Computer implementation method for generating lane boundary models for routes traveled by autonomous vehicles.

Abstract

A computer-implemented method for generating a lane boundary model of a path along which an autonomous vehicle travels. The present invention provides a computer-implemented method for generating a lane boundary model of a path along which an autonomous vehicle travels. This computer-implemented method includes steps of acquiring a three-dimensional LiDAR point cloud of a path and an image of the path along which the autonomous vehicle travels, detecting lane boundaries in the image of the path using a machine learning model, and generating a lane boundary model based on a plurality of points of the three-dimensional LiDAR point cloud of the path that are positionally corresponding to the lane boundaries detected at that time.

Description

【Technical Field】 【0001】 The subject matter of the present disclosure relates to a computer-implemented method, a non-transitory computer-readable medium, and an autonomous vehicle for generating a lane boundary model of a path on which the autonomous vehicle travels. 【Background Art】 【0002】 An autonomous vehicle (AV) navigates a path using various sensors. For example, an AV can navigate a path using an image captured by the AV. To appropriately navigate the path, it is important for the AV to know the position of the lane boundaries on the path. 【0003】 A machine learning model can be used to infer the position of lane boundaries in an image. However, the lane boundaries are only useful for that image. It is difficult to transfer the lane boundaries to other lanes of its path or other modalities. 【Summary of the Invention】 【Problems to be Solved by the Invention】 【0004】 The subject matter of the present disclosure aims to improve the prior art by alleviating such problems. 【Means for Solving the Problems】 【0005】 According to an aspect of the present disclosure, there is provided a computer-implemented method for generating a lane boundary model of a path on which an autonomous vehicle travels, the method including: obtaining a 3D LiDAR point cloud of the path and an image of the path on which the autonomous vehicle travels; detecting lane boundaries in the image of the path using a machine learning model; and generating a lane boundary model based on a plurality of points of the 3D LiDAR point cloud of the path that are positionally corresponding to the detected lane boundaries. 【0006】 By generating lane boundaries based on a 3D LiDAR point cloud of the route, the lane boundary model can be easily transferred to other images and other modalities. 【0007】 Training a machine learning model to automatically detect lane boundaries from images of a route traveled by an autonomous vehicle means automatically detecting occlusionless lane boundaries, lane boundaries without occlusions, or lane boundaries with one or more occlusions removed from images of a route traveled by an autonomous vehicle. In this specification, the term “occlusion” may be used to mean any part of a lane boundary that is hidden or obscured and not visible due to another feature on the route. Examples of such features include other vehicles, road construction, and puddles. 【0008】 In this specification, the term “lane boundary” may be used to define the boundary surface between an area on which an AV (Avenue) can travel and an area on which an AV cannot travel. The boundary surface may be structural, such as a curb between a road and a sidewalk / pedestrian walkway (pavement). The boundary surface may be non-structural, such as lane markings on a road. 【0009】 In one embodiment, the step of acquiring a 3D LiDAR point cloud of the route may include the step of capturing a plurality of LiDAR points of the route with the autonomous vehicle, and the step of integrating the plurality of LiDAR points to generate a 3D point cloud of the route. 【0010】 In one embodiment, multiple LiDAR points and images may be paired. 【0011】 In one embodiment, the step of acquiring an image of the route may include the step of capturing an image of the route by the autonomous vehicle. 【0012】 In one embodiment, the step of generating the lane boundary model based on a plurality of points in the 3D LiDAR point cloud of the path that are spatially corresponding to the detected lane boundary may include the steps of: constructing a 3D point cloud of the lane boundary by selecting a plurality of points that are spatially corresponding to the identified lane boundary from the integrated 3D point cloud; clustering the plurality of points in the 3D point cloud of the lane boundary into one or more clusters using the distance between points; and constructing an optimal spline for each cluster as the lane boundary model. 【0013】 In one embodiment, the distance between points may be calculated by determining the distance between each point and its neighboring points, and if the distance between each point is less than a distance threshold, clustering the multiple points into the cluster. 【0014】 In one embodiment, the distance threshold may be weighted according to the direction to each of the adjacent points. 【0015】 In one embodiment, the distance threshold may be weighted to increase in a first direction and decrease in a second direction, the first direction being parallel to the direction of travel of the autonomous vehicle and the second direction being perpendicular to the direction of travel of the autonomous vehicle. 【0016】 In one embodiment, the step of constructing the optimal spline may include the steps of iteratively selecting a random set of points from a plurality of points in the cluster, constructing an optimal spline for each iteratively selected set, calculating the distance from the optimal spline to the points in the set for each set, and selecting the optimal spline with the smallest distance as the optimal spline for the cluster. 【0017】 In one embodiment, the distance may be the total distance. 【0018】 In one embodiment, the distance may be an average distance. 【0019】 In one embodiment, the machine learning model may include a neural network. 【0020】 In one embodiment, the neural network may be a convolutional neural network. 【0021】 According to one aspect of the present disclosure, there is provided a temporary or non-temporary computer-readable medium storing instructions that, when executed by a processor, cause the processor to execute the method according to any one of the preceding claims. 【0022】 According to one aspect of the present disclosure, there is provided an autonomous vehicle including a non-temporary computer-readable medium. BRIEF DESCRIPTION OF THE DRAWINGS 【0023】 The subject matter of the present disclosure is best described with reference to the accompanying drawings. 【0024】 [Figure 1] Shows a schematic diagram of an AV according to one or more embodiments. [Figure 2] Shows a flowchart outlining a computer-implemented method for generating a lane boundary model of a path along which an autonomous vehicle travels according to one or more embodiments, a computer-implemented method for generating training data for training a machine learning model to automatically detect lane boundaries from an image of a path along which an autonomous vehicle travels according to one or more embodiments, and a computer-implemented method for training a machine learning model to automatically detect lane boundaries from an image of a path along which an autonomous vehicle travels according to one or more embodiments. [Figure 3] Shows a 3D LiDAR point cloud of a path captured by the AV of FIG. 1. [Figure 4] Shows the 3D LiDAR point cloud of FIG. 3 overlaid on an image captured by the AV. [Figure 5] Shows a view similar to FIG. 4 of a 3D LiDAR point cloud overlaid on an image captured by the AV (image of a different path from FIG. 4). [Figure 6] Fig. 4 shows a 3D LiDAR point cloud in which points corresponding to the lane boundaries of the image are highlighted. [Figure 7] Fig. 6 shows a diagram similar to Fig. 6 of a path different from the path of Fig. 5. [Figure 8] Fig. 8 shows a 3D LiDAR point cloud of the lane boundary of the path of Fig. 4. [Figure 9] Fig. 8 shows a diagram similar to Fig. 8 of the lane boundary of the path of Fig. 5. [Figure 10] Fig. 14 shows an image representing the generation of a free space model. [Figure 11] Fig. 17 shows a flowchart that details the steps of a computer-implemented method for generating a lane boundary model for a path on which an autonomous vehicle travels according to one or more embodiments. 【Best Mode for Carrying Out the Invention】 【0025】 The embodiments described in this specification are embodied as a set of instructions stored as electronic data on one or more storage media. Specifically, the instructions may be provided on a temporary or non-temporary computer-readable medium. When executed by a processor, the processor is configured to execute various methods described in the following embodiments. Thus, these methods may be computer-implemented methods. In particular, the processor and the storage containing the instructions may be incorporated into a vehicle. The vehicle may be an AV. 【0026】 The following embodiments provide specific exemplary examples, but those exemplary examples should not be construed as limiting, and the scope of protection is defined by the claims. The features of a particular embodiment may be used in combination with the features of other embodiments within the scope that does not extend the subject matter beyond the content of this disclosure. 【0027】 Referring to Figure 1, the AV10 may include a number of sensors 12, 13. Sensor 12 may be mounted on the roof of the AV10. Sensor 13 may be mounted on the front grille of the AV10. Sensors 12, 13 may be communicatively connected to a computer 14. The computer 14 may be mounted on the AV10. The computer 14 may include a processor 16 and memory 18. The memory may include the non-temporary computer-readable medium described above. Alternatively, the non-temporary computer-readable medium may be located remotely and communicatively linked to the computer 14 via the cloud 20. The computer 14 may be communicatively linked to one or more actuators 22 to control the actuators 22 and move the AV10. The actuators may include, for example, motors, brake systems, power steering systems, etc. 【0028】 Sensors 12 and 13 may include various sensor types. Examples of sensor types include LiDAR sensors, RADAR sensors, and cameras. Each sensor type may be called a sensor modality. Each sensor type may record data associated with the sensor modality. For example, a LiDAR sensor may record LiDAR modality data. If sensor 13 is mounted on the front of the AV10, sensor 13 may include a LiDAR sensor so that it can more easily detect features such as lane boundaries behind certain obstructions (e.g., a vehicle) when mounted close to the ground. 【0029】 The data may capture various scenes that the AV10 encounters. For example, the scenes may be visible scenes around the AV10, including roads, buildings, weather, and objects (other vehicles, pedestrians, animals, etc.). 【0030】 Referring to Figure 2, one or more embodiments provide a computer implementation method for training a machine learning model to automatically detect lane boundaries from images of the route on which an AV vehicle travels. As part of this method, a computer implementation method for generating a lane boundary model may be provided. 【0031】 To generate a 3D LiDAR point cloud of a path, the method may include, in step 100, the step of acquiring multiple LiDAR points along the path being traveled by the AV10. Step 100 may also include the step of acquiring one or more images of the same path. Images may be captured by a roof-mounted sensor 12 in the form of one or more cameras. LiDAR points may be captured by a sensor 13 in the form of a LiDAR sensor. The LiDAR sensor 13 mounted on the front of the AV10 allows the LiDAR beam to pass under certain objects, such as parked vehicles that are obstructing the lane boundary. LiDAR points and images may be paired. In other words, the LiDAR points and images captured by the AV10 may be synchronized, as they are linked to the same system clock of the computer 14. 【0032】 Step 102 provides a machine learning model. The machine learning model may include a neural network. The neural network may be a convolutional neural network. The machine learning model may be trained to automatically detect lane boundaries from images of the route that an autonomous vehicle will travel. First, the machine learning model is trained using manually labeled training data of images that have manual labels representing lane boundaries on the route in the images, as described below. 【0033】 Step 104 includes the step of using a machine learning model to identify one or more lane boundaries in one or more images captured in step 100. The lane boundaries are stored as automatically generated labels as shown in 106. 【0034】 Referring briefly to Figure 3, multiple LiDAR points may be integrated to form a 3D LiDAR point cloud of path 200. 【0035】 Referring further to Figure 2, step 108 renders an annotated map 110 by annotating the 3D LiDAR point cloud of route 200 with 108. As described below, the annotated map 110 includes a lane boundary model. 【0036】 Referring to Figure 4, a 3D LiDAR point cloud of lane boundary 202 may be constructed by selecting multiple points from the integrated 3D point cloud of route 200 that are positionally corresponding to the identified lane boundary. Since the images and LiDAR used to derive the lane boundary using a machine learning model are paired, points from the 3D point cloud of the route that are positionally corresponding to the lane boundary can be selected as the 3D LiDAR point cloud of lane boundary 202. Figure 5 shows the same method for constructing 3D point clouds of lane boundary 202 for different routes. 【0037】 Referring to Figure 6, multiple points from the 3D point cloud of lane boundaries are clustered into clusters 202A-D for each specific lane boundary. Since there are four lane boundaries in Figure 6, there are four clusters 202A-D. The step of clustering points into clusters is performed using the point-to-point distance. More specifically, the point-to-point distance may be the distance between each point and its neighbors. If the distance between points is less than a distance threshold, multiple points are clustered into a cluster, and the distance threshold is weighted according to the direction to each neighbor. 【0038】 The distance threshold is weighted to increase in the first direction and decrease in the second direction. The first direction may be parallel to the direction of movement of the autonomous vehicle, and the second direction may be perpendicular to the direction of movement of the autonomous vehicle. 【0039】 For example, select a first point. Identify the points adjacent to the first point. Measure the distance between the first point and each identified adjacent point. The distance can be derived from the positional information of the LiDAR scan. The distance may also be a real-world distance. 【0040】 The distance threshold may be weighted to increase non-linearly. For example, in a plan view, the distance threshold may appear elliptical. For instance, if the first adjacent point is X mm away from the first point in the longitudinal direction of AV's movement, and the second adjacent point is also X mm away from the first point in the transverse direction of AV's movement, then the first adjacent point may be within the boundary of the distance threshold, but the second adjacent point may not be within the boundary of the distance threshold. 【0041】 Figure 7 shows a diagram similar to Figure 6, showing clusters obtained for lane boundaries of different routes. Figure 7 contains five clusters 202A to E. 【0042】 Referring to Figure 8, the spline generation algorithm is used to construct the optimal spline for each cluster. To construct the optimal spline, first, a random set of points is iteratively selected from multiple points within the cluster. 10 3 There may be a certain number of iterations. For example, there may be about 2000 iterations. In each iteration, a different set of points is randomly selected. Then, the optimal spline is constructed for each set. 【0043】 Next, the distance from the optimal spline to each point in the set is calculated. This distance can be the total distance or the average distance. For example, the unit distance between each point in the set and the spline may be calculated. These unit distances may be the smallest distance between each point and the spline. In other words, the point on the spline closest to each individual point is used to measure the unit distance. The total distance is obtained by summing all the unit distances. The average distance is obtained by obtaining the total distance and dividing it by the number of points in the set. The optimal spline with the smallest distance is selected as the optimal splines 204A-D of the cluster. All the optimal splines 204A-D of the image are combined to form a lane boundary model. Figure 9 shows a similar diagram to Figure 8, where the same method is used for different paths. 【0044】 Referring further to Figure 2, as explained above, in 108, the lane boundary model is used to annotate the 3D LiDAR point cloud (map) of the route, forming an annotated map 110. 【0045】 The method may also include a computer implementation method for generating training data to train a machine learning model to automatically detect lane boundaries from images of the route traveled by an autonomous vehicle. 【0046】 In step 112, a lane boundary model is projected onto multiple traversals of the route based on a 3D LiDAR point cloud (map) of the route. Hereinafter, the term “traversal” is used to mean different paths along the same route. While the lane boundary model should be valid for all traversals, certain parts of the lane boundary may be obscured because objects appear along the route in some traversals but not in others. In particular, dynamic objects may differ between traversals. 【0047】 Therefore, in step 114, one or more portions of one or more lane boundaries may be removed to eliminate the obstructing portion of the lane boundary. One or more portions of the lane boundary are automatically removed. 【0048】 Referring to Figure 10, the automatic removal of one or more portions of the lane boundary may include a step of generating a free-space model based on the image using a further machine learning model. The free-space model includes one or more free-space regions and one or more occluded regions. The further machine learning model may include a neural network. The neural network may be a convolutional neural network. 【0049】 Further machine learning models are trained using the initially manually annotated free-space boundary and image pairs. Boundary 208 separates one or more free-space regions 210 from one or more occluded regions 212. Therefore, further machine learning models are trained to generate free-space boundary 208 between one or more free-space regions 210 and one or more occluded regions 212. 【0050】 Referring to Figures 2 and 10, the method may include in step 114 the steps of removing one or more portions of the lane boundary that overlap with one or more occluding areas and retaining one or more portions of the lane boundary that overlap with one or more free space areas. Thus, the method generates an occluding lane boundary in the form of ground truth 116 that includes only one or more portions of the retained one or more lane boundaries. It also generates a training example 118 that includes an image annotated with the occluding lane boundary. In step 120, the raw image of the driving line and the image annotated with the occluding lane boundary may be used as a training pair for training the machine learning model 102. 【0051】 Referring further to Figure 2, in step 122, as described above, the path images may be manually annotated first in order to train the machine learning model 102. 【0052】 Step 124 may include the step of capturing new images by AV10 while driving along a different route. Since point clouds are not needed during estimation, it is not necessary to capture LiDAR data at this stage. 【0053】 In step 126, the machine learning model 102 is used to perform inferences on the new image and identify the new lane boundary. The new lane boundary is shown in 128. 【0054】 In step 130, the performance of the machine learning model 102 is evaluated by comparing the new lane boundary 128 with ground truth 116. Based on the comparison, an accuracy score for the new lane boundary may be determined. More specifically, the accuracy score may be determined in one or more of the following ways: In other words, the accuracy score may be equivalent to any of the precision score, recall score, or F1 score. Based on the comparison, the precision score may be determined. Based on the comparison, the recall score may be determined. Based on the comparison, the F1 score may be determined. 【0055】 The aforementioned scores are calculated using different formulas that utilize the false negative (FN), false positive (FP), true negative (TN), and true positive (TP) results obtained through comparison. These different formulas may be as follows: 【0056】 Accuracy score = (TP + TN) / (TP + FP + FN + TN) Simply put, the accuracy score is the ratio of correctly predicted observations to the total number of observations. 【0057】 Precision score = TP / (TP + FP) Therefore, the precision score is the ratio of correctly predicted positive observations to the total predicted positive observations. 【0058】 Recall score = TP / (TP+FN) Therefore, the recall score is the ratio of correctly predicted positive observations to all actual observations within the class. 【0059】 F1 score = 2 × (Recall score × Precision score) / (Recall score + Precision score) Therefore, the F1 score is a weighted average of the precision score and the recall score. 【0060】 In step 130, if the accuracy score is above the accuracy threshold, the machine learning model 102 is considered to have made an accurate prediction. In other words, the machine learning model is considered accurate. If the accuracy score is below the accuracy threshold, the machine learning model 102 is considered to have not made an accurate prediction. In other words, the machine learning model is considered inaccurate. 【0061】 If the machine learning model 102 is deemed not to have made accurate predictions, step 112 is repeated during further iterations of the method, using a new image as the above image, to project the lane boundary model onto the new image 124. Then, steps 114, 118, and 120 can be performed to improve the accuracy of the machine learning model 102 for the trajectory from which the new image was acquired. In 116, a new ground truth may be generated for the new image. 【0062】 To summarize the above, referring to Figure 11, a computer implementation method for generating a lane boundary model of a route traveled by an autonomous vehicle is provided, which includes the steps of: acquiring a 3D LiDAR point cloud of the route and an image of the route traveled by the autonomous vehicle (S300); detecting lane boundaries in the image of the route using a machine learning model (S302); and generating a lane boundary model based on a plurality of points in the 3D LiDAR point cloud of the route that are positionally corresponding to the detected lane boundaries (S304). 【0063】 While the embodiments described above are provided to illustrate the subject matter of this disclosure, the features of the embodiments should not be construed as limiting the scope of protection. To avoid doubt, the scope of protection is defined by the appended claims. 【0064】 The following includes one or more items that provide information related to the subject matter of this disclosure. 【0065】 item Item 1. A computer implementation method for generating training data to train a machine learning model to automatically detect lane boundaries from images of a route traveled by an autonomous vehicle, comprising: acquiring images captured by the autonomous vehicle while traveling along the route; projecting a lane boundary model onto the images based on a 3D LiDAR point cloud of the route; and generating example training data for training the machine learning model by automatically removing one or more portions of the lane boundaries corresponding to one or more occluded portions in the images. 【0066】 Item 2. The computer implementation method according to Item 1, wherein the step of automatically removing one or more portions of a lane boundary includes the steps of: generating a free-space model based on the image, which includes one or more free-space regions and one or more occluded regions, using a further machine learning model; deleting one or more portions of the lane boundary that overlap with one or more of the occluded regions; and retaining one or more portions of the lane boundary that overlap with one or more of the free-space regions. 【0067】 Item 3. The computer implementation method according to Item 2, further comprising the step of generating ground truth including one or more retained portions of the lane boundary. 【0068】 Item 4. The further machine learning model includes a neural network, as described in item 2 or 3, and is a computer implementation method. 【0069】 Item 5. The computer implementation method described in Item 4, wherein the neural network is a convolutional neural network. 【0070】 Item 6. A computer implementation method according to any one of items 1 to 5, further comprising the steps of: acquiring new images captured by an autonomous vehicle while traveling along different routes; and determining new lane boundaries by performing inference on the new images using a machine learning model. 【0071】 Item 7. A computer implementation method according to Item 6, as dependent on Item 3, further comprising the steps of comparing the new lane boundary with the ground truth, and determining the accuracy score of the new lane boundary based on the comparison. 【0072】 Item 8. The computer implementation method described in Item 7, wherein the accuracy score is determined as one or more of the following: a comparative precision score, and / or a comparative recall score, and / or a comparative F1 score. 【0073】 Item 9. The computer implementation method according to item 7 or 8, further comprising the steps of determining that the machine learning model is accurate if the accuracy score is equal to or greater than an accuracy threshold, and determining that the machine learning model is inaccurate if the accuracy score is less than the accuracy threshold. 【0074】 Item 10. The computer implementation method according to claim 9, further comprising the steps of, if the machine learning model is determined to be inaccurate, projecting a lane boundary model onto an image based on a 3D LiDAR point cloud of the path, and generating training data examples for training the machine learning model by automatically removing one or more portions of the lane boundary corresponding to one or more occluding portions in the image, repeating these steps using a new image as the image. 【0075】 Item 11. A computer implementation method according to any one of items 1 to 10, further comprising the steps of: acquiring a plurality of LiDAR points along the route by the autonomous vehicle; and integrating the plurality of LiDAR points to generate a three-dimensional point cloud of the route. 【0076】 Item 12. The computer implementation method described in Item 11, wherein multiple LiDAR points and the images are paired. 【0077】 Item 13. A computer implementation method according to Item 11 or 12, further comprising: identifying lane boundaries from captured images using a machine learning algorithm; constructing a 3D point cloud of lane boundaries by selecting a plurality of points that are positionally corresponding to the identified lane boundaries from an integrated 3D point cloud; clustering the plurality of points of the 3D point cloud of lane boundaries into one or more clusters using the distance between points; and constructing an optimal spline for each cluster as the lane boundary model. 【0078】 Item 14. The computer implementation method described in Item 13, wherein the distance between points is calculated by determining the distance between each point and its adjacent points, and if the distance between each point is less than a distance threshold, clustering the multiple points into the cluster. 【0079】 Item 15. The computer implementation method according to Item 134, wherein the distance threshold is weighted according to the direction to each adjacent point. 【0080】 Item 16. The computer implementation method according to Item 15, wherein the distance threshold is weighted to increase in a first direction and decrease in a second direction, the first direction being parallel to the direction of movement of the autonomous vehicle and the second direction being perpendicular to the direction of movement of the autonomous vehicle. 【0081】 Item 17. A computer implementation method according to any one of items 13 to 16, wherein the step of constructing the optimal spline includes the steps of iteratively selecting a random set of points from a plurality of points in the cluster, constructing an optimal spline for each iteratively selected set, calculating the distance from the optimal spline to the points in the set for each set, and selecting the optimal spline with the smallest distance as the optimal spline for the cluster. 【0082】 Item 18. The computer implementation method described in Item 17, wherein the distance is the total distance. 【0083】 Item 19. The computer implementation method described in Item 17, wherein the distance is the average distance. 【0084】 Item 20. The machine learning model includes a neural network, and is implemented using a computer method as described in any one of items 1 to 19. 【0085】 Item 21. The computer implementation method described in Item 20, wherein the neural network is a convolutional neural network. 【0086】 Item 22. A computer implementation method for training a machine learning model to automatically detect lane boundaries from images of a route traveled by an autonomous vehicle, comprising the steps of: obtaining a plurality of training data examples of route lines in which lane boundaries without obstructions are identified; and training a machine learning model to automatically detect lane boundaries from images of a route traveled by an autonomous vehicle. 【0087】 Item 23. The step of obtaining multiple training data examples is the computer implementation method described in Item 22, which includes the computer implementation method described in any one of Items 1 to 20. 【0088】 Item 24. A temporary or non-temporary computer-readable medium that, when executed by a processor, stores instructions causing the processor to perform any of the methods described in the preceding items. 【0089】 Item 25. Autonomous vehicles, including non-temporary computer-readable media as defined in Item 24.

Claims

[Claim 1] A computer implementation method for generating a lane boundary model for a route traveled by an autonomous vehicle, The steps include acquiring a 3D LiDAR point cloud of the route and an image of the route traveled by the autonomous vehicle, A step of detecting lane boundaries in the image of the route using a machine learning model, The steps include generating a lane boundary model based on a plurality of points in the 3D LiDAR point cloud of the path that are positionally corresponding to the detected lane boundary, and Includes, The step of generating the lane boundary model based on a plurality of points in the 3D LiDAR point cloud of the path that is spatially corresponding to the detected lane boundary is: The steps include constructing a 3D point cloud of the lane boundary by selecting a plurality of points that are positionally corresponding to the identified lane boundary from the integrated 3D LiDAR point cloud, The steps include: clustering multiple points of the three-dimensional point cloud of the lane boundary into one or more clusters using the distance between points; The steps include constructing the optimal spline for each cluster as the lane boundary model, and Computer implementation methods, including those mentioned above. [Claim 2] The step of acquiring a 3D LiDAR point cloud of the aforementioned path is: The autonomous vehicle captures multiple LiDAR points along the route, The steps include: integrating multiple LiDAR points to generate a three-dimensional point cloud of the path; The computer implementation method according to claim 1, including the method described in claim 1. [Claim 3] The computer implementation method according to claim 2, wherein a plurality of LiDAR points and the image are paired. [Claim 4] The computer implementation method according to claim 1, wherein the step of acquiring an image of the route includes the step of capturing an image of the route by the autonomous vehicle. [Claim 5] The distance between the aforementioned points is, Determine the distance between each point and its adjacent points, If the distance between each point is less than the distance threshold, the multiple points are clustered into the aforementioned cluster. The computer implementation method according to claim 1, calculated by the method described above. [Claim 6] The computer implementation method according to claim 5, wherein the distance threshold is weighted according to the direction to each adjacent point. [Claim 7] The computer implementation method according to claim 6, wherein the distance threshold is weighted to increase in a first direction and decrease in a second direction, the first direction being parallel to the direction of movement of the autonomous vehicle and the second direction being perpendicular to the direction of movement of the autonomous vehicle. [Claim 8] The step of constructing the aforementioned optimal spline is: The steps include: iteratively selecting a random set of points from multiple points within the aforementioned cluster; The steps include constructing the optimal spline for each set selected iteratively, For each set, the steps include calculating the distance from the optimal spline to a point within the set, The steps include selecting the optimal spline with the smallest distance as the optimal spline for the cluster, and The computer implementation method according to claim 1, including the method described in claim 1. [Claim 9] The computer implementation method according to claim 8, wherein the distance is the total distance. [Claim 10] The computer implementation method according to claim 8, wherein the distance is the average distance. [Claim 11] The computer implementation method according to claim 1, wherein the machine learning model includes a neural network. [Claim 12] The computer implementation method according to claim 11, wherein the neural network is a convolutional neural network. [Claim 13] A temporary or non-temporary computer-readable medium, which, when executed by a processor, stores instructions causing the processor to perform the method according to claim 1. [Claim 14] An autonomous vehicle comprising a temporary computer-readable medium as described in claim 13.

Images