Lane line point cloud splicing method, device, equipment, medium and vehicle

Abstract

Embodiments of the present application provide a lane line point cloud splicing method, device, equipment, medium and vehicle, and relate to the technical field of lane lines. The method comprises: acquiring each frame of lane line point cloud and each frame of combined inertial navigation data in real time; performing track prediction on each frame of combined inertial navigation data, determining the vehicle body pose corresponding to each frame of combined inertial navigation data, and storing the vehicle body pose corresponding to each frame of combined inertial navigation data and the corresponding first timestamp in a pose container; determining the vehicle body pose corresponding to the current frame of lane line point cloud through a pose difference value according to the second timestamp corresponding to the current frame of lane line point cloud and the pose container; storing the current frame of lane line point cloud and the corresponding vehicle body pose in a fixed-length point cloud queue; and splicing all lane line point clouds in the point cloud queue to obtain multiple frames of lane line splicing point clouds. Through the method of the embodiments, a lane line long enough and reliable in precision can be formed, and the probability of vehicle positioning deviating from the real position due to false matching is reduced.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of lane line technology, and in particular to a lane line point cloud stitching method, device, equipment, medium and vehicle. Background Technology

[0002] Currently, forward-facing cameras are one of the fundamental sensors for achieving autonomous driving in intelligent vehicles. Visual perception algorithms can provide rich road information, such as lane line information used for positioning. Lane lines play an extremely important role in autonomous driving systems for intelligent vehicles, and high-precision and robust lane line mapping has always been one of the key research technologies.

[0003] In related technologies, commonly used lane line recognition algorithms often encounter situations where poor data leads to missed or false detections of lane lines, or excessive noise. Specifically, when the accuracy of the identified single-frame lane line point cloud is poor, the vehicle's localization process frequently results in mismatches between the detected lane line point cloud and the high-precision map, causing it to deviate from its true position. Therefore, reducing the probability of mismatches when single-frame lane line point cloud data is poor is a key technical problem that this invention aims to solve. Summary of the Invention

[0004] This invention provides a method, apparatus, device, medium, and vehicle for stitching lane line point clouds, aiming to stitch together multiple frames of lane line point clouds based on vehicle pose, thereby reducing the probability of mismatch causing vehicle positioning to deviate from the true position.

[0005] The first aspect of this invention provides a method for stitching lane line point clouds, the method comprising:

[0006] Real-time acquisition of lane line point cloud for each frame;

[0007] Real-time acquisition of each frame of combined inertial navigation data;

[0008] For each frame of combined inertial navigation data, a trajectory is calculated to determine the vehicle body pose corresponding to each frame of combined inertial navigation data, and the vehicle body pose and the corresponding first timestamp are stored in the pose container.

[0009] Based on the second timestamp corresponding to the lane line point cloud of the current frame and the pose container, the vehicle pose corresponding to the lane line point cloud of the current frame is determined by the pose difference.

[0010] The current frame lane line point cloud and the corresponding vehicle body pose are stored in a fixed-length point cloud queue, which includes multiple historical frame lane line point clouds and the corresponding vehicle body poses.

[0011] All lane line point clouds in the point cloud queue are stitched together to obtain multi-frame stitched lane line point clouds.

[0012] Optionally, the combined inertial navigation data includes at least: vehicle speed and angular acceleration; the step of calculating the trajectory of each frame of combined inertial navigation data to determine the vehicle pose corresponding to each frame of combined inertial navigation data includes:

[0013] Based on the vehicle speed and angular acceleration in the previous frame of combined inertial navigation data, the time interval between the two frames of combined inertial navigation data, and the vehicle displacement and rotation matrix corresponding to the previous frame of combined inertial navigation data, the vehicle pose corresponding to the current frame of combined inertial navigation data is determined.

[0014] Optionally, determining the vehicle pose corresponding to the current frame of combined inertial navigation data based on the vehicle speed and angular acceleration in the previous frame of combined inertial navigation data, the time interval between the two frames of combined inertial navigation data, and the vehicle displacement and rotation matrix corresponding to the previous frame of combined inertial navigation data includes:

[0015] The vehicle pose corresponding to the current frame combined inertial navigation data is obtained using the first, second, and third formulas:

[0016] First formula: p n =p n-1 +R n-1 ·v n-1 ·Δt

[0017] Second formula:

[0018] The third formula:

[0019] Where, p n-1 and p n R represents the vehicle displacement corresponding to the previous frame of combined inertial navigation data and the current frame of combined inertial navigation data, respectively. n-1 and R n These are the rotation matrices corresponding to the previous frame combined inertial navigation data and the current frame combined inertial navigation data, respectively. n-1 Δt is the vehicle speed in the previous frame of combined inertial navigation data, and Δt is the time interval between the two frames of combined inertial navigation data. It is the angular acceleration T in the previous frame of combined inertial navigation data. n It is the vehicle pose corresponding to the current frame combined inertial navigation data.

[0020] Optionally, determining the vehicle pose corresponding to the lane line point cloud of the current frame based on the second timestamp corresponding to the lane line point cloud of the current frame and the pose container through the pose difference includes:

[0021] Based on the second timestamp corresponding to the lane line point cloud of the current frame, the first timestamp corresponding to the combined inertial navigation data of the two frames adjacent to the second timestamp is determined in the pose container, and the vehicle pose corresponding to the first timestamp corresponding to the combined inertial navigation data of the two frames is determined respectively.

[0022] The compensation factor is determined based on the second timestamp corresponding to the lane line point cloud of the current frame and the first timestamp corresponding to the combined inertial navigation data of the two preceding and following frames.

[0023] The vehicle pose corresponding to the lane line point cloud in the current frame is determined based on the compensation factor and the vehicle pose corresponding to the first timestamps of the combined inertial navigation data of the two frames before and after.

[0024] Optionally, determining the compensation factor based on the second timestamp corresponding to the lane line point cloud of the current frame and the first timestamp corresponding to the combined inertial navigation data of the two preceding and following frames includes:

[0025] The compensation factor is determined using the fourth formula:

[0026] The fourth formula: s=(t next -t c ) / / (t next -t prev )

[0027] The step of determining the vehicle pose corresponding to the lane line point cloud in the current frame based on the compensation factor and the vehicle pose corresponding to the first timestamps of the combined inertial navigation data of the preceding and following frames includes:

[0028] The vehicle pose corresponding to the lane line point cloud in the current frame is obtained using the fifth, sixth, and seventh formulas:

[0029] The fifth formula: p c =p prev ·s+p next ·(1-s)

[0030] The sixth formula: R c =R next ·R prev ·s

[0031] The seventh formula:

[0032] Where s is the compensation factor, t prev and t next These are the first timestamps corresponding to the two consecutive frames of combined inertial navigation data adjacent to the second timestamp, t c p is the second timestamp corresponding to the lane line point cloud of the current frame. prev and p nextp represents the vehicle displacement in the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames, respectively. c R represents the vehicle displacement corresponding to the lane line point cloud in the current frame. prev , and R next R represents the rotation matrix in the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames, respectively. c T is the rotation matrix corresponding to the lane line point cloud in the current frame. c The vehicle pose corresponding to the lane line point cloud in the current frame.

[0033] Optionally, the step of stitching together all lane line point clouds in the point cloud queue to obtain multi-frame stitched lane line point clouds includes:

[0034] The vehicle pose corresponding to the other lane line point cloud in each frame of the point cloud queue is subjected to matrix operation with the inverse of the vehicle pose corresponding to the lane line point cloud in the current frame, respectively, to obtain the coordinate system transformation relationship between the other lane line point cloud in each frame and the lane line point cloud in the current frame; wherein, the other lane line point cloud is the lane line point cloud in the point cloud queue other than the lane line point cloud in the current frame.

[0035] Each of the other lane line point clouds in each frame is projected onto the vehicle coordinate system of the current frame lane line point cloud through its corresponding coordinate system transformation relationship, resulting in the multi-frame lane line stitched point cloud located in the vehicle coordinate system of the current frame lane line point cloud.

[0036] Optionally, after determining the vehicle pose corresponding to the lane line point cloud of the current frame through the pose difference based on the second timestamp corresponding to the lane line point cloud of the current frame and the pose container, the method further includes:

[0037] Delete the timestamp before the second timestamp corresponding to the lane line point cloud of the current frame and the vehicle pose corresponding to the timestamp in the pose container.

[0038] Optionally, the method further includes:

[0039] The multi-frame lane line stitched point cloud is rasterized and downsampled to obtain the downsampled multi-frame lane line stitched point cloud.

[0040] Optionally, the step of rasterizing and downsampling the multi-frame lane line stitched point cloud to obtain a downsampled multi-frame lane line stitched point cloud includes:

[0041] The multi-frame lane line stitched point cloud is rasterized to obtain multiple grids;

[0042] Downsampling is performed on all point clouds in each of the multiple grids to obtain a single target point cloud for each grid. The downsampled multi-frame lane line stitching point cloud is then determined based on the single target point cloud of each grid.

[0043] Optionally, the real-time acquisition of each frame of lane line point cloud includes:

[0044] Real-time acquisition of lane line point cloud data output by the vehicle-mounted forward-looking camera for each frame, wherein the lane line point cloud data includes at least: initial lane line point cloud, lane line start point and lane line length;

[0045] Based on the lane line starting point and the lane line length, a lane line point cloud segment with a preset distance is selected from the initial lane line point cloud, and the lane line point cloud segment with the preset distance is used as the lane line point cloud for each frame.

[0046] A second aspect of the present invention provides a lane line point cloud stitching device, the device comprising:

[0047] The point cloud acquisition module is used to acquire lane line point clouds in real time for each frame.

[0048] The data acquisition module is used to acquire each frame of combined inertial navigation data in real time;

[0049] The first determining module is used to perform trajectory estimation on each frame of combined inertial navigation data, determine the vehicle body pose corresponding to each frame of combined inertial navigation data, and store the vehicle body pose corresponding to each frame of combined inertial navigation data and the corresponding first timestamp into the pose container.

[0050] The second determining module is used to determine the vehicle pose corresponding to the lane line point cloud of the current frame by means of the pose difference based on the second timestamp corresponding to the lane line point cloud of the current frame and the pose container.

[0051] The queue storage module is used to store the current frame lane line point cloud and the vehicle body pose corresponding to the current frame lane line point cloud into a point cloud queue of fixed length. The point cloud queue includes multiple historical frame lane line point clouds and the vehicle body pose corresponding to the multiple historical frame lane line point clouds.

[0052] The stitching module is used to stitch together all lane line point clouds in the point cloud queue to obtain multi-frame stitched lane line point clouds.

[0053] Optionally, the combined inertial navigation data includes at least: vehicle speed and angular acceleration; the first determining module includes:

[0054] The first determining submodule is used to determine the vehicle pose corresponding to the current frame of combined inertial navigation data based on the vehicle speed and angular acceleration in the previous frame of combined inertial navigation data, the time interval between the two frames of combined inertial navigation data, and the vehicle displacement and rotation matrix corresponding to the previous frame of combined inertial navigation data.

[0055] Optionally, the first determining submodule includes:

[0056] The first vehicle pose determination submodule is used to obtain the vehicle pose corresponding to the current frame combined inertial navigation data through the first formula, the second formula and the third formula.

[0057] First formula: p n =p n-1 +R n-1 ·v n-1 ·Δt

[0058] Second formula:

[0059] The third formula:

[0060] Where, p n-1 and p n R represents the vehicle displacement corresponding to the previous frame of combined inertial navigation data and the current frame of combined inertial navigation data, respectively. n-1 and R n These are the rotation matrices corresponding to the previous frame combined inertial navigation data and the current frame combined inertial navigation data, respectively. n-1 Δt is the vehicle speed in the previous frame of combined inertial navigation data, and Δt is the time interval between the two frames of combined inertial navigation data. It is the angular acceleration T in the previous frame of combined inertial navigation data. n It is the vehicle pose corresponding to the current frame combined inertial navigation data.

[0061] Optionally, the second determining module includes:

[0062] The timestamp determination submodule is used to determine, in the pose container, the first timestamp corresponding to the combined inertial navigation data of the two frames adjacent to the second timestamp, and the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames respectively, based on the second timestamp corresponding to the lane line point cloud of the current frame;

[0063] The factor determination submodule is used to determine the compensation factor based on the second timestamp corresponding to the lane line point cloud of the current frame and the first timestamp corresponding to the combined inertial navigation data of the two preceding and following frames.

[0064] The second vehicle pose determination submodule is used to determine the vehicle pose corresponding to the lane line point cloud in the current frame based on the compensation factor and the vehicle pose corresponding to the first timestamps of the combined inertial navigation data of the previous and next frames.

[0065] Optionally, the factor determination submodule includes:

[0066] The compensation factor determination submodule is used to determine the compensation factor using the fourth formula:

[0067] The fourth formula: s=(t next -t c ) / (t next -t prev )

[0068] The second vehicle body pose determination submodule includes:

[0069] The third vehicle pose determination submodule is used to obtain the vehicle pose corresponding to the lane line point cloud of the current frame through the fifth formula, the sixth formula and the seventh formula.

[0070] The fifth formula: p c =p prev ·s+p next ·(1-s)

[0071] The sixth formula: R c =R next ·R prev ·s

[0072] The seventh formula:

[0073] Where s is the compensation factor, t prev and t next These are the first timestamps corresponding to the two consecutive frames of combined inertial navigation data adjacent to the second timestamp, t c p is the second timestamp corresponding to the lane line point cloud of the current frame. prev and p next p represents the vehicle displacement in the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames, respectively. c R represents the vehicle displacement corresponding to the lane line point cloud in the current frame. prev and R next R represents the rotation matrix in the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames, respectively. c T is the rotation matrix corresponding to the lane line point cloud in the current frame. c The vehicle pose corresponding to the lane line point cloud in the current frame.

[0074] Optionally, the splicing module includes:

[0075] The transformation relationship determination submodule is used to perform matrix operations on the vehicle pose corresponding to the other lane line point cloud in each frame of the point cloud queue and the inverse of the vehicle pose corresponding to the current frame lane line point cloud, respectively, to obtain the coordinate system transformation relationship between the other lane line point cloud in each frame and the current frame lane line point cloud; wherein, the other lane line point cloud is the lane line point cloud in the point cloud queue other than the current frame lane line point cloud;

[0076] The stitching submodule is used to project the other lane line point clouds of each frame onto the vehicle coordinate system of the current frame lane line point cloud through their respective coordinate system transformation relationships, so as to obtain the multi-frame lane line stitched point cloud located in the vehicle coordinate system of the current frame lane line point cloud.

[0077] Optionally, the device further includes:

[0078] The deletion module is used to delete the timestamps before the second timestamp of the lane line point cloud in the current frame and the vehicle pose corresponding to the timestamps from the pose container after determining the vehicle pose corresponding to the lane line point cloud in the current frame by means of the pose difference based on the second timestamp corresponding to the lane line point cloud in the current frame and the pose container.

[0079] Optionally, the device further includes:

[0080] The downsampling processing module is used to perform rasterization and downsampling processing on the multi-frame lane line stitched point cloud to obtain the downsampled multi-frame lane line stitched point cloud.

[0081] Optionally, the downsampling processing module includes:

[0082] The rasterization submodule is used to rasterize the multi-frame lane line stitched point cloud to obtain multiple grids.

[0083] The downsampling processing submodule is used to perform downsampling processing on all point clouds in each of the multiple grids to obtain a single target point cloud for each grid, and to determine the downsampled multi-frame lane line stitching point cloud based on the single target point cloud of each grid.

[0084] Optionally, the point cloud acquisition module includes:

[0085] The data acquisition submodule is used to acquire lane line point cloud data output by the vehicle-mounted forward-looking camera in real time for each frame of lane line point cloud data. The lane line point cloud data includes at least: initial lane line point cloud, lane line start point and lane line length.

[0086] A selection submodule is used to select a lane line point cloud segment at a preset distance from the initial lane line point cloud based on the lane line starting point and the lane line length, and use the lane line point cloud segment at the preset distance as the lane line point cloud for each frame.

[0087] A third aspect of the present invention provides an electronic device, the electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the lane line point cloud stitching method of the first aspect of the present invention.

[0088] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the lane line point cloud stitching method of the first aspect of the present invention.

[0089] A fifth aspect of the present invention provides a vehicle, the vehicle including a positioning system, the positioning system being used to execute the lane line point cloud stitching method of the first aspect of the present invention.

[0090] The lane line point cloud stitching method provided in this invention performs trajectory calculation on each frame of combined inertial navigation data acquired in real time to obtain the vehicle pose corresponding to each frame of combined inertial navigation data, and stores it and the corresponding first timestamp in a pose container. Then, for the current frame of lane line point cloud acquired in real time, the vehicle pose corresponding to the current frame of lane line point cloud can be obtained through pose difference based on the second timestamp and the pose container. The current frame of lane line point cloud and the corresponding vehicle pose are stored in a point cloud queue including multiple consecutive historical frame lane line point clouds and their corresponding vehicle poses. Then, all lane line point clouds in the point cloud queue are stitched together to obtain multi-frame stitched lane line point clouds. This accumulates multiple consecutive single-frame lane line point clouds, i.e., stitches together the lane line point clouds of historical frames. By using the lane line point clouds of historical frames, the weight of lane line point clouds with poor single-frame quality is reduced, forming a sufficiently long and accurate multi-frame stitched lane line point cloud, thereby reducing the probability of vehicle positioning deviating from the true position due to mismatch. Attached Figure Description

[0091] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0092] Figure 1This is a schematic diagram illustrating the impact of the mounting height of the forward-looking camera on the z-axis on the width of the identified lane lines in related technologies.

[0093] Figure 2 This is a schematic diagram illustrating the effect of the roll angle of the forward-looking camera around the x-axis on the width of the identified lane lines in related technologies.

[0094] Figure 3 This is a schematic diagram illustrating the impact of road cross slope on the width of the identified lane lines in related technologies;

[0095] Figure 4 This is a flowchart illustrating a lane line point cloud stitching method according to an embodiment of the present invention;

[0096] Figure 5 This is a single-frame lane line point cloud extracted from semantic segmentation in a vehicle coordinate system, provided by an embodiment of the present invention;

[0097] Figure 6 This is a stitched image of a multi-frame lane line point cloud provided by an embodiment of the present invention;

[0098] Figure 7 This is a schematic diagram of lane line point cloud rasterization provided in an embodiment of the present invention;

[0099] Figure 8 This is a schematic flowchart illustrating a lane line point cloud stitching method according to an embodiment of the present invention;

[0100] Figure 9 This is a structural block diagram of a lane line point cloud stitching device provided in an embodiment of the present invention;

[0101] Figure 10 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0102] The technical solutions of the embodiments of the present invention will be clearly and completely described below 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.

[0103] As mentioned earlier, current lane line recognition technology cannot reliably guarantee accuracy. In practical applications, it has been found that the accuracy of the fitted single-frame lane line point cloud is poor at both near and far ends. This is because the perception accuracy of lane lines is related to factors such as the mounting height of the forward-looking camera on the z-axis, the roll angle around the x-axis, and the road cross slope (see [reference]). Figure 1 , 2 3). Among them, Figure 1This is a schematic diagram illustrating the impact of the mounting height of the forward-looking camera on the z-axis on the width of the identified lane lines in related technologies; for example... Figure 1 As shown, assuming the actual installation height of the forward-looking camera is 1.32 meters, the lane width is 5 meters, and the z-axis calibration result is 1.34 meters, then 5 ÷ 1.32 × 1.34 = 5.07 meters. In other words, a 0.02-meter error in the z-axis results in a lane width error of 0.07 meters. Figure 2 This is a schematic diagram illustrating the impact of the roll angle of the forward-looking camera around the x-axis on the width of the identified lane lines in related technologies; for example... Figure 2 As shown, assuming the actual roll angle of the forward-looking camera is 75.21 degrees and the calibration error is 0.5 degrees, then tan75.71° × 1.32 = 5.18 meters; tan74.71° × 1.32 = 4.82 meters. In other words, a 0.5° roll angle error causes a lane width error of 0.18 meters. Figure 3 This is a schematic diagram illustrating the impact of road cross slope on the identified lane width in related technologies; for example... Figure 3 As shown, assuming a road cross slope of 1.5%, the road height increment is 5 × 1.5% = 0.075 meters. Therefore, 5 ÷ (1.32 + 0.075) × 1.32 = 4.73 meters. This means that a 1.5% cross slope causes a lane width error of 0.27 meters. It is evident that if we discard the lane line point cloud information at the near and far ends of a single frame and only use the lane line point cloud information located in the middle section, the number of lane line point clouds available for matching with high-precision maps is too small. This often leads to mismatches during vehicle localization, causing deviations from the true location.

[0104] Therefore, in order to at least partially solve one or more of the above-mentioned problems and other potential problems, this invention proposes a lane line point cloud stitching method. The method calculates the vehicle pose corresponding to each frame of combined inertial navigation data by trajectory estimation, that is, the position and attitude change of the vehicle in three-dimensional space. By acquiring the lane line point cloud of the current frame, the method performs pose linear interpolation using the timestamp corresponding to the lane line point cloud of the current frame and each vehicle pose to obtain the vehicle pose corresponding to each frame of lane line point cloud. Finally, the lane line point cloud of the historical frames is projected onto the current frame using the vehicle pose to complete the stitching of the lane line point cloud, so as to form a lane line that is long enough and has reliable accuracy, thereby reducing the probability of mismatch causing the vehicle positioning to deviate from the true position.

[0105] refer to Figure 4 , Figure 4 This is a flowchart illustrating a lane line point cloud stitching method according to an embodiment of the present invention. Figure 4As shown, the lane line point cloud stitching method of this embodiment can be applied to a positioning system, which can be an in-vehicle positioning system or a positioning system in a vehicle backend server; this embodiment does not impose any limitations on this. The lane line point cloud stitching method of this embodiment may include the following steps:

[0106] Step S1: Acquire lane line point cloud for each frame in real time.

[0107] In this embodiment, lane line point clouds for each frame can be obtained in real time from the vehicle-mounted forward-looking camera. Specifically, the vehicle-mounted forward-looking camera in this embodiment can provide lane line point clouds corresponding to each captured image frame after semantic segmentation, i.e., provide each frame of lane line point clouds. Each frame of lane line point cloud semantically segmented by the vehicle-mounted forward-looking camera can be fitted by a cubic curve equation described by four parameters a0, a1, a2, and a3, as shown below:

[0108] y = a0 + a1x + a2x 2 +a3x 3

[0109] Where a0 is the constant term of the cubic curve parameter, a1 is the linear term of the cubic curve parameter, a2 is the quadratic term of the cubic curve parameter, θ3 is the cubic term of the cubic curve parameter, x is the longitudinal distance from the lane line point cloud to the vehicle, y is the lateral distance from the lane line point cloud to the vehicle, (x, y) is the coordinate of any point in the lane line point cloud, x is determined by the vehicle-mounted forward-looking camera, y is calculated by the cubic curve equation, and "the vehicle" is the origin of the vehicle coordinate system. Furthermore, in this embodiment, the vehicle coordinate system used for the lane line point cloud is such that x points towards the front of the vehicle, y is perpendicular to x pointing towards the left side of the vehicle body, and the origin is at the center of the rear axle of the vehicle, following the right-hand rule. Figure 5 As shown, Figure 5 This is a single-frame lane line point cloud extracted from semantic segmentation in a vehicle coordinate system, provided by an embodiment of the present invention.

[0110] Step S2: Acquire the combined inertial navigation data for each frame in real time.

[0111] In this embodiment, each frame of combined inertial navigation data collected by the vehicle-mounted combined inertial navigation system can be acquired in real time. In this embodiment, the vehicle-mounted combined inertial navigation system can collect the vehicle's INS data (i.e., combined inertial navigation data) in real time and send each frame of combined inertial navigation data to the positioning system for calculating the vehicle's position and attitude in three-dimensional space. Specifically, the vehicle-mounted forward-looking camera and the vehicle-mounted combined inertial navigation system in this embodiment need to be time-synchronized; each frame of lane line point cloud and each frame of combined inertial navigation data needs to be synchronized based on the same clock source to ensure time synchronization.

[0112] Step S3: Perform trajectory estimation on each frame of combined inertial navigation data to determine the vehicle pose corresponding to each frame of combined inertial navigation data, and store the vehicle pose and the corresponding first timestamp corresponding to each frame of combined inertial navigation data into the pose container.

[0113] In this embodiment, after the positioning system acquires each frame of combined inertial navigation data provided by the vehicle-mounted combined inertial navigation system, it can perform trajectory calculation on each frame of combined inertial navigation data. Specifically, the vehicle pose at the initial moment can be set as an identity matrix to obtain the relationship between two adjacent frames of combined inertial navigation data. By calculating the trajectory from the starting point using the acquired combined inertial navigation data, the vehicle pose corresponding to each frame of combined inertial navigation data can be determined, thus obtaining the vehicle pose at the current moment. In this embodiment, the starting moment is the moment when the lane line point cloud stitching method is started, and the starting point is the vehicle position corresponding to the start of the lane line point cloud stitching method.

[0114] After determining the vehicle pose corresponding to each frame of combined inertial navigation data, the vehicle pose corresponding to each frame of combined inertial navigation data and the first timestamp corresponding to each frame of combined inertial navigation data can be stored in a pose container. In this embodiment, the first timestamp is the timestamp corresponding to the combined inertial navigation data, and the pose container stores the vehicle pose corresponding to each of multiple frames of combined inertial navigation data and the first timestamp corresponding to each of the multiple frames of combined inertial navigation data.

[0115] Step S4: Determine the vehicle pose corresponding to the lane line point cloud of the current frame by means of the pose difference, based on the second timestamp corresponding to the lane line point cloud of the current frame and the pose container.

[0116] In this embodiment, lane line point clouds can be acquired in real time for each frame, and the most recently acquired lane line point cloud is the current frame's lane line point cloud. However, due to the inconsistency between the operating frequency of the vehicle-mounted forward-looking camera and the operating frequency of the vehicle-mounted integrated inertial navigation system, the first timestamp corresponding to any frame of integrated inertial navigation data is different from the second timestamp corresponding to the lane line point cloud. Therefore, the positioning system can determine the vehicle pose corresponding to the current frame's lane line point cloud by using the pose difference, based on the second timestamp corresponding to the current frame's lane line point cloud, the first timestamp in the pose container, and the vehicle pose in the pose container. Specifically, the vehicle pose corresponding to each frame's lane line point cloud can be determined by using the linear difference between the second timestamp corresponding to the lane line point cloud and the first timestamp corresponding to the integrated inertial navigation data. In this embodiment, the second timestamp is the timestamp corresponding to the lane line point cloud, which is also the timestamp corresponding to the image data acquired by the vehicle-mounted forward-looking camera.

[0117] Step S5: Store the current frame lane line point cloud and the corresponding vehicle pose in a fixed-length point cloud queue. The point cloud queue includes multiple historical frame lane line point clouds and the corresponding vehicle poses.

[0118] In this embodiment, after obtaining the vehicle pose corresponding to the lane line point cloud of the current frame, the lane line point cloud of the current frame and its corresponding vehicle pose can be stored in a point cloud queue of fixed length. Specifically, the point cloud in this embodiment corresponds to a fixed-length circular queue, which includes multiple historical frame lane line point clouds and their corresponding vehicle poses. In particular, the lane line point cloud of the latest frame and its corresponding vehicle pose can be inserted into the tail of the point cloud queue each time, i.e., the current frame lane line point cloud and its corresponding vehicle pose can be stored in the tail of the point cloud queue. The fixed length in this embodiment is a pre-defined number of frames for lane line point cloud stitching. The fixed length of the point cloud queue can be set according to project requirements, such as 20, 30, or 40 frames. This embodiment does not limit the specific value of the fixed length.

[0119] Step S6: Stitch together all lane line point clouds in the point cloud queue to obtain multi-frame stitched lane line point clouds.

[0120] In this embodiment, after storing the current frame lane line point cloud and its corresponding vehicle pose into the point cloud queue, all lane line point clouds in the point cloud queue can be stitched together using the vehicle pose in the point cloud queue to obtain a multi-frame stitched lane line point cloud. That is, the current frame lane line point cloud in the point cloud queue is stitched together with the lane line point clouds of several consecutive historical frames preceding the current frame lane line point cloud in the point cloud queue. This stitching of multiple consecutive frames of lane line point clouds is then applied to the vehicle coordinate system of the current frame lane line point cloud, resulting in a multi-frame stitched lane line point cloud, thus enriching the lane line point cloud information.

[0121] In this embodiment, the vehicle pose corresponding to each frame of combined inertial navigation data obtained from the trajectory calculation and its corresponding first timestamp are stored in a pose container. After obtaining the lane line point cloud of the current frame, the vehicle pose corresponding to the lane line point cloud of the current frame is determined by the pose difference based on the second timestamp and the pose container. The lane line point cloud of the current frame and its corresponding vehicle pose are then stored in a fixed-length point cloud queue. All lane line point clouds in the point cloud queue are then stitched together, thereby stitching multiple consecutive historical frame lane line point clouds into the current frame to form a multi-frame stitched lane line point cloud. This embodiment uses point cloud data from historical frames to reduce the weight of point clouds with poor quality in a single frame, forming sufficiently long and reliable lane lines, thereby reducing the probability of mismatches causing the vehicle positioning to deviate from the true position.

[0122] High-speed driving tests showed that the multi-frame lane line stitching point cloud obtained by the method in this embodiment maintained stability and high accuracy on both straight and curved roads. Because historical frame information was used, even if the quality of a particular frame's point cloud is poor, its proportion is only 1 / fixed length of the total, thus greatly reducing its impact on matching accuracy. In other words, the matching effect of the multi-frame lane line point cloud obtained by this embodiment is far superior to that of the single-frame lane line point cloud.

[0123] In conjunction with the above embodiments, in one implementation, the present invention also provides a lane line point cloud stitching method. In this method, the combined inertial navigation data includes at least: vehicle speed and angular acceleration; the step S3 above, "performing trajectory calculation on each frame of combined inertial navigation data to determine the vehicle pose corresponding to each frame of combined inertial navigation data," specifically includes step S31:

[0124] Step S31: Determine the vehicle pose corresponding to the current frame of combined inertial navigation data based on the vehicle speed and angular acceleration in the previous frame of combined inertial navigation data, the time interval between the two frames of combined inertial navigation data, and the vehicle displacement and rotation matrix corresponding to the previous frame of combined inertial navigation data.

[0125] In this embodiment, it can be assumed that the vehicle maintains a constant speed during the time interval between two frames of combined inertial navigation data. For example, assuming the frequency of the onboard combined inertial navigation system is 100Hz, the time interval between two frames of combined inertial navigation data is 10ms, and the vehicle maintains a constant speed during this 10ms time interval. This embodiment can obtain the vehicle's accurate speed in real time by measuring the acceleration and angular velocity using the inertial measurement unit in the onboard combined inertial navigation system, combined with the vehicle's wheel speed pulse signal. This allows for the calculation of the vehicle's position and attitude corresponding to each frame of combined inertial navigation data through trajectory estimation. Specifically, the combined inertial navigation data obtained in this embodiment includes at least the current vehicle speed and angular acceleration. Therefore, when determining the vehicle's attitude corresponding to each frame of combined inertial navigation data, the vehicle's attitude corresponding to the current frame of combined inertial navigation data can be determined based on the vehicle speed and angular acceleration in the previous frame of combined inertial navigation data, the time interval between the two frames of combined inertial navigation data, and the vehicle displacement and rotation matrix corresponding to the previous frame of combined inertial navigation data.

[0126] In an optional embodiment, step S31 above can specifically obtain the vehicle pose corresponding to the current frame combined inertial navigation data through the first formula, the second formula, and the third formula;

[0127] First formula: p n =p n-1 +R n-1 ·v n-1 ·Δt

[0128] Second formula:

[0129] The third formula:

[0130] Where, p n-1 and p n These represent the vehicle displacements corresponding to the combined inertial navigation data of the previous frame and the combined inertial navigation data of the current frame, respectively, R. n-1 and R n These are the rotation matrices corresponding to the combined inertial navigation data of the previous frame and the combined inertial navigation data of the current frame, respectively, v n-1 The vehicle speed is the value in the previous frame of combined inertial navigation data, and Δt is the time interval between two frames of combined inertial navigation data. It is the angular acceleration T in the previous frame of combined inertial navigation data. n It is the vehicle pose corresponding to the current frame's combined inertial navigation data.

[0131] In this embodiment, the vehicle pose at the initial moment can be set as an identity matrix. The combined inertial navigation data can be used to calculate the trajectory from the starting point using the first and second formulas mentioned above. The translational and rotational changes between frames can be calculated using the vehicle speed and angular acceleration of the previous moment. The pose at the current moment can be calculated using the third formula mentioned above, that is, the vehicle pose corresponding to the combined inertial navigation data of the current frame.

[0132] In conjunction with the above embodiments, in one implementation, the present invention also provides a lane line point cloud stitching method. In this method, step S4 specifically includes steps S41 to S43:

[0133] Step S41: Based on the second timestamp corresponding to the lane line point cloud of the current frame, determine the first timestamp corresponding to the combined inertial navigation data of the two frames adjacent to the second timestamp in the pose container, and the vehicle pose corresponding to the first timestamp corresponding to the combined inertial navigation data of the two frames respectively.

[0134] In this embodiment, because the operating frequencies of the integrated inertial navigation system and the forward-looking camera are inconsistent, the sampling timestamps of the integrated inertial navigation data and the lane line point cloud in any given frame are different. Therefore, in order to obtain the vehicle pose corresponding to the lane line point cloud in the current frame, under the assumption that the vehicle moves at a constant speed during the time interval between two frames of integrated inertial navigation data, linear interpolation can be used to interpolate the vehicle pose calculated from the trajectory, thereby obtaining the vehicle pose corresponding to the lane line point cloud located between two consecutive frames of integrated inertial navigation data, i.e., the vehicle pose corresponding to the lane line point cloud in the current frame.

[0135] In this embodiment, firstly, in the pose container, based on the second timestamp corresponding to the lane line point cloud of the current frame, the first timestamp corresponding to the combined inertial navigation data of the two frames preceding and following frames that is adjacent to the second timestamp corresponding to the lane line point cloud of the current frame is found; and in the pose container, the vehicle pose corresponding to the first timestamp corresponding to the combined inertial navigation data of the two frames preceding and following frames that is adjacent to the second timestamp corresponding to the lane line point cloud of the current frame is found.

[0136] Step S42: Determine the compensation factor based on the second timestamp corresponding to the lane line point cloud of the current frame and the first timestamp corresponding to the combined inertial navigation data of the two preceding and following frames.

[0137] In this embodiment, after obtaining the first timestamp corresponding to the second timestamp of the lane line point cloud in the current frame and the combined inertial navigation data of the two frames before and after it, a compensation factor representing the timestamp ratio and scale can be determined based on the second timestamp corresponding to the lane line point cloud in the current frame and the first timestamp corresponding to the combined inertial navigation data of the two frames before and after it.

[0138] In one optional embodiment, step S42 above can be specifically determined by the fourth formula to determine the compensation factor;

[0139] The fourth formula: s=(t next -t c ) / (t next -t orev )

[0140] Where s is the compensation factor, t prev and t next These are the first timestamps corresponding to the combined inertial navigation data of the two frames preceding and following the second timestamp of the lane line point cloud in the current frame, respectively. c This is the second timestamp corresponding to the lane line point cloud in the current frame.

[0141] Step S43: Determine the vehicle pose corresponding to the lane line point cloud in the current frame based on the compensation factor and the vehicle pose corresponding to the first timestamps of the combined inertial navigation data of the previous and next frames.

[0142] In this embodiment, after determining the compensation factor, the vehicle pose corresponding to the current lane line point cloud can be determined based on the compensation factor and the vehicle pose corresponding to the first timestamps of the combined inertial navigation data of the two frames before and after the current frame, which are adjacent to the second timestamp of the lane line point cloud in the pose container.

[0143] In one optional embodiment, step S43 can specifically obtain the vehicle pose corresponding to the lane line point cloud of the current frame through the fifth formula, the sixth formula, and the seventh formula:

[0144] The fifth formula: p c =p prev ·s+p next ·(1-s)

[0145] The sixth formula: R c =R next ·R prev ·s

[0146] The seventh formula:

[0147] Where s is the compensation factor, p prev and p next p represents the vehicle displacement in the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames, respectively. c R represents the vehicle displacement corresponding to the lane line point cloud in the current frame. prev and R next R represents the rotation matrix in the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames, respectively. c T is the rotation matrix corresponding to the lane line point cloud in the current frame. c This represents the vehicle pose corresponding to the lane line point cloud in the current frame.

[0148] In other words, in this embodiment, the pose container can first be configured according to the timestamp t corresponding to the current lane line point cloud. c Find the timestamp t of the combined inertial navigation data of the two adjacent frames. prev and t next and pose p prev R prev and p next R next Then, linear interpolation is performed using the fourth to seventh formulas to obtain the vehicle pose corresponding to the lane line point cloud in the current frame, thereby obtaining the vehicle pose corresponding to the lane line point cloud in each frame.

[0149] In conjunction with the above embodiments, in one implementation, the present invention also provides a lane line point cloud stitching method. In this method, step S6 may specifically include steps S61 and S62:

[0150] Step S61: Perform matrix operations on the vehicle pose corresponding to the other lane line point cloud in each frame of the point cloud queue and the inverse of the vehicle pose corresponding to the lane line point cloud in the current frame, respectively, to obtain the coordinate system transformation relationship between the other lane line point cloud in each frame and the lane line point cloud in the current frame.

[0151] In this embodiment, since the point cloud queue is a fixed-length circular queue, after inserting the current frame lane line point cloud and its corresponding vehicle pose into the tail of the point cloud queue, the historical frame lane line point cloud and its corresponding vehicle pose at the original head position in the point cloud queue will leave the point cloud queue, thus ensuring the fixed length of the point cloud queue. After inserting the current frame lane line point cloud and its corresponding vehicle pose into the tail of the point cloud queue to obtain a new point cloud queue, the vehicle pose corresponding to each other lane line point cloud in the point cloud queue can be matrix-operated with the inverse of the vehicle pose corresponding to the current frame lane line point cloud at the tail of the point cloud queue, thereby obtaining the coordinate system transformation relationship between each other lane line point cloud and the current frame lane line point cloud. In this embodiment, the other lane line point clouds refer to the lane line point clouds in the point cloud queue other than the current frame lane line point cloud.

[0152] For example, we can iterate through all lane line point clouds and their corresponding vehicle poses in the point cloud queue. For each frame of other lane line point clouds, after retrieving a frame of other lane line point clouds and its corresponding vehicle pose, we perform a matrix operation on the vehicle pose corresponding to the other lane line point clouds in that frame and the inverse of the vehicle pose corresponding to the current frame's lane line point cloud. This yields the transformation relationship between the vehicle coordinate systems of the two frames, i.e., the coordinate system transformation relationship between the other lane line point clouds in that frame and the current frame's lane line point clouds. Specifically, the coordinate system transformation relationship can be obtained through the eighth formula:

[0153] The eighth formula:

[0154] Among them, T current For the vehicle pose corresponding to the lane line point cloud in the current frame, T n For the vehicle pose corresponding to the point clouds of other lane lines in the nth frame. This represents the coordinate system transformation relationship between the point clouds of other lane lines in the nth frame and the point cloud of lane lines in the current frame, which is also the pose transformation matrix from the nth frame to the current frame.

[0155] Step S62: Project the other lane line point clouds of each frame onto the vehicle coordinate system of the current frame lane line point cloud through their respective coordinate system transformation relationships to obtain the multi-frame lane line stitched point cloud located in the vehicle coordinate system of the current frame lane line point cloud.

[0156] In this embodiment, after obtaining the coordinate system transformation relationship between the point clouds of other lane lines in each frame and the point cloud of the current frame, the point clouds of other lane lines in each frame can be projected onto the vehicle coordinate system of the point cloud of the current frame through their respective coordinate system transformation relationships. This results in a multi-frame stitched point cloud of lane lines located in the vehicle coordinate system of the point cloud of the current frame, thus completing the stitching of the multi-frame lane line point clouds. In this embodiment, the projection transformation between the two coordinate systems can be performed after obtaining each coordinate system transformation relationship; alternatively, the projection transformation of each corresponding coordinate system can be performed uniformly after obtaining all the coordinate system transformation relationships in the point cloud queue. This embodiment does not impose any restrictions on this approach.

[0157] In this embodiment, after determining the coordinate system transformation relationship of each historical frame by inversely determining the vehicle pose corresponding to the lane line point cloud of each historical frame and the vehicle pose corresponding to the lane line point cloud of the current frame, the lane line point clouds of each historical frame can be projected onto the vehicle coordinate system of the lane line point cloud of the current frame through their respective coordinate system transformation relationship, thereby obtaining a new lane line point cloud located in the vehicle coordinate system of the lane line point cloud of the current frame, thus completing the point cloud stitching of multiple frames of lane lines.

[0158] In one embodiment, such as Figure 6 As shown, Figure 6 This is a stitched image of a multi-frame lane line point cloud provided by an embodiment of the present invention. Figure 6 In the diagram, black represents the single-frame lane line point cloud at the current moment, while white represents the stitched multi-frame lane line point cloud. By using point cloud data from historical frames, the weight of poor-quality point clouds in a single frame is reduced, forming a sufficiently long and accurate lane line, thereby reducing the probability of mismatch causing the vehicle positioning to deviate from the true position.

[0159] In conjunction with the above embodiments, in one implementation, the present invention also provides a lane line point cloud stitching method. In this method, after step S4, step S4.5 may be further included:

[0160] Step S4.5: Delete the timestamp before the second timestamp corresponding to the lane line point cloud of the current frame and the vehicle pose corresponding to the timestamp in the pose container.

[0161] In this embodiment, as the vehicle pose data and corresponding first timestamp data in the pose container increase, the size of the pose container grows continuously. This reduces the search speed within the pose container, thus affecting the efficiency of the lane line point cloud stitching method provided in this embodiment. Therefore, in order to maintain the size of the pose container within a reasonable and as small a range as possible, after each pose difference is completed to obtain the vehicle pose corresponding to the current frame lane line point cloud, the data before the second timestamp corresponding to the pose difference in the current frame lane line point cloud is deleted from the pose container; that is, the timestamps before the second timestamp corresponding to the current frame lane line point cloud are deleted from the pose container, and the vehicle poses corresponding to the timestamps before the second timestamp corresponding to the current frame lane line point cloud are deleted from the pose container.

[0162] In conjunction with the above embodiments, in one implementation, the present invention also provides a lane line point cloud stitching method. In addition to the steps described above, this method may further include step S7:

[0163] Step S7: Rasterize and downsample the multi-frame lane line stitched point cloud to obtain the downsampled multi-frame lane line stitched point cloud.

[0164] In this embodiment, since the number of point clouds after stitching together multiple lane line point clouds is very large, it will increase the computational resources and time consumption, affecting the running efficiency of the algorithm. Moreover, the more points there are, the slower the projection stitching speed becomes. Furthermore, the computation speed will decrease as the number of point clouds increases during subsequent iterative nearest point matching and backend pose optimization. Therefore, this embodiment needs to perform rasterization processing on the stitched multi-frame lane line stitched point cloud to achieve downsampling, thereby obtaining a downsampled multi-frame lane line stitched point cloud.

[0165] In one optional embodiment, step S7 may specifically include steps S71 and S72:

[0166] Step S71: Rasterize the multi-frame lane line stitched point cloud to obtain multiple grids.

[0167] In this embodiment, the projected multi-frame lane line stitched point cloud can be represented in a two-dimensional plane using an occupancy grid map, that is, the multi-frame lane line stitched point cloud is rasterized to obtain multiple grids. In this embodiment, the area of ​​each grid can be set, for example, the area occupied by each grid can be 0.1m × 0.1m. This embodiment does not limit the specific value of the area occupied by each grid.

[0168] Step S72: Downsample all point clouds in each of the multiple grids to obtain a single target point cloud for each grid, and determine the downsampled multi-frame lane line stitching point cloud based on the single target point cloud of each grid.

[0169] In this embodiment, after rasterization to obtain multiple grids, downsampling can be performed on all point clouds in each grid to obtain a single target point cloud for each grid. Thus, based on the single target point cloud in each grid, the downsampled multi-frame lane line stitching point cloud can be determined.

[0170] For example, one could take all the point clouds within each grid cell (i.e., all the points within each grid cell), multiply the point cloud coordinates (x, y) by 10, round down, and then multiply by 0.1. This would downsample all the point clouds within each grid cell into a single target point cloud. Since there is only one target point cloud per grid cell, this results in a single target point cloud for each grid cell. Specifically, the single target point cloud within each grid cell can be obtained using the ninth formula:

[0171] The ninth formula: Where (x, y)′ is the single target point cloud of the raster, and (x, y) are the coordinates of all point clouds within the raster.

[0172] In one embodiment, such as Figure 7 As shown, Figure 7 This is a schematic diagram of lane line point cloud rasterization provided in an embodiment of the present invention. Figure 7 In this model, each grid cell occupies an area of ​​0.1m × 0.1m. Downsampling is used to select the top-left vertex of each cell to represent all point clouds within that cell. Figure 7 The point in the upper left corner of the middle grid is the single target point cloud of that grid.

[0173] In conjunction with the above embodiments, in one implementation, the present invention also provides a lane line point cloud stitching method. In this method, step S1 may specifically include steps S11 and S12:

[0174] Step S11: Acquire lane line point cloud data for each frame output by the vehicle-mounted forward-looking camera in real time. The lane line point cloud data includes at least: initial lane line point cloud, lane line start point, and lane line length.

[0175] In this embodiment, in addition to providing lane line point clouds in real time, the vehicle-mounted forward-looking camera can also simultaneously provide information on the lane line start point (start_x) and lane line length (line_range). That is, this embodiment can acquire each frame of lane line point cloud data output by the vehicle-mounted forward-looking camera in real time. The lane line point cloud data includes at least: an initial lane line point cloud, a lane line start point, and a lane line length. In other words, each frame of lane line point cloud data acquired in this embodiment includes at least: an initial lane line point cloud frame, a lane line start point for each frame, and a lane line length for each frame.

[0176] Step S12: Based on the lane line starting point and the lane line length, select a lane line point cloud segment with a preset distance in the initial lane line point cloud, and use the lane line point cloud segment with the preset distance as the lane line point cloud for each frame.

[0177] As mentioned above, the accuracy of each frame of lane line point cloud output by the current vehicle-mounted forward-looking camera is poor at the near and far ends, with only the middle segment having good accuracy. In order to achieve higher accuracy positioning, in this embodiment, after acquiring each frame of lane line point cloud data, a lane line point cloud segment with a preset distance can be selected from the initial lane line point cloud of that frame based on the lane line starting point, lane line length, and preset distance in the lane line point cloud data of that frame. The lane line point cloud segment with the preset distance is used as the lane line point cloud of each frame for subsequent stitching of single-frame lane line point clouds.

[0178] In this embodiment, the preset distance can be any distance segment from the starting point of the lane line, set arbitrarily according to requirements. For example, the preset distance can be 5-10m, where 5m represents the distance from the starting point of the preset distance to 5m from the starting point of the lane line, and 10m represents the distance from the ending point of the preset distance to 10m from the starting point of the lane line. This embodiment does not limit the specific range of the preset distance. It should be noted that the ending point of the preset distance is generally less than the length of the lane line. When the ending point of the preset distance is greater than the length of the lane line, the lane line point cloud segment of the preset distance is all lane line point cloud segments after the starting point of the preset distance in the initial lane line point cloud.

[0179] In an optional implementation, after obtaining lane line point cloud segments at a preset distance, a point cloud can be taken at fixed distance intervals (e.g., the fixed distance interval can be 0.5m, which is not limited in this embodiment) within the lane line point cloud segments at the preset distance, thereby obtaining each frame of lane line point cloud for subsequent lane line point cloud stitching.

[0180] In another embodiment, such as Figure 8 As shown, Figure 8 This is a schematic flowchart illustrating a lane line point cloud stitching method according to an embodiment of the present invention.

[0181] exist Figure 8The method begins by acquiring image data and INS data in real time. Image data refers to lane-related information from the vehicle's forward-facing camera after semantic segmentation; that is, acquiring lane-line point clouds for each frame in real time. INS data is combined inertial navigation data, acquired for each frame in real time. Both INS data and image data need to be synchronized using the same clock source.

[0182] Secondly, a single-frame lane line point cloud is obtained through cubic curve fitting. Then, trajectory calculation is performed based on INS data to obtain the vehicle pose corresponding to each frame of combined inertial navigation data, and the vehicle pose and corresponding timestamp for each frame of combined inertial navigation data are stored in the pose container. The steps "cubic curve fitting of a single-frame lane line point cloud" and "trajectory calculation, recording timestamps and poses" can be performed simultaneously or sequentially, without restriction.

[0183] Next, the vehicle pose corresponding to the lane line point cloud in the current frame is obtained through pose difference: Since the operating frequency of the INS and the camera are inconsistent, the sampling timestamps of the INS data and image data in any frame are different. Therefore, in order to obtain the vehicle pose corresponding to the image data sampling time, under the assumption of uniform motion (the frequency of the combined inertial navigation is 100Hz, so it is assumed that the vehicle moves at a uniform speed within 10ms), linear interpolation is used to interpolate the pose calculated from the trajectory, thereby obtaining the pose corresponding to the lane line point cloud located between two consecutive frames of combined inertial navigation data (i.e., the vehicle pose corresponding to the lane line point cloud in the current frame).

[0184] After obtaining the vehicle pose corresponding to the lane line point cloud of the current frame, the single-frame lane line point cloud (i.e. the current frame lane line point cloud) along with the corresponding vehicle pose can be stored in a fixed-length circular queue. Each time, the latest frame lane line point cloud and the corresponding vehicle pose are inserted into the tail of the circular queue. This circular queue is the point cloud queue used for stitching the lane line point clouds of historical frames.

[0185] Then, based on the vehicle pose, the lane line point clouds of historical frames in the circular queue are projected onto the vehicle coordinate system of the current frame's lane line point cloud. This can be done by iterating through all lane line point cloud data in the circular queue, retrieving a lane line point cloud frame and its corresponding pose each time, and then performing a matrix operation on the inverse of the vehicle pose corresponding to that frame's lane line point cloud and the current frame's lane line point cloud to obtain the transformation relationship between the two coordinate systems. The lane line point cloud of that frame can then be projected onto the vehicle coordinate system of the current frame's lane line point cloud using this transformation relationship. Similarly, the lane line point clouds of other frames are projected onto the vehicle coordinate system of the current frame's lane line point cloud in the same way, thus obtaining a new lane line point cloud located in the coordinate system of the current frame, completing the stitching of lane line point clouds from multiple frames.

[0186] Finally, the stitched multi-frame lane line point cloud is rasterized and downsampled. After stitching, the number of points is very large, increasing computational resources and time consumption, thus affecting the algorithm's efficiency. The more points, the slower the projection stitching speed, and the computation speed decreases further with the increase in the number of points during subsequent iterative nearest-point matching and backend pose optimization. Therefore, it is necessary to rasterize the stitched multi-frame lane line point cloud to achieve downsampling. Specifically, the projected stitched point cloud can be represented in a two-dimensional plane using an occupancy grid, with each grid occupying an area of ​​0.1m × 0.1m. For all points within each grid, their coordinates (x, y) are multiplied by 10, rounded, and then multiplied by 0.1. This downsamples the point cloud within that grid to a single point, thus achieving downsampling. This concludes the method.

[0187] The lane line point cloud stitching method provided in this embodiment calculates the vehicle's position and attitude changes in three-dimensional space using an onboard inertial navigation system, acquires lane line information using an onboard camera, and then uses image timestamps to perform pose linear interpolation to obtain the vehicle pose corresponding to each frame of lane lines. Finally, the lane line point cloud of historical frames is projected onto the current frame to complete the stitching. In other words, multiple consecutive single-frame lane lines are accumulated, and the lane lines of historical frames are stitched onto the current frame through projection transformation. By using the point cloud data from historical frames, the weight of poor-quality point clouds in single frames is reduced, forming sufficiently long and reliable lane lines, thereby reducing the probability of mismatches causing the vehicle's positioning to deviate from its true position.

[0188] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0189] Based on the same inventive concept, one embodiment of the present invention provides a lane line point cloud stitching device 900. (See reference) Figure 9 , Figure 9 This is a structural block diagram of a lane line point cloud stitching device provided in an embodiment of the present invention. Figure 9 As shown, the device 900 includes:

[0190] The point cloud acquisition module 901 is used to acquire the lane line point cloud in real time for each frame.

[0191] Data acquisition module 902 is used to acquire each frame of combined inertial navigation data in real time;

[0192] The first determining module 903 is used to perform trajectory estimation on each frame of combined inertial navigation data, determine the vehicle body pose corresponding to each frame of combined inertial navigation data, and store the vehicle body pose corresponding to each frame of combined inertial navigation data and the corresponding first timestamp into the pose container.

[0193] The second determining module 904 is used to determine the vehicle pose corresponding to the lane line point cloud of the current frame by means of the pose difference based on the second timestamp corresponding to the lane line point cloud of the current frame and the pose container.

[0194] The queue storage module 905 is used to store the current frame lane line point cloud and the vehicle body pose corresponding to the current frame lane line point cloud into a point cloud queue of fixed length. The point cloud queue includes multiple historical frame lane line point clouds and the vehicle body pose corresponding to the multiple historical frame lane line point clouds.

[0195] The stitching module 906 is used to stitch together all lane line point clouds in the point cloud queue to obtain multi-frame stitched lane line point clouds.

[0196] Optionally, the combined inertial navigation data includes at least: vehicle speed and angular acceleration; the first determining module 903 includes:

[0197] The first determining submodule is used to determine the vehicle pose corresponding to the current frame of combined inertial navigation data based on the vehicle speed and angular acceleration in the previous frame of combined inertial navigation data, the time interval between the two frames of combined inertial navigation data, and the vehicle displacement and rotation matrix corresponding to the previous frame of combined inertial navigation data.

[0198] Optionally, the first determining submodule includes:

[0199] The first vehicle pose determination submodule is used to obtain the vehicle pose corresponding to the current frame combined inertial navigation data through the first formula, the second formula and the third formula.

[0200] First formula: p n =p n-1 +R n-1 ·v n-1 ·Δt

[0201] Second formula:

[0202] The third formula:

[0203] Where, p n-1 and p n R represents the vehicle displacement corresponding to the previous frame of combined inertial navigation data and the current frame of combined inertial navigation data, respectively. n-1 and R nThese are the rotation matrices corresponding to the previous frame combined inertial navigation data and the current frame combined inertial navigation data, respectively. n-1 Δt is the vehicle speed in the previous frame of combined inertial navigation data, and Δt is the time interval between the two frames of combined inertial navigation data. It is the angular acceleration T in the previous frame of combined inertial navigation data. n It is the vehicle pose corresponding to the current frame combined inertial navigation data.

[0204] Optionally, the second determining module 904 includes:

[0205] The timestamp determination submodule is used to determine, in the pose container, the first timestamp corresponding to the combined inertial navigation data of the two frames adjacent to the second timestamp, and the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames respectively, based on the second timestamp corresponding to the lane line point cloud of the current frame;

[0206] The factor determination submodule is used to determine the compensation factor based on the second timestamp corresponding to the lane line point cloud of the current frame and the first timestamp corresponding to the combined inertial navigation data of the two preceding and following frames.

[0207] The second vehicle pose determination submodule is used to determine the vehicle pose corresponding to the lane line point cloud in the current frame based on the compensation factor and the vehicle pose corresponding to the first timestamps of the combined inertial navigation data of the previous and next frames.

[0208] Optionally, the factor determination submodule includes:

[0209] The compensation factor determination submodule is used to determine the compensation factor using the fourth formula:

[0210] The fourth formula: s=(t next -t c ) / (t next -t prev )

[0211] The second vehicle body pose determination submodule includes:

[0212] The third vehicle pose determination submodule is used to obtain the vehicle pose corresponding to the lane line point cloud of the current frame through the fifth formula, the sixth formula and the seventh formula.

[0213] The fifth formula: p c =p prev ·s+p next ,(1-s)

[0214] The sixth formula: R c =R next ·R prev ·s

[0215] The seventh formula:

[0216] Where s is the compensation factor, t prev and t next These are the first timestamps corresponding to the two consecutive frames of combined inertial navigation data adjacent to the second timestamp, t c p is the second timestamp corresponding to the lane line point cloud of the current frame. prev and p next p represents the vehicle displacement in the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames, respectively. c R represents the vehicle displacement corresponding to the lane line point cloud in the current frame. prev and R next R represents the rotation matrix in the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames, respectively. c T is the rotation matrix corresponding to the lane line point cloud in the current frame. c The vehicle pose corresponding to the lane line point cloud in the current frame.

[0217] Optionally, the splicing module 906 includes:

[0218] The transformation relationship determination submodule is used to perform matrix operations on the vehicle pose corresponding to the other lane line point cloud in each frame of the point cloud queue and the inverse of the vehicle pose corresponding to the current frame lane line point cloud, respectively, to obtain the coordinate system transformation relationship between the other lane line point cloud in each frame and the current frame lane line point cloud; wherein, the other lane line point cloud is the lane line point cloud in the point cloud queue other than the current frame lane line point cloud;

[0219] The stitching submodule is used to project the other lane line point clouds of each frame onto the vehicle coordinate system of the current frame lane line point cloud through their respective coordinate system transformation relationships, so as to obtain the multi-frame lane line stitched point cloud located in the vehicle coordinate system of the current frame lane line point cloud.

[0220] Optionally, the device 900 further includes:

[0221] The deletion module is used to delete the timestamps before the second timestamp of the lane line point cloud in the current frame and the vehicle pose corresponding to the timestamps from the pose container after determining the vehicle pose corresponding to the lane line point cloud in the current frame by means of the pose difference based on the second timestamp corresponding to the lane line point cloud in the current frame and the pose container.

[0222] Optionally, the device 900 further includes:

[0223] The downsampling processing module is used to perform rasterization and downsampling processing on the multi-frame lane line stitched point cloud to obtain the downsampled multi-frame lane line stitched point cloud.

[0224] Optionally, the downsampling processing module includes:

[0225] The rasterization submodule is used to rasterize the multi-frame lane line stitched point cloud to obtain multiple grids.

[0226] The downsampling processing submodule is used to perform downsampling processing on all point clouds in each of the multiple grids to obtain a single target point cloud for each grid, and to determine the downsampled multi-frame lane line stitching point cloud based on the single target point cloud of each grid.

[0227] Optionally, the point cloud acquisition module 901 includes:

[0228] The data acquisition submodule is used to acquire lane line point cloud data output by the vehicle-mounted forward-looking camera in real time for each frame of lane line point cloud data. The lane line point cloud data includes at least: initial lane line point cloud, lane line start point and lane line length.

[0229] A selection submodule is used to select a lane line point cloud segment at a preset distance from the initial lane line point cloud based on the lane line starting point and the lane line length, and use the lane line point cloud segment at the preset distance as the lane line point cloud for each frame.

[0230] Based on the same inventive concept, another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the lane line point cloud stitching method as described in any of the above embodiments of the present invention.

[0231] Based on the same inventive concept, another embodiment of the present invention provides an electronic device 1000, such as... Figure 10 As shown. Figure 10 This is a schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device includes a memory 1002, a processor 1001, and a computer program stored in the memory and executable on the processor. When executed by the processor, the program implements the steps in the lane line point cloud stitching method described in any of the above embodiments of the present invention.

[0232] Based on the same inventive concept, another embodiment of the present invention provides a vehicle including a positioning system for performing steps in the lane line point cloud stitching method as described in any of the above embodiments of the present invention.

[0233] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0234] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0235] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0236] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0237] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0238] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0239] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the present invention.

[0240] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0241] The present invention provides a detailed description of a lane line point cloud stitching method, apparatus, device, medium, and vehicle. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for stitching lane line point clouds, characterized in that, The method includes: Real-time acquisition of lane line point cloud for each frame; Real-time acquisition of each frame of combined inertial navigation data; For each frame of combined inertial navigation data, a trajectory is calculated to determine the vehicle body pose corresponding to each frame of combined inertial navigation data, and the vehicle body pose and the corresponding first timestamp are stored in the pose container. Based on the second timestamp corresponding to the lane line point cloud of the current frame and the pose container, the vehicle pose corresponding to the lane line point cloud of the current frame is determined by the pose difference. The current frame lane line point cloud and the corresponding vehicle body pose are stored in a fixed-length point cloud queue, which includes multiple historical frame lane line point clouds and the corresponding vehicle body poses. All lane line point clouds in the point cloud queue are stitched together to obtain multi-frame stitched lane line point clouds. The step of determining the vehicle pose corresponding to the lane line point cloud in the current frame based on the second timestamp corresponding to the lane line point cloud in the current frame and the pose container through the pose difference includes: Based on the second timestamp corresponding to the lane line point cloud of the current frame, the first timestamp corresponding to the combined inertial navigation data of the two frames adjacent to the second timestamp is determined in the pose container, and the vehicle pose corresponding to the first timestamp corresponding to the combined inertial navigation data of the two frames is determined respectively. The compensation factor is determined based on the second timestamp corresponding to the lane line point cloud of the current frame and the first timestamp corresponding to the combined inertial navigation data of the two preceding and following frames. Based on the compensation factor and the vehicle pose corresponding to the first timestamps of the combined inertial navigation data of the previous and next frames, the vehicle pose corresponding to the lane line point cloud of the current frame is determined. The step of determining the compensation factor based on the second timestamp corresponding to the lane line point cloud of the current frame and the first timestamp corresponding to the combined inertial navigation data of the two preceding and following frames includes: The compensation factor is determined using the fourth formula: The fourth formula: The step of determining the vehicle pose corresponding to the lane line point cloud in the current frame based on the compensation factor and the vehicle pose corresponding to the first timestamps of the combined inertial navigation data of the preceding and following frames includes: The vehicle pose corresponding to the lane line point cloud in the current frame is obtained using the fifth, sixth, and seventh formulas: The fifth formula: The sixth formula: The seventh formula: Where s is the compensation factor, and These are the first timestamps corresponding to the two consecutive frames of combined inertial navigation data adjacent to the second timestamp. The second timestamp corresponding to the lane line point cloud of the current frame. and These represent the vehicle displacements in the vehicle pose corresponding to the first timestamps of the combined inertial navigation data from the preceding and following frames, respectively. This represents the vehicle displacement corresponding to the lane line point cloud in the current frame. and These are the rotation matrices in the vehicle pose corresponding to the first timestamps of the combined inertial navigation data from the two consecutive frames, respectively. This is the rotation matrix corresponding to the lane line point cloud in the current frame. The vehicle pose corresponding to the lane line point cloud in the current frame.

2. The lane line point cloud stitching method according to claim 1, characterized in that, The combined inertial navigation data includes at least: vehicle speed and angular acceleration; the step of calculating the trajectory of each frame of combined inertial navigation data to determine the vehicle body pose corresponding to each frame of combined inertial navigation data includes: Based on the vehicle speed and angular acceleration in the previous frame of combined inertial navigation data, the time interval between the two frames of combined inertial navigation data, and the vehicle displacement and rotation matrix corresponding to the previous frame of combined inertial navigation data, the vehicle pose corresponding to the current frame of combined inertial navigation data is determined.

3. The lane line point cloud stitching method according to claim 2, characterized in that, The step of determining the vehicle pose corresponding to the current frame of combined inertial navigation data based on the vehicle speed and angular acceleration in the previous frame of combined inertial navigation data, the time interval between the two frames of combined inertial navigation data, and the vehicle displacement and rotation matrix corresponding to the previous frame of combined inertial navigation data includes: The vehicle pose corresponding to the current frame combined inertial navigation data is obtained using the first, second, and third formulas: First formula: Second formula: The third formula: in, and These are the vehicle displacements corresponding to the previous frame of combined inertial navigation data and the current frame of combined inertial navigation data, respectively. and These are the rotation matrices corresponding to the combined inertial navigation data of the previous frame and the combined inertial navigation data of the current frame, respectively. It is the vehicle speed in the previous frame of combined inertial navigation data. t is the time interval between the two frames of combined inertial navigation data. It is the angular acceleration in the previous frame of combined inertial navigation data. It is the vehicle pose corresponding to the current frame combined inertial navigation data.

4. The lane line point cloud stitching method according to claim 1, characterized in that, The step of stitching together all lane line point clouds in the point cloud queue to obtain multi-frame stitched lane line point clouds includes: The vehicle pose corresponding to the other lane line point cloud in each frame of the point cloud queue is subjected to matrix operation with the inverse of the vehicle pose corresponding to the lane line point cloud in the current frame, respectively, to obtain the coordinate system transformation relationship between the other lane line point cloud in each frame and the lane line point cloud in the current frame; wherein, the other lane line point cloud is the lane line point cloud in the point cloud queue other than the lane line point cloud in the current frame. Each of the other lane line point clouds in each frame is projected onto the vehicle coordinate system of the current frame lane line point cloud through its corresponding coordinate system transformation relationship, so as to obtain the multi-frame lane line stitched point cloud located in the vehicle coordinate system of the current frame lane line point cloud.

5. The lane line point cloud stitching method according to claim 1, characterized in that, After determining the vehicle pose corresponding to the lane line point cloud of the current frame through pose difference based on the second timestamp corresponding to the lane line point cloud of the current frame and the pose container, the method further includes: Delete the timestamp before the second timestamp corresponding to the lane line point cloud of the current frame and the vehicle pose corresponding to the timestamp in the pose container.

6. The lane line point cloud stitching method according to claim 1, characterized in that, The method further includes: The multi-frame lane line stitched point cloud is rasterized and downsampled to obtain the downsampled multi-frame lane line stitched point cloud.

7. The lane line point cloud stitching method according to claim 6, characterized in that, The step of rasterizing and downsampling the multi-frame lane line stitched point cloud to obtain a downsampled multi-frame lane line stitched point cloud includes: The multi-frame lane line stitched point cloud is rasterized to obtain multiple grids; Downsampling is performed on all point clouds in each of the multiple grids to obtain a single target point cloud for each grid. The downsampled multi-frame lane line stitching point cloud is then determined based on the single target point cloud of each grid.

8. The lane line point cloud stitching method according to claim 1, characterized in that, The real-time acquisition of lane line point clouds for each frame includes: Real-time acquisition of lane line point cloud data output by the vehicle-mounted forward-looking camera for each frame, wherein the lane line point cloud data includes at least: initial lane line point cloud, lane line start point and lane line length; Based on the lane line starting point and the lane line length, a lane line point cloud segment with a preset distance is selected from the initial lane line point cloud, and the lane line point cloud segment with the preset distance is used as the lane line point cloud for each frame.

9. A lane line point cloud stitching device, characterized in that, The device includes: The point cloud acquisition module is used to acquire lane line point clouds in real time for each frame. The data acquisition module is used to acquire each frame of combined inertial navigation data in real time; The first determining module is used to perform trajectory estimation on each frame of combined inertial navigation data, determine the vehicle body pose corresponding to each frame of combined inertial navigation data, and store the vehicle body pose corresponding to each frame of combined inertial navigation data and the corresponding first timestamp into the pose container. The second determining module is used to determine the vehicle pose corresponding to the lane line point cloud of the current frame by means of the pose difference based on the second timestamp corresponding to the lane line point cloud of the current frame and the pose container. The queue storage module is used to store the current frame lane line point cloud and the vehicle body pose corresponding to the current frame lane line point cloud into a point cloud queue of fixed length. The point cloud queue includes multiple historical frame lane line point clouds and the vehicle body pose corresponding to the multiple historical frame lane line point clouds. The stitching module is used to stitch together all lane line point clouds in the point cloud queue to obtain multi-frame stitched lane line point clouds. The second determining module includes: The timestamp determination submodule is used to determine, in the pose container, the first timestamp corresponding to the combined inertial navigation data of the two frames adjacent to the second timestamp, and the vehicle pose corresponding to the first timestamp of the combined inertial navigation data of the two frames respectively, based on the second timestamp corresponding to the lane line point cloud of the current frame; The factor determination submodule is used to determine the compensation factor based on the second timestamp corresponding to the lane line point cloud of the current frame and the first timestamp corresponding to the combined inertial navigation data of the two preceding and following frames. The second vehicle pose determination submodule is used to determine the vehicle pose corresponding to the lane line point cloud in the current frame based on the compensation factor and the vehicle pose corresponding to the first timestamps of the combined inertial navigation data of the previous and next frames, respectively. The factor determination submodule includes: The compensation factor determination submodule is used to determine the compensation factor using the fourth formula: The fourth formula: The second vehicle body pose determination submodule includes: The third vehicle pose determination submodule is used to obtain the vehicle pose corresponding to the lane line point cloud of the current frame through the fifth formula, the sixth formula and the seventh formula. The fifth formula: The sixth formula: The seventh formula: Where s is the compensation factor, and These are the first timestamps corresponding to the two consecutive frames of combined inertial navigation data adjacent to the second timestamp. The second timestamp corresponding to the lane line point cloud of the current frame. and These represent the vehicle displacements in the vehicle pose corresponding to the first timestamps of the combined inertial navigation data from the preceding and following frames, respectively. This represents the vehicle displacement corresponding to the lane line point cloud in the current frame. and These are the rotation matrices in the vehicle pose corresponding to the first timestamps of the combined inertial navigation data from the two consecutive frames, respectively. This is the rotation matrix corresponding to the lane line point cloud in the current frame. The vehicle pose corresponding to the lane line point cloud in the current frame.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the computer program is executed by the processor, it implements the lane line point cloud stitching method as described in any one of claims 1 to 8.

11. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the lane line point cloud stitching method as described in any one of claims 1 to 8.

12. A vehicle, characterized in that, It includes a positioning system for performing the lane line point cloud stitching method as described in any one of claims 1 to 8.

Images