A method and system for initializing vehicle positioning using road boundary lines as constraints

Abstract

This invention discloses a method and system for vehicle initialization and positioning using road boundary line constraints. The method includes: S1: acquiring forward view image data, INS data, and dynamic map data; S2: fitting the data according to the cubic curve equation to obtain lane line point clouds and road boundary line point clouds; S3: searching the INS data to obtain all possible initial poses of the vehicle; S4: calculating the vehicle's forward view frame initial pose based on the INS data, projecting the lane line point clouds onto the dynamic map data, performing ICP matching, and obtaining the matching result; S5: filtering initial poses that meet the matching conditions and verifying boundary constraints. If the initial poses meet the matching conditions and pass the boundary constraint verification, proceed to S6; otherwise, repeat S1 to S5; S6: using the initial pose that passes the boundary constraint verification as the optimal initial pose to complete the vehicle initialization and positioning. This invention avoids incorrect lane positioning during initialization.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving positioning technology, and in particular to a method and system for initializing vehicle positioning by using road boundary lines as constraints. Background Technology

[0002] In the field of autonomous driving positioning, most algorithms achieve vehicle localization by incorporating GPS, inertial navigation systems, and camera or LiDAR sensors. During vehicle localization, the vehicle's position needs to be initialized to obtain the current positioning result, which is then used to determine the vehicle's positioning during subsequent driving. Therefore, determining the vehicle's initial position is crucial to the accuracy of subsequent vehicle localization.

[0003] In autonomous driving technology, determining the vehicle's initial position means that the accuracy of the vehicle's initial positioning needs to be improved; current initial positioning methods include two types: camera-based initial positioning methods and lidar-based initial positioning methods.

[0004] In the field of autonomous driving, GNSS, IMU positioning, and map matching positioning technologies are widely used. However, GNSS and IMU devices are extremely susceptible to interference while the vehicle is in motion. In open or obstructed areas, positioning may be inaccurate, and lane departures may occur, leading to reduced initial positioning accuracy, impacting vehicle decision-making and planning, and threatening driving safety. Map matching technology is a process of performing a high-precision local map search given the vehicle's pose information. It first uses onboard GNSS and IMU to determine the initial position and define the local search range of the high-precision map. Then, data collected by camera sensors or LiDAR is transformed and matched with the high-precision map data to the same coordinate system.

[0005] When GNSS and IMU data contain significant errors, the pose obtained from map matching will also have large errors, affecting subsequent positioning. Map-based data fusion positioning technology combines the above two algorithms, fusing the positioning results obtained from map matching with GNSS / IMU data to further improve the accuracy and reliability of positioning.

[0006] During map matching and localization, the vehicle pose needs to be initialized to obtain the current vehicle location result, which is then used for subsequent vehicle positioning during driving. Therefore, determining the vehicle's initial pose is crucial to the accuracy of subsequent map matching and localization fusion.

[0007] Existing vehicle positioning initialization methods are prone to lane deviation when encountering environments with weak textures or few reference obstacles or other features. This can lead to the GPS or combined inertial navigation system following the wrong lane deviating from the lane during the positioning process, thus affecting the positioning accuracy and subsequent driving safety. Summary of the Invention

[0008] To address the problem in existing technologies where vehicles mistakenly position themselves in the wrong lane during initial INS data, a method and system for vehicle initial positioning using road boundary lines are proposed. This method constrains the vehicle's initial position, preventing it from misaligning during initial positioning.

[0009] The technical solution of the present invention is as follows:

[0010] This invention discloses a method for initializing vehicle positioning by constraining road boundary lines, comprising the following steps:

[0011] S1: Acquire vehicle front view image data, INS data, and dynamic map data;

[0012] S2: Fit the vehicle front view image data according to the cubic curve equation to obtain lane line point cloud and road boundary line point cloud;

[0013] S3: Find an INS data frame with a timestamp no less than the current vehicle front view image data timestamp and its previous INS data frame in the INS data, and obtain all possible initial poses of the vehicle after judgment;

[0014] S4: For all possible poses, the current vehicle forward-looking frame initial pose is obtained by calculating the trajectory based on the previous frame INS data. The lane line point cloud in the forward-looking frame is projected onto the dynamic map data, and ICP matching is performed with the lane line point cloud in the dynamic map data to obtain the matching result.

[0015] S5: Filter the initial poses that meet the matching conditions from the matching results, and perform boundary constraint verification on the initial poses that meet the matching conditions. If the initial poses meet the matching conditions and pass the boundary constraint verification, proceed to step S6; if the initial poses do not meet the matching conditions or fail the boundary constraint verification, repeat steps S1 to S5.

[0016] S6: Use the initial pose verified by boundary constraints as the optimal initial pose to complete the vehicle initialization positioning.

[0017] Preferably, in S1, the vehicle front view image data is lane line and road boundary line information after semantic segmentation.

[0018] The INS data and the vehicle front view image data are synchronized with the same clock source.

[0019] Preferably, the equation of the cubic curve in S2 is expressed as follows:

[0020] y = a0 + a1x + a2x 2 +a3x 3 (1)

[0021] In equation (1), a0, a1, a2, and a3 are the coefficients of the zero-order, 1-order, 2-order, and 3-order terms of the cubic curve equation, respectively, and x is the longitudinal coordinate of the vehicle; a0 represents the lateral distance between the road boundary line and the vehicle. When a0 is positive, it represents the lateral distance between the left road boundary line and the vehicle. When a0 is negative, it represents the lateral distance between the right road boundary line and the vehicle.

[0022] Furthermore, the method for obtaining the lane line point cloud is as follows:

[0023] The cubic curve equation is used to fit a single frame of lane lines and road boundary lines. Then, a 5-10 meter segment is taken from the fitted single frame of lane lines, and a point is taken at each preset interval to discretize it into a point cloud.

[0024] Preferably, the method for determining all possible vehicle initial poses in S3 is as follows:

[0025] If the pose of the previous frame of INS data is a fixed solution, then this pose is the possible pose of the vehicle in the previous frame.

[0026] If the pose of the previous frame of INS data is a floating-point solution, then the pose is projected onto the center line of the map, and the multiple poses obtained by the projection are the possible poses of the vehicle in the previous frame.

[0027] Preferably, the trajectory calculation process in S4 is as follows:

[0028] p n =p n-1 +R n-1 ·v n-1 ·Δt (2)

[0029]

[0030] In the formula, p n-1 R represents the vehicle's position at the previous moment. n-1 The rotation matrix v represents the vehicle's rotation at the previous moment. n-1 This indicates the vehicle's speed at the previous moment. p represents the angular acceleration of the vehicle at the previous moment. n R represents the current position of the vehicle. n This represents the rotation matrix of the vehicle at the current moment, where Δt is the time difference between the current moment and the previous moment.

[0031] Using equations (2) and (3), the INS data is used to extrapolate the trajectory from the previous frame. The vehicle's wheel speed and angular acceleration at the previous moment are used to calculate the vehicle's translational and rotational changes between the two frames, and the vehicle's initial pose in the current forward-looking frame is calculated.

[0032]

[0033] Preferably, the matching conditions in S5 are as follows:

[0034] The number of matched points exceeds the first threshold of the total number of forward-looking perception point clouds, the median error is less than the second threshold, and the number of matched points is greater than the third threshold.

[0035] Furthermore, the boundary constraints in S5 are verified as follows:

[0036] Find the two points in the point cloud of the left road boundary line that are closest to the initial pose that meet the matching conditions, calculate the distance d1 between the vehicle and the line segment formed by these two points, calculate the distance |a0| between the cubic curve and the vehicle body, and compare the difference between d1 and |a0| with the fourth threshold.

[0037] Find the two points in the point cloud of the right road boundary line that are closest to the initial pose that meet the matching conditions. Calculate the distance d2 between the vehicle and the line segment formed by these two points. Calculate the distance between the cubic curve and the vehicle body. d2 and The difference is compared with the fourth threshold; This represents the lateral distance between the right-hand road boundary line and the vehicle.

[0038] If the difference between d1 and |a0|, and d2 and |a0|, If the difference between the values ​​is less than the fourth threshold, then the initial pose at this time is considered to be the optimal pose.

[0039] Another aspect of the present invention discloses a vehicle initialization positioning system constrained by road boundary lines, which adopts the vehicle initialization positioning method constrained by road boundary lines as described above, including: a vehicle-mounted forward-looking camera, a vehicle-mounted integrated inertial navigation module, a data fitting module, a data search module, a data processing module, a data filtering module, and an initialization module;

[0040] The aforementioned vehicle-mounted forward-view camera is used to acquire forward-view image data of the vehicle;

[0041] The vehicle-mounted integrated inertial navigation module is used to acquire INS data and dynamic map data;

[0042] The data fitting module is used to fit the cubic curve equation to obtain lane line point clouds and road boundary line point clouds.

[0043] The data lookup module calculates the current vehicle forward-looking frame initial pose based on the trajectory of the previous INS frame data.

[0044] The data processing module is used to calculate the current vehicle forward-looking frame initial pose based on the previous frame of INS data for all possible poses, project the lane line point cloud in the forward-looking frame onto the dynamic map data, and perform ICP matching with the lane line point cloud in the dynamic map data to obtain the matching result.

[0045] The data filtering module filters initial poses that meet the matching conditions from the matching results, performs boundary constraint verification on the initial poses that meet the matching conditions, and if the initial poses meet the matching conditions and pass the boundary constraint verification, they enter the initialization module; if they do not meet the matching conditions or fail the boundary constraint verification, the working process of the vehicle-mounted forward-looking camera, vehicle-mounted integrated inertial navigation module, data fitting module, data search module, data processing module, and data filtering module is repeated.

[0046] The initialization module is used to take the initial pose verified by boundary constraints as the optimal initial pose to complete the vehicle initialization positioning.

[0047] More specifically, the vehicle initialization positioning system constrained by road boundary lines also includes a data container; the data container is used to save the acquired INS data in the order of timestamps.

[0048] The beneficial effects of this invention are:

[0049] 1. The vehicle's current location is obtained by acquiring front view image data, INS data, and dynamic map data, providing accurate data for vehicle initialization.

[0050] 2. By fitting the vehicle's front view image data with a cubic curve equation, point clouds of lane lines and road boundary lines are obtained, improving the accuracy of the acquired data.

[0051] 3. Based on the INS data, obtain all possible initial poses of the vehicle; then filter all initial poses and constrain the vehicle's initial position by determining whether the initial pose is the optimal initial pose, so as to avoid incorrect initial positioning that may affect subsequent positioning and driving.

[0052] 4. When the optimal initial pose cannot be obtained, the process of searching for the optimal initial pose is repeated to avoid obtaining an incorrect vehicle initial position, thereby improving the positioning accuracy and precision of vehicle initialization. Attached Figure Description

[0053] Figure 1 A flowchart of a vehicle initialization positioning method using road boundary lines as constraints, provided for an embodiment.

[0054] Figure 2 This is a schematic diagram of the vehicle coordinate system provided for an embodiment.

[0055] Figure 3 This is a schematic diagram illustrating the distance between the boundary line of a cubic curve road and the vehicle body, provided as an example.

[0056] Figure 4 This is a schematic diagram of the possible vehicle poses when the pose of the INS provided in the embodiment is a floating-point solution.

[0057] Figure 5 This is a schematic diagram illustrating the distance between a vehicle and the left-hand road boundary line, provided as an example.

[0058] Figure 6 This is a schematic diagram showing the comparison of initial poses with and without boundary constraints provided in the embodiment. Detailed Implementation

[0059] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0060] In environments with limited surrounding features, such as highways and urban expressways, when using lane line point clouds and high-precision map lane line point clouds for map matching and positioning, the accuracy of map matching and positioning is highly dependent on the initial position because the lane line point clouds of each lane are extremely similar. In practical applications, it has been found that after the positioning program starts, if the initial INS data is inaccurate—for example, if the INS positions the wrong lane—the vehicle's initial positioning will deviate from its current true position following the INS data, resulting in consistently inaccurate subsequent vehicle positioning and affecting driving functions.

[0061] To address the aforementioned issues, we consider using information beyond the road lane lines to constrain the vehicle's initial position. The forward-facing camera, while segmenting lane line information, can also segment the guardrails, boundaries, and curbs on both sides of the road. These features are unique and stably exist on both sides of the road. Based on the lateral distance between the vehicle's current position and these features, the vehicle's initial position can be effectively constrained, preventing incorrect lane initialization.

[0062] This embodiment provides a method for initializing vehicle positioning by using road boundary lines as constraints, such as... Figure 1 As shown, it includes the following steps:

[0063] S1: Acquire vehicle front view image data, INS data, and dynamic map data;

[0064] S2: Fit the vehicle front view image data according to the cubic curve equation to obtain lane line point cloud and road boundary line point cloud;

[0065] S3: Find an INS data frame with a timestamp no less than the current vehicle front view image data timestamp and its previous INS data frame in the INS data, and obtain all possible initial poses of the vehicle after judgment;

[0066] S4: For all possible poses, the current vehicle forward-looking frame initial pose is obtained by calculating the trajectory based on the previous frame INS data. The lane line point cloud in the forward-looking frame is projected onto the dynamic map data, and ICP matching is performed with the lane line point cloud in the dynamic map data to obtain the matching result.

[0067] S5: Filter the initial poses that meet the matching conditions from the matching results, and perform boundary constraint verification on the initial poses that meet the matching conditions. If the initial poses meet the matching conditions and pass the boundary constraint verification, proceed to step S6; if the initial poses do not meet the matching conditions or fail the boundary constraint verification, repeat steps S1 to S5.

[0068] S6: Use the initial pose verified by boundary constraints as the optimal initial pose to complete the vehicle initialization positioning.

[0069] In this embodiment, in S1, the vehicle front view image data is lane line and road boundary line information after semantic segmentation.

[0070] The INS data and the vehicle front view image data are synchronized with the same clock source, thus avoiding data deviations and inaccurate positioning due to time asynchrony.

[0071] In this embodiment, as Figure 2 As shown, the origin is located at the center of the rear axle of the vehicle, x points to the front of the vehicle, and y is perpendicular to x and points to the left side of the vehicle, following the right-hand rule.

[0072] In this embodiment, the equation of the cubic curve in S2 is expressed as follows:

[0073] y = a0 + a1x + a2x 2 +a3x 3 (1)

[0074] In equation (1), a0, a1, a2, and a3 are the coefficients of the zero-order, first-order, second-order, and third-order terms of the cubic curve equation, respectively, and x is the longitudinal coordinate of the vehicle; a0 represents the lateral distance between the road boundary line and the vehicle. A positive a0 represents the lateral distance between the left road boundary line and the vehicle, and a negative a0 represents the lateral distance between the right road boundary line and the vehicle. The method of obtaining lane line point clouds and road boundary line point clouds through cubic curve equations is simple and quick.

[0075] More specifically, such as Figure 3As shown, the two road boundary lines are located on the left and right sides of the vehicle, respectively.

[0076] The equations of the cubic curves are expressed as follows:

[0077] The road boundary line located on the left side of the vehicle:

[0078] y1=a0+a1x+a2x 2 +a3x 3

[0079] In the formula, a0 is positive and represents the lateral distance between the left road boundary line and the vehicle.

[0080] The road boundary line located on the right side of the vehicle:

[0081]

[0082] In the formula, A negative value indicates the lateral distance between the right-hand road boundary line and the vehicle. They correspond to a0, a1, a2, and a3 respectively.

[0083] More specifically, the method for obtaining lane line point clouds is as follows:

[0084] A single frame of lane lines and road boundary lines is fitted using a cubic curve equation. Then, a 5-10 meter segment is taken from the fitted single frame of lane lines, and a point is taken at each preset interval to discretize it into a point cloud. The preset interval can be 0.5m. The point cloud discretized at this distance can ensure high accuracy.

[0085] In this embodiment, the method for determining all possible vehicle initial poses in S3 is as follows:

[0086] If the pose of the previous frame of INS data is a fixed solution, then this pose is the possible pose of the vehicle in the previous frame.

[0087] If the pose of the previous frame's INS data is a floating-point solution, then project that pose onto the center line of the map. The resulting multiple poses represent the possible poses of the vehicle in the previous frame; for example... Figure 4 As shown, assuming the car is traveling on a road with 4 lanes in the same direction, the current INS floating-point solution is located in lane 3. Projecting this floating-point solution onto the center line of the 4 lanes yields 4 possible initial poses.

[0088] In this embodiment, the trajectory calculation process in S4 is as follows:

[0089] p n =p n-1 +R n-1 ·v n-1 ·Δt (2)

[0090]

[0091] In the formula, p n-1 R represents the vehicle's position at the previous moment. n-1 The rotation matrix v represents the vehicle's rotation at the previous moment. n-1 This indicates the vehicle's speed at the previous moment. p represents the angular acceleration of the vehicle at the previous moment. n R represents the current position of the vehicle. n This represents the rotation matrix of the vehicle at the current moment, where Δt is the time difference between the current moment and the previous moment.

[0092] Using equations (2) and (3), the INS data is used to extrapolate the trajectory from the previous frame. The vehicle's wheel speed and angular acceleration at the previous moment are used to calculate the vehicle's translational and rotational changes between the two frames, and the vehicle's initial pose in the current forward-looking frame is calculated.

[0093]

[0094] In this embodiment, the matching conditions in S5 are as follows:

[0095] The number of matched points is above a first threshold of the total number of forward-looking perception point clouds, the median error is less than a second threshold, and the number of matched points is greater than a third threshold; the first threshold can be 50%, the second threshold can be 0.8m, and the third threshold can be 50; in this way, some initial poses that do not meet the conditions are excluded.

[0096] More specifically, the boundary constraint verification in S5 is as follows:

[0097] Find the two points in the point cloud of the left road boundary line that are closest to the initial pose that meet the matching conditions, calculate the distance d1 between the vehicle and the line segment formed by these two points, calculate the distance |a0| between the cubic curve and the vehicle body, and compare the difference between d1 and |a0| with the fourth threshold. This represents the lateral distance between the right-hand road boundary line and the vehicle.

[0098] Find the two points in the point cloud of the right road boundary line that are closest to the initial pose that meet the matching conditions. Calculate the distance d2 between the vehicle and the line segment formed by these two points. Calculate the distance between the cubic curve and the vehicle body. d2 and The difference is compared with the fourth threshold.

[0099] If the difference between d1 and |a0|, and d2 and |a0|, If the differences between the initial pose and the target pose are all less than the threshold, then the initial pose is considered the optimal pose. This process eliminates all initial poses that do not meet the conditions, thereby improving the accuracy and precision of initial positioning.

[0100] In this embodiment, as Figure 5 As shown, select two boundary line points closest to the vehicle in the kdtree point cloud of the left boundary line. Assume that the straight line passing through these two points is l. Determine whether the projection of the vehicle on l is between these two points. If it is, calculate the vertical distance d1 from the vehicle to l, which is used as the distance between the vehicle and the left boundary line. The calculation method for the right boundary line d2 is the same.

[0101] In an embodiment, such as Figure 6 As shown, assuming the INS locates the wrong lane, since the lane line point clouds are quite similar, the ICP matching result on the wrong lane may also meet the requirements, causing the vehicle's initial pose to follow the INS wrong lane. After adding boundary constraints, the distance from the initial pose of the wrong lane to the map boundary line is much different from the distance of the boundary line fitted by the forward-looking camera, so it is filtered out.

[0102] In another specific embodiment, a vehicle initialization positioning system constrained by road boundary lines is disclosed, which adopts the vehicle initialization positioning method constrained by road boundary lines as described in Embodiment 1, including: vehicle-mounted forward-looking camera, vehicle-mounted combined inertial navigation module, data fitting module, data search module, data processing module, data filtering module, and initialization module.

[0103] The aforementioned vehicle-mounted forward-view camera is used to acquire forward-view image data of the vehicle;

[0104] The vehicle-mounted integrated inertial navigation module is used to acquire INS data and dynamic map data;

[0105] The data fitting module is used to fit the cubic curve equation to obtain lane line point clouds and road boundary line point clouds.

[0106] The data lookup module calculates the current vehicle forward-looking frame initial pose based on the trajectory of the previous INS frame data.

[0107] The data processing module is used to calculate the current vehicle forward-looking frame initial pose based on the previous frame of INS data for all possible poses, project the lane line point cloud in the forward-looking frame onto the dynamic map data, and perform ICP matching with the lane line point cloud in the dynamic map data to obtain the matching result.

[0108] The data filtering module filters initial poses that meet the matching conditions from the matching results, performs boundary constraint verification on the initial poses that meet the matching conditions, and if the initial poses meet the matching conditions and pass the boundary constraint verification, they enter the initialization module; if they do not meet the matching conditions or fail the boundary constraint verification, the working process of the vehicle-mounted forward-looking camera, vehicle-mounted integrated inertial navigation module, data fitting module, data search module, data processing module, and data filtering module is repeated.

[0109] The initialization module is used to take the initial pose verified by boundary constraints as the optimal initial pose to complete the vehicle initialization positioning.

[0110] In another specific embodiment, a method for initializing vehicle positioning by constraining road boundary lines is disclosed, including the following steps:

[0111] S1: Acquire vehicle front view image data through the vehicle-mounted front-view camera; acquire INS data and dynamic map data through the vehicle-mounted integrated inertial navigation module; save the INS data in the data container according to the timestamp order;

[0112] S2: The data processing module fits the obtained vehicle front view image data according to the cubic curve equation to obtain lane line point cloud and road boundary line point cloud;

[0113] S3: The data processing module searches for a frame of INS data in the data container that is not less than the timestamp of the current vehicle front view image data and its previous frame of INS data. After judgment, all possible poses of the vehicle initial pose are obtained.

[0114] S4: The vehicle-mounted integrated inertial navigation module calculates the current vehicle forward-looking frame initial pose based on the previous frame of INS data. The data processing module projects the lane line point cloud in the forward-looking frame onto the dynamic map data and performs ICP matching with the lane line point cloud in the map to obtain the matching result.

[0115] S5: The data processing module filters the initial poses that meet the matching conditions from the matching results, and performs boundary constraint verification on the initial poses that meet the matching conditions. If the initial poses meet the matching conditions and pass the boundary constraint verification, then proceed to step S6; if the initial poses do not meet the matching conditions or fail the boundary constraint verification, then repeat steps S1 to S5.

[0116] S6: Use the initial pose verified by boundary constraints as the optimal initial pose to complete the vehicle initialization positioning.

[0117] In this embodiment, the inertial measurement unit in the vehicle-mounted integrated inertial navigation module can measure the vehicle's acceleration and angular velocity. Combined with the vehicle's wheel speed pulse signal, the vehicle's accurate speed can be obtained in real time, thereby calculating the vehicle's pose for each frame through trajectory estimation.

[0118] The acquired dynamic map data can be represented as high-precision map data.

[0119] Multiple driving tests on highways have shown that using road boundary lines to constrain the vehicle's initial pose can effectively prevent inaccurate inertial navigation positioning from causing the vehicle to miss its lane during initial pose.

[0120] 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. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.

Claims

1. A method for initializing vehicle positioning by constraining road boundary lines, characterized in that, Includes the following steps: S1: Acquire vehicle front view image data, INS data, and dynamic map data; S2: Fit the vehicle front view image data according to the cubic curve equation to obtain lane line point clouds and road boundary line point clouds; the expression of the cubic curve equation is as follows: y=a0+a1x+a2x 2 +a3x 3 (1) In equation (1), a0, a1, a1, and a3 are the coefficients of the zero-order, first-order, second-order, and third-order terms of the cubic curve equation, respectively, and x is the longitudinal coordinate of the vehicle; a0 represents the lateral distance between the road boundary line and the vehicle. When a0 is positive, it represents the lateral distance between the left road boundary line and the vehicle; when a0 is negative, it represents the lateral distance between the right road boundary line and the vehicle. The method for obtaining lane line point clouds is as follows: The cubic curve equation is used to fit a single frame of lane lines and road boundary lines. Then, a 5-10 meter segment is taken from the fitted single frame of lane lines, and a point is taken at each preset interval to discretize it into a point cloud. S3: Find an INS data frame with a timestamp no less than the current vehicle front view image data timestamp and its previous INS data frame in the INS data, and obtain all possible initial poses of the vehicle after judgment; S4: For all possible poses, the current vehicle forward-looking frame initial pose is obtained by calculating the trajectory based on the previous frame INS data. The lane line point cloud in the forward-looking frame is projected onto the dynamic map data, and ICP matching is performed with the lane line point cloud in the dynamic map data to obtain the matching result. S5: Filter the initial poses that meet the matching conditions from the matching results, and perform boundary constraint verification on the initial poses that meet the matching conditions. If the initial poses meet the matching conditions and pass the boundary constraint verification, proceed to step S6; if the initial poses do not meet the matching conditions or fail the boundary constraint verification, repeat steps S1 to S5. S6: Use the initial pose verified by boundary constraints as the optimal initial pose to complete the vehicle initialization positioning.

2. The method for initializing vehicle positioning by means of road boundary lines according to claim 1, characterized in that, In S1, the vehicle front view image data is lane line and road boundary line information after semantic segmentation; The INS data and the vehicle front view image data are synchronized with the same clock source.

3. The method for initializing vehicle positioning by means of road boundary lines according to claim 1, characterized in that, In S1, the acquired INS data is saved in the order of timestamps.

4. The method for initializing vehicle positioning by means of road boundary lines according to claim 1, characterized in that, The method for determining all possible initial poses of the vehicle in S3 is as follows: If the pose of the previous frame of INS data is a fixed solution, then this pose is the possible pose of the vehicle in the previous frame. If the pose of the previous frame of INS data is a floating-point solution, then the pose is projected onto the center line of the map, and the multiple poses obtained by the projection are the possible poses of the vehicle in the previous frame.

5. The method for initializing vehicle positioning by means of road boundary lines according to claim 1, characterized in that, The trajectory calculation process in S4 is as follows: p n =p n-1 +R n-1 ·v n-1 ·Δt (2) In the formula, p n-1 R represents the vehicle's position at the previous moment. n-1 The rotation matrix v represents the vehicle's rotation at the previous moment. n-1 This indicates the vehicle's speed at the previous moment. p represents the angular acceleration of the vehicle at the previous moment. n R represents the current position of the vehicle. n This represents the rotation matrix of the vehicle at the current moment, where Δt is the time difference between the current moment and the previous moment. Using equations (2) and (3), the INS data is used to extrapolate the trajectory from the previous frame. The vehicle's wheel speed and angular acceleration at the previous moment are used to calculate the vehicle's translational and rotational changes between the two frames, and the vehicle's initial pose in the current forward-looking frame is calculated.

6. The method for initializing vehicle positioning by means of road boundary lines according to claim 1, characterized in that, The matching conditions in S5 are as follows: The number of matched points exceeds the first threshold of the total number of forward-looking perception point clouds, the median error is less than the second threshold, and the number of matched points is greater than the third threshold.

7. The method for initializing vehicle positioning by means of road boundary lines according to claim 1, characterized in that, The boundary constraints in S5 are verified as follows: Find the two points in the point cloud of the left road boundary line that are closest to the initial pose that meet the matching conditions, calculate the distance d1 between the vehicle and the line segment formed by these two points, calculate the distance |a0| between the cubic curve and the vehicle body, and compare the difference between d1 and |a0| with the fourth threshold. Find the two points in the point cloud of the right road boundary line that are closest to the initial pose that meet the matching conditions. Calculate the distance d2 between the vehicle and the line segment formed by these two points. Calculate the distance between the cubic curve and the vehicle body. d2 and The difference is compared with the fourth threshold; This refers to the lateral distance between the right-hand road boundary line and the vehicle. If the difference between d1 and |a0|, and d2 and |a0|, If the difference between the values ​​is less than the fourth threshold, then the initial pose at this time is considered to be the optimal pose.

8. A vehicle initialization positioning system constrained by road boundary lines, employing the vehicle initialization positioning method constrained by road boundary lines as described in any one of claims 1 to 7, characterized in that, include: Vehicle-mounted forward-looking camera, vehicle-mounted integrated inertial navigation module, data fitting module, data search module, data processing module, data filtering module, initialization module; The aforementioned vehicle-mounted forward-view camera is used to acquire forward-view image data of the vehicle; The vehicle-mounted integrated inertial navigation module is used to acquire INS data and dynamic map data; The data fitting module is used to fit the cubic curve equation to obtain lane line point clouds and road boundary line point clouds. The data lookup module calculates the current vehicle forward-looking frame initial pose based on the trajectory of the previous INS frame data. The data processing module is used to calculate the current vehicle forward-looking frame initial pose based on the previous frame of INS data for all possible poses, project the lane line point cloud in the forward-looking frame onto the dynamic map data, and perform ICP matching with the lane line point cloud in the dynamic map data to obtain the matching result. The data filtering module filters initial poses that meet the matching conditions from the matching results, performs boundary constraint verification on the initial poses that meet the matching conditions, and if the initial poses meet the matching conditions and pass the boundary constraint verification, they enter the initialization module; if they do not meet the matching conditions or fail the boundary constraint verification, the working process of the vehicle-mounted forward-looking camera, vehicle-mounted integrated inertial navigation module, data fitting module, data search module, data processing module, and data filtering module is repeated. The initialization module is used to take the initial pose verified by boundary constraints as the optimal initial pose to complete the vehicle initialization positioning.

9. A vehicle initialization positioning system constrained by road boundary lines according to claim 8, characterized in that, It also includes data containers; The data container is used to save the acquired INS data in the order of timestamps.

Images