A multi-frame lane line point cloud rasterization method, device, equipment and medium

Abstract

The application provides a multi-frame lane line point cloud rasterization method, device, equipment and medium, and the method comprises the following steps: acquiring a visual point cloud at a current time; converting the visual point cloud at a historical time to the current time; rasterizing and removing redundant visual points; filtering out the same number of visual point clouds by fixing the number of grids, and removing the redundant visual points at different times. The application splices the lane line point cloud segmented by the vehicle-mounted camera semantics in multiple frames, accumulates and converts the historical visual information to the current time, and simplifies the redundant point cloud by using the occupancy grid, thereby reducing the weight of the error point cloud of a certain frame, ensuring that there is enough visual lane line point cloud at any time, removing the redundant visual points at different times, and making the positioning accuracy higher and the robustness stronger.

Description

Technical Field

[0001] This invention pertains to automotive autonomous driving technology, specifically high-precision positioning technology for automotive autonomous driving. Background Technology

[0002] Lane line information is one of the most crucial pieces of information for achieving high-precision positioning in autonomous driving. Matching lane lines identified by onboard cameras with lane lines on high-precision maps enables accurate vehicle positioning. Onboard cameras offer advantages such as strong real-time perception and cost-effectiveness, but are susceptible to factors like lighting, weather, and unclear road surfaces. Therefore, using lane lines identified by onboard cameras, combined with existing lane line recognition algorithms, can easily lead to missed or false detections in complex road conditions or severe weather, resulting in inaccurate vehicle positioning.

[0003] To address the above shortcomings, researchers in this field have developed some improved technical solutions. For example, one literature reports a lane line fusion method based on intelligent cameras and high-precision map positioning. This method uses an onboard camera to obtain lane line information through semantic segmentation and fits it into a cubic curve. The cubic curve is then discretized into a point cloud at certain intervals, and subsequently matched with the corresponding map point cloud to calibrate the current vehicle position coordinates. However, this method only uses lane line information acquired by the camera at the current moment; when false positives or missed detections occur, the positioning will suffer from significant deviations.

[0004] Another example is a point cloud rasterization method proposed in the literature, which establishes a raster based on the point cloud coordinates and calculates the coordinates of the raster center point. Within each raster, the center point of the current raster approximates all points within that raster. This method, by approximating all points within the raster using the center point, results in a certain degree of reduction in the accuracy of the coordinate points.

[0005] It is evident that while the above technologies have made some progress, the current recognition algorithms based on semantic segmentation of visual cameras are greatly affected by weather, lighting, and other factors, and the accuracy cannot reach 100%. Therefore, using single lane line recognition results for localization results in the disadvantage of weak robustness of the localization results. Furthermore, there are issues such as the inability to balance the quantity, quality, and accuracy of visual point cloud processing, indicating that there is still significant room for improvement. Summary of the Invention

[0006] To address the problems existing in the prior art and improve the accuracy and robustness of vehicle positioning, this invention proposes a multi-frame lane line point cloud rasterization method, device, equipment, and medium. The method involves stitching together multiple frames of the lane line point cloud semantically segmented by the vehicle camera, accumulating and transforming historical visual information to the current time, and using an occupied grid to simplify redundant point clouds. This point cloud will be used for vehicle positioning in high-precision maps.

[0007] The technical solution of the present invention is as follows:

[0008] In a first aspect, the present invention provides a method for rasterizing multi-frame lane line point clouds, the method comprising: first acquiring the visual point cloud at the current moment; then converting the visual point cloud at a historical moment to the current moment; and finally rasterizing to remove redundant visual points, thereby filtering out the same number of visual point clouds by using a fixed number of raster cells, thus removing redundant visual points at different moments.

[0009] Further preferred, the rasterization process for removing redundant visual points includes reading the visual point cloud coordinates, calculating the raster index point coordinates, and establishing a raster based on the coordinates; if raster index points with the same coordinates appear, it indicates that there are multiple visual points in the raster, and only the coordinates of the visual point closest to the vehicle are retained, while the rest with the same values ​​are removed, so that each raster retains only one visual point.

[0010] Further preferred, the step of selecting the same number of visual point clouds by fixing the number of grids involves counting the number of grids. When the number of grids exceeds a given limit, visual points that are more than a preset value away from the vehicle are discarded, i.e., a certain number of forward-looking points are selected according to the distance. The preset value should not be greater than the diameter of the maximum curvature of the target lane in the actual use scenario.

[0011] In a further preferred embodiment, the given value is set according to the distance from the visual point to the vehicle body in ascending order of the number of grids. By fixing the number of grids, the number of visual point clouds is adjusted, thereby adjusting the number of iterations in the matching process.

[0012] Further optimization yields the following formula for calculating the coordinates of the raster index point:

[0013]

[0014] Where Grid_x, Grid_y, and Grid_z represent the grid index points of the visual point on the x-axis, y-axis, and z-axis, respectively, and scale represents the scale factor of the grid size.

[0015] Further optimization also includes calculating the distance from the grid index point to the origin of the coordinate system, i.e., the center of the vehicle body:

[0016] distance=Grid_x*Grid_x+Grid_y*Grid_y+Grid_z*Grid_z.

[0017] Further preferred, acquiring the visual point cloud at the current moment involves discretizing a curve segment within a predetermined distance region in front of the vehicle at the current moment into a point cloud. This predetermined distance region is chosen because the lane line fitting based on semantic segmentation may have significant errors compared to the actual situation at closer and farther distances. Since visual cameras are typically mounted on the windshield, and closer lane lines are beyond their visible range, a near-distance threshold of 3m is selected. Because lane lines fitted by semantic segmentation at farther distances may have significant errors compared to the actual situation, a maximum distance of 10m from the vehicle is selected as the maximum distance between the visual point and the vehicle. This predetermined distance region is set within a range of 3 to 10 meters based on the current capabilities of visual perception algorithms to ensure accuracy. Therefore, since the visual information obtained from camera semantic segmentation cannot guarantee accuracy, and the effect of cubic curve fitting of lane lines at near and far ends is unsatisfactory, it is preferable to use a curve segment 3m to 10m in front of the vehicle at the current moment as the predetermined distance region.

[0018] Further optimized, the step of transforming the visual point cloud from historical moments to the current moment involves calculating the vehicle's transformation relationship in the coordinate system based on the vehicle's wheel speed, heading angle, instantaneous angular velocity, and instantaneous linear velocity. This process saves the vehicle's pose corresponding to the visual point cloud at each moment, thereby projecting the point cloud from historical moments onto the current moment. This step accumulates and transforms historical visual information to the current time, thus reducing the weight of erroneous point clouds caused by camera misses or false detections at a particular moment.

[0019] Further optimization yields the following projection formula for projecting historical point clouds onto the current moment:

[0020]

[0021] Among them, P t P represents the point cloud coordinates at time t. t-1 Represents the point cloud coordinates at time t-1. Let be the transformation matrix of the k-th visual point from the vehicle coordinate system to the world coordinate system at time t.

[0022] In a second aspect, the present invention provides a multi-frame lane line point cloud rasterization device, comprising:

[0023] The acquisition module is used to acquire the visual point cloud at the current moment.

[0024] The conversion module is used to convert the visual point cloud from a historical moment to the current moment.

[0025] The rasterization module is used to rasterize and remove redundant visual points. By using a fixed number of raster cells, it selects the same number of visual point clouds and removes redundant visual points at different times.

[0026] In a third aspect, the present invention also provides an electronic device, comprising: a processor; and a memory storing a program, wherein the program includes instructions that, when executed by the processor, cause the processor to perform the multi-frame lane line point cloud rasterization method described above.

[0027] In a fourth aspect, the present invention also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the multi-frame lane line point cloud rasterization method described above.

[0028] The advantages of this invention are as follows:

[0029] 1. This invention effectively avoids the large errors between the cubic curves of lane lines that are too close or too far from the vehicle and the actual situation due to insufficient fitting, by taking only the discrete point cloud of the visual lane line in a set distance area (e.g., 3-10m) in front of the vehicle at the current moment. This would affect the matching effect between the visual point cloud and the high-precision map point cloud, and thus affect vehicle positioning.

[0030] 2. This invention projects point clouds from historical moments onto the current moment, which not only reduces the matching weight of a single moment but also makes better use of historical information throughout the entire journey, resulting in higher positioning accuracy and stronger robustness.

[0031] 3. This invention simplifies the grid while preserving the original visual point information. Before simplification, each grid may contain several point clouds; after simplification, each grid retains only one point. This ensures that the quantity is maintained without sacrificing the quality and accuracy of the visual point cloud. Furthermore, a fixed number of grids guarantees sufficient visual points for matching regardless of vehicle operating conditions. Simultaneously, adjusting the number of grids allows for the management of computational resources.

[0032] Through practical verification, the robustness and accuracy of positioning can be guaranteed under various road conditions and weather environments. Furthermore, the number of visual point clouds can be adjusted by fixing the number of grids, thereby adjusting the number of iterations in the matching process and thus optimizing CPU utilization. Attached Figure Description

[0033] Figure 1 A flowchart illustrating a specific embodiment of the multi-frame lane line point cloud rasterization method of the present invention;

[0034] Figure 2 The visual lane line point cloud is obtained from semantic segmentation at the current moment in the vehicle coordinate system.

[0035] Figure 3 This is a visual point cloud representation of projecting a historical moment onto the current moment.

[0036] Figure 4 This is a schematic diagram of the rasterization.

[0037] Figure 5 This is a rasterized point cloud representation of the viewpoint's position. Compared to... Figure 3 Visual point clouds from direct projection, Figure 4 Overlapping redundant points were removed while maintaining accuracy and precision.

[0038] Figure 6 This is a schematic diagram of the electronic device described in this invention. Detailed Implementation

[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. It should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. These embodiments are provided to provide a more thorough and complete understanding of the present invention. The accompanying drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Based on the embodiments of the present invention, technical solutions obtained by those skilled in the art without creative effort are all within the scope of protection of the present invention.

[0040] See Figure 1 The specific implementation of the present invention consists of the following three steps:

[0041] 1. Obtain the current visual discrete point cloud, see [link / reference] Figure 2 :

[0042] Visual lane lines are represented by cubic curves: y = C0 + C1x + C2x 2 +C3x 3 The positive x-axis is directly in front of the vehicle, and the positive y-axis is to the right. Starting 3m in front of the vehicle, the curve is discretized into a point cloud P(ax, ay, az) every 0.5m. The initial vehicle pose is set as the identity matrix: T0. The pose change ΔT at the next moment is calculated using wheel speed, angular velocity, and acceleration, thus obtaining the pose T1 at the next moment. Similarly, the pose Tt at that moment is obtained. If the number of visual point clouds at the current moment is 0, the pose at that moment also needs to be saved separately so that the historical point cloud can be projected onto the current moment later.

[0043] 2. Project the historical point cloud onto the current moment:

[0044] The visual point cloud at each moment is saved, corresponding to the vehicle's pose at that moment. Then, the vehicle's transformation in the coordinate system is calculated using the onboard inertial navigation system, thus projecting the historical point cloud onto the current moment. The projection formula is:

[0045]

[0046] Among them, P t P represents the point cloud coordinates at time t. t-1 Let T represent the point cloud coordinates at time t-1, and T be the transformation matrix at a certain time. The point cloud at time t-1 is projected to time t using the above formula.

[0047] from Figure 3 As can be seen, the point cloud after projection has many overlapping and redundant visual points. We need to perform occupancy rasterization to remove the redundant points.

[0048] 3. Rasterization downsampling, which involves rasterizing to remove redundant visual points.

[0049] After completing steps 1 and 2 above, the number of point clouds will be very large, and many overlapping visual points with almost equal coordinates will appear within a very small coordinate system. This not only increases the computation time and affects the efficiency of the algorithm, but also increases the matching weight in a certain range. Therefore, we use a grid-based approach to remove redundant visual points with similar coordinates.

[0050] The specific implementation method is as follows:

[0051] Iterate through all viewpoints P in sequence, and calculate the corresponding grid index point (grid_point) based on the coordinates of each viewpoint. The calculation formula is as follows:

[0052]

[0053] Where Grid_x, Grid_y, and Grid_z represent the raster index points of the visual point on the x, y, and z axes, respectively. `scale` represents the scale factor of the raster size, which is set to 0.1 here. A raster corresponding to the current visual point cloud is constructed based on the index points, with a size of 0.1m * 0.1m. Simultaneously, the distance from the raster index point to the origin of the coordinate system (i.e., the center of the vehicle body) is calculated.

[0054] distance=Grid_x*Grid_x+Grid_y*Grid_y+Grid_z*Grid_z.

[0055] In summary, by creating a raster at the point cloud, index points will be calculated and a raster will be created only where there are point clouds.

[0056] Before raster simplification of redundant point clouds, each raster may contain several point clouds. After simplification, each raster retains only one point.

[0057] 4. Select the same number of visual point clouds by fixing the number of grid cells.

[0058] Select a certain number of foresight points according to distance.

[0059] Finally, the coordinates of the point and the calculated grid index points are stored in ascending order of distance. If the same distance is found, it is determined whether the corresponding grid index point coordinates are all the same. If they are the same, it indicates that there are multiple points within the grid; only one point is retained, and the rest are discarded. The number of grids is counted; when the number of grids exceeds a certain value, points that are farther away are discarded. That is, only the number of grids is counted, and they are sorted by distance, retaining a certain number of point clouds. Finally, the points within the grid are used as the visual point cloud at the current moment for calculating high-precision vehicle positioning.

[0060] Here, CPU usage is adjusted by setting the number of grid cells. For example, if you want to retain 100 grid cells, the processed point cloud will contain 100 points. These 100 points will then be used for map matching and localization, and the CPU usage might be 10%. If set to 1000, the usage might be 60%. For practical use, a setting of 300 might be suitable.

[0061] Here, discarding points that are too far away means discarding visual points that are farther from the vehicle than a preset value. This preset value should not be greater than the diameter of the target lane's maximum curvature in the actual usage scenario. For example, when the actual usage scenario is a highway ramp, it is set to 50 meters to avoid exceeding the diameter.

[0062] Figure 4 This is a rasterization diagram. The coordinates of the raster index point are calculated based on the coordinates of the viewpoint, and a raster is set with a grid size of 0.1m x 0.1m. When multiple viewpoint clouds exist within the raster, only one point is selected.

[0063] See the point cloud effect after rasterization where the visual point occupies the rasterized area. Figure 5 Compared to Figure 3 The visual point cloud projected directly shows that overlapping redundant points have been removed while maintaining accuracy and precision.

[0064] As can be seen, by utilizing historical frames and filtering a certain number of visual point clouds with a fixed number of grids, this invention not only reduces the weight of erroneous point clouds in a certain frame, ensuring that there are enough visual lane line point clouds at any time, but also eliminates redundant visual points at different times.

[0065] The following embodiment is an implementation of a multi-frame lane line point cloud rasterization device, which includes:

[0066] The acquisition module is used to acquire the visual point cloud at the current moment. Specifically, it takes the curve segment from 3m to 10m in front of the vehicle body at the current moment and discretizes it into a point cloud.

[0067] The transformation module is used to transform the visual point cloud from historical moments to the current moment. Specifically, it calculates the transformation relationship of the vehicle in the coordinate system through the onboard inertial navigation system, saves the vehicle pose corresponding to the visual point cloud at each moment, and then projects the point cloud from the historical moment to the current moment. The projection relationship is described in the above embodiments.

[0068] The rasterization module is used to rasterize and remove redundant visual points. It filters out visual point clouds with a fixed number of raster cells, eliminating redundant visual points at different times. The redundant visual point removal process includes: reading the visual point cloud coordinates, calculating the raster index point coordinates, and creating a raster based on the coordinates. If raster index points with the same coordinates appear, only the visual point coordinates closest to the vehicle are retained, and the rest with the same values ​​are removed. The rasterization module filters out a certain number of visual point clouds with a fixed number of raster cells. This involves counting the number of raster cells; when the number of raster cells exceeds a given limit, visual points farther from the vehicle are discarded. The given limit is a certain number of raster cells retained based on the increasing distance between the visual points and the vehicle. By setting a fixed number of raster cells, the number of visual point clouds is adjusted, thereby adjusting the number of iterations in the matching process.

[0069] The electronic device proposed in this invention includes: a processor; and a memory storing a program, wherein the program includes instructions that, when executed by the processor, cause the processor to perform the multi-frame lane line point cloud rasterization method described above. Figure 6 As shown, the electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102.

[0070] The structure of the electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0071] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0072] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the figure, but this does not mean that there is only one bus or one type of bus.

[0073] The memory 103 may be a ROM or other type of static storage device capable of storing static information and instructions, RAM or other type of dynamic storage device capable of storing information and instructions, or it may be an EEPROM, CD-ROM or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0074] The applicant of this invention has provided a detailed description of the embodiments of the invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred embodiments of the invention. The detailed description is only to help better understand the spirit of the invention and is not intended to limit the scope of protection of the invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, but without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

[0075] Based on the embodiments of the present invention described above, and through the above description, those skilled in the art can make various changes and modifications without departing from the technical concept of the present invention. The scope of the present invention is not limited to the specific embodiments described, but should be determined according to the scope of the claims.

Claims

1. A method for rasterizing multi-frame lane line point clouds, characterized in that, The method includes: Obtain the visual point cloud at the current moment; Transforming the visual point cloud from a historical moment to the current moment involves calculating the vehicle's transformation relationship in the coordinate system based on the vehicle's wheel speed, heading angle, instantaneous angular velocity, and instantaneous linear velocity, saving the vehicle pose corresponding to the visual point cloud at each moment, and thus projecting the point cloud from the historical moment to the current moment. Rasterization downsampling, also known as rasterization, removes redundant visual points at different times. This includes reading the visual point cloud coordinates, calculating the raster index point coordinates, and building a raster based on the coordinates. If raster index points with the same coordinates are found, only the visual point coordinates closest to the vehicle are retained, and the rest with the same values ​​are removed. Each raster retains one visual point. The same number of visual point clouds are selected by filtering out a fixed number of grids. Specifically, the number of grids is counted. When the number of grids is greater than a given limit, visual points that are more than a preset value away from the vehicle are discarded. That is, the forward viewpoint is selected according to the distance. The preset value should not be greater than the diameter of the maximum curvature of the target lane in the actual use scenario. The given limit is to retain a set number of grid points in ascending order of distance between the visual point and the vehicle body. By setting a fixed number of grid points, the number of visual point clouds is adjusted, thereby adjusting the number of iterations in the matching process.

2. The multi-frame lane line point cloud rasterization method according to claim 1, characterized in that, The formula for calculating the coordinates of a raster index point is as follows: Where Grid_x, Grid_y, and Grid_z represent the grid index points of the visual point on the x-axis, y-axis, and z-axis, respectively, and scale represents the scale factor of the grid size.

3. The multi-frame lane line point cloud rasterization method according to claim 2, characterized in that, This also includes calculating the distance from the grid index point to the origin of the coordinate system, i.e., the center of the vehicle body: distance=Grid_x * Grid_x+Grid_y * Grid_y+Grid_z * Grid_z.

4. The multi-frame lane line point cloud rasterization method according to claim 1, characterized in that, The process of obtaining the visual point cloud at the current moment involves taking a curve segment of a predetermined distance region in front of the vehicle body at the current moment and discretizing it into a point cloud. The predetermined distance region is set based on the capabilities of the visual perception algorithm used.

5. The multi-frame lane line point cloud rasterization method according to claim 4, characterized in that, The projection formula for projecting historical point clouds onto the current moment is: = * * in, This represents the point cloud coordinates at time t. This represents the point cloud coordinates at time t-1. This represents the k-th visual point at time t. Let be the transformation matrix of the k-th visual point from the vehicle coordinate system to the world coordinate system at time t.

6. A multi-frame lane line point cloud rasterization device, characterized in that, include: The acquisition module is used to acquire the visual point cloud at the current moment; The conversion module is used to convert the visual point cloud of the historical moment to the current moment. Specifically, it calculates the transformation relationship of the vehicle in the coordinate system by the vehicle's wheel speed, heading angle, instantaneous angular velocity and instantaneous linear velocity, saves the vehicle pose corresponding to the visual point cloud at each moment, and thus projects the point cloud of the historical moment to the current moment. The rasterization module is used to rasterize and remove redundant visual points. It includes reading the visual point cloud coordinates, calculating the raster index point coordinates, and building a raster based on the coordinates. If raster index points with the same coordinates are found, only the visual point coordinates closest to the vehicle are retained, and the rest with the same values ​​are removed. Each raster retains one visual point. By filtering out the same number of visual point clouds with a fixed number of grids, redundant visual points at different times are eliminated. Specifically, the number of grids is counted, and when the number of grids exceeds a given limit, visual points that are more than a preset value away from the vehicle are discarded, i.e., the forward viewpoint is selected based on distance. The preset value should not exceed the diameter of the maximum curvature of the target lane in the actual use scenario. The given limit is the number of grid points retained according to the order of distance between the visual point and the vehicle from small to large. By setting a fixed number of grids, the number of visual point clouds is adjusted, thereby adjusting the number of iterations in the matching process.

7. The multi-frame lane line point cloud rasterization device according to claim 6, characterized in that, The given limit is to retain a set number of grid points in ascending order of distance between the visual point and the vehicle body. By setting a fixed number of grid points, the number of visual point clouds is adjusted, thereby adjusting the number of iterations in the matching process.

8. The multi-frame lane line point cloud rasterization device according to claim 7, characterized in that, The formula for calculating the coordinates of a raster index point is as follows: Where Grid_x, Grid_y, and Grid_z represent the grid index points of the visual point on the x-axis, y-axis, and z-axis, respectively, and scale represents the scale factor of the grid size.

9. The multi-frame lane line point cloud rasterization device according to claim 8, characterized in that, This also includes calculating the distance from the grid index point to the origin of the coordinate system, i.e., the center of the vehicle body: distance=Grid_x * Grid_x+Grid_y * Grid_y+Grid_z * Grid_z.

10. The multi-frame lane line point cloud rasterization device according to any one of claims 6-9, characterized in that, The acquisition module acquires the visual point cloud at the current moment by taking the curve segment of the area in front of the vehicle body at a set distance at the current moment, discretizing it into a point cloud, and the set distance area is set according to the visual perception algorithm capability used.

11. The multi-frame lane line point cloud rasterization device according to any one of claims 6-9, characterized in that, The conversion module calculates the transformation relationship of the vehicle in the coordinate system through the vehicle-mounted integrated inertial navigation system, saves the vehicle pose corresponding to the visual point cloud at each moment, and then projects the point cloud of the historical moment to the current moment.

12. The multi-frame lane line point cloud rasterization device according to claim 11, characterized in that, The projection formula for projecting historical point clouds onto the current moment is: = * * in, This represents the point cloud coordinates at time t. This represents the point cloud coordinates at time t-1. Let be the transformation matrix of the k-th visual point from the vehicle coordinate system to the world coordinate system at time t.

13. An electronic device, comprising: processor; And a memory for storing programs, characterized in that the programs include instructions that, when executed by the processor, cause the processor to perform the multi-frame lane line point cloud rasterization method according to any one of claims 1-5.

14. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, in, The computer instructions are used to cause the computer to execute the multi-frame lane line point cloud rasterization method according to any one of claims 1-5.

Images