then, the vector of the left lane... PL2−PL1, Right Lane Vector PR2−PR1, the roadside broadcast information directly includes the standard vectors of each lane calculated in its coordinate system.

[0073] Please continue reading. Figure 2 As shown, the lane positioning method further includes:

[0074] Step S203: Calculate vector similarity and determine the vehicle's lane.

[0075] Specifically, the vehicle calculates the cosine similarity between its own left and right lane vectors and the standard vectors of each lane broadcast on the roadside, and sets a similarity threshold s. If the cosine similarity between the standard vector of the lane and the lane vector of the vehicle is greater than the similarity threshold, the vehicle is determined to be in that lane, and the lane ID is output.

[0076] The lane determination process for vehicles is adjusted based on vehicle speed and historical matching success rate:

[0077] The vehicle speed v0 is compared with the preset speed v1, and the similarity threshold is adjusted according to the comparison result. If v0≤v1, the similarity threshold is not adjusted. If v0>v1, the similarity threshold is adjusted to s1, s1=s+η×tanh[(v0-v1) / v1], where η is the preset correction coefficient.

[0078] The historical matching success rate p0 is compared with the preset success rate p1, and the preset correction coefficient is adjusted according to the comparison result. If p0 < p1, the preset correction coefficient is adjusted to η1, where η1 = 0.5 × η. If p0 ≥ p1, the preset correction coefficient is not adjusted.

[0079] Specifically, the historical matching success rate refers to the proportion of times the vehicle-side lane vector and the roadside standard vector successfully match within a recent period (such as the past 10 matches) out of the total number of matches.

[0080] Specifically, in this embodiment, the preset speed is 80km / h, the preset success rate is 80%, the preset correction coefficient is 0.06, and the similarity threshold is 0.85.

[0081] Specifically, through multi-level comparison logic of "preliminary matching of lane numbers" and "projection vector similarity calculation", the ability to distinguish and locate in complex lane alignment scenarios is significantly improved, effectively overcoming the shortcomings of traditional solutions that are easily confused due to simple logic.

[0082] Please continue reading. Figure 1 As shown, the graph-free lane-level localization method based on the vehicle-road-cloud cross-domain complex system further includes:

[0083] Step S105: Output the vehicle's lane location result.

[0084] Specifically, the lane ID is output to the vehicle control system.

[0085] Specifically, the high-precision lane-level positioning results are output in real time in a standardized format to directly serve intelligent driving decisions. At the same time, a systematic status monitoring and maintenance mechanism ensures that the positioning system can operate stably and reliably for a long time.

[0086] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is impossible to exhaustively list all the implementation methods here. All obvious variations or modifications derived from the technical solutions of the present invention are still within the protection scope of the present invention.

Claims

1. A graph-free lane-level localization method based on a complex cross-domain vehicle-road-cloud system, characterized in that, include: Configure vehicle-side equipment, roadside RSU, and vehicle-road-cloud collaborative interaction module; Road images are collected using vehicle-mounted devices, and lane features are extracted. The lane information table is obtained from the roadside RSU and uploaded to the vehicle-road-cloud collaborative platform to verify the lane information table. The verified lane information table is then broadcast to the vehicle. The extracted lane features are matched with the lane information table received by the vehicle to locate the vehicle's lane. Output the vehicle's lane location result.

2. The mapless lane-level localization method based on a complex cross-domain vehicle-road-cloud system according to claim 1, characterized in that, The configuration process of the vehicle-side equipment is as follows: A high-definition camera is installed on the inside of the windshield at the front of the vehicle. The installation height h should be between 1.5 meters and 2.0 meters, and the pitch angle θ should be between 15° and 20°. After installation, the Zhang Zhengyou calibration method is used. Using a checkerboard calibration board of known size, multiple images of the calibration board in different poses are acquired to calculate the internal parameter matrix K of the camera. At the same time, based on the installation height and angle, the external parameter matrix of the camera relative to the vehicle coordinate system is determined. The external parameter matrix includes the rotation matrix R and the translation vector t.

3. The mapless lane-level localization method based on a vehicle-road-cloud cross-domain complex system according to claim 2, characterized in that, The roadside RSU configuration process is as follows: At critical locations where GPS signals are about to fail, RSUs should be deployed at intervals of no more than 500 meters. Each RSU should integrate a lidar, a high-definition camera, and a V2X communication module. All RSUs should be connected to a unified vehicle-road-cloud collaborative management platform through the network and register their geographic location ID and coverage area in the platform to form a collaborative perception network. The configuration process of the vehicle-road-cloud collaborative interaction module is as follows: Deploy a collaborative interaction module on a cloud server. This module needs to pre-set a lane information verification rule base, including "lane number change detection" and "line order logic conflict detection". At the same time, the platform needs to establish synchronization with a high-precision clock server to ensure that all devices are synchronized in time and issue communication certificates to vehicle-side and roadside devices.

4. The mapless lane-level localization method based on a complex cross-domain vehicle-road-cloud system according to claim 3, characterized in that, The vehicle-mounted camera continuously acquires RGB road images at a rate of 30 frames per second, and preprocesses each frame as follows: The grayscale image was converted using a weighted average method: Gray = 0.299R + 0.587G + 0.114B, and the grayscale image was filtered using a 5×5 Gaussian kernel with a standard deviation σ = 1.

2. The inverse perspective transformation matrix H from the image plane to the ground plane is calculated based on the internal and external parameter matrices. The preprocessed image is multiplied by this matrix H to transform the image from a frontal view to a top-down view with the center of the vehicle's rear axle as the origin.

5. The mapless lane-level localization method based on a complex cross-domain vehicle-road-cloud system according to claim 4, characterized in that, Edge detection is performed using the Canny operator on the overhead view image, and probabilistic Hough transform is used to detect straight line segments. The parameters are set as follows: distance resolution = 1 pixel, angle resolution = 1 degree, accumulator threshold = 150. The detected line segments are clustered and extended to fit complete left and right lane lines. For each fitted line, the line type is identified by analyzing its pixel continuity: continuous pixel segments with a length of 5-6 meters are identified as solid lines, and those with a gap of 3-4 meters are identified as dashed lines. Finally, the equations and line type sequences of the left and right lane lines of the lane where the vehicle is located are output.

6. The graph-free lane-level localization method based on a complex cross-domain vehicle-road-cloud system according to claim 5, characterized in that, The RSU's lidar and camera simultaneously collect data. The lidar point cloud is used to accurately calculate the three-dimensional spatial position of each lane line, while the camera images are used to assist in identifying the lane line color and type. The lane colors are yellow and white, and the lane types are solid and dashed. Through spatiotemporal synchronization and coordinate transformation, the data from the two sensors are fused to generate a lane information table that includes the number of lanes, the geometric coordinates of each lane line, and the line type order.

7. The mapless lane-level localization method based on a complex cross-domain vehicle-road-cloud system according to claim 6, characterized in that, The RSU will upload the generated lane information table to the vehicle-road-cloud collaborative platform and verify the generated lane information table. For lane information that passes verification, the RSU encapsulates it into a standardized V2X message, and the broadcast frequency is dynamically adjusted according to real-time traffic flow density.

8. The mapless lane-level localization method based on a complex cross-domain vehicle-road-cloud system according to claim 7, characterized in that, The vehicle parses the roadside broadcast message it receives, extracts the "number of lanes" field, and compares it with the number of lanes fitted by the vehicle itself. If the numbers match, the next step of fine comparison is performed; if they do not match, the vehicle's image processing flow is triggered to be re-executed. The maximum number of re-executions is 3. If they still do not match, the positioning fails and a "perception anomaly" is reported.

9. The mapless lane-level localization method based on a complex cross-domain vehicle-road-cloud system according to claim 8, characterized in that, In the vehicle-view coordinate system, with the center projection point of the rear axle of the vehicle as the origin O and the vehicle's forward direction as the X-axis, points PL1, PL2 on the left lane and points PR1, PR2 on the right lane are taken at distances of D1 = 10 meters and D2 = 30 meters from the origin, respectively. The left lane vector... PL2−PL1, Right Lane Vector PR2−PR1, the roadside broadcast information directly includes the standard vectors of each lane calculated in its coordinate system.

10. The mapless lane-level localization method based on a vehicle-road-cloud cross-domain complex system according to claim 9, characterized in that, The vehicle calculates the cosine similarity between its own left and right lane vectors and the standard vectors of each lane broadcast on the roadside, and sets a similarity threshold. If the cosine similarity between the standard vector of the lane and the lane vector of the vehicle is greater than the similarity threshold, the vehicle is determined to be in that lane, and the lane ID is output.

Images