Lane line recognition method and device
By dividing lane lines into segments and using initial feature vector clustering and directional attribute classification, the problem of low recognition accuracy of redundant and noisy lane lines in high-precision maps is solved, and efficient abnormal lane line recognition is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- AUTONAVI SOFTWARE CO LTD
- Filing Date
- 2023-07-24
- Publication Date
- 2026-05-29
AI Technical Summary
Existing technologies rely on manually defined rules to identify redundant and noisy lane lines in high-precision maps, resulting in low accuracy and low automation efficiency.
The lane lines are divided into multiple segments, clustered using initial feature vectors that do not reflect direction, and then classified according to direction attributes to identify abnormal lane lines.
It effectively and accurately identifies abnormal lane markings, improving recognition accuracy and enhancing automation efficiency.
Smart Images
Figure CN116843936B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of map technology, and in particular to a lane line recognition method and apparatus. Background Technology
[0002] During the production of high-precision map data, a large number of redundant and noisy lane lines may be generated. For example, multiple lane lines may be output for the same data collection object, and the directions of the multiple lane lines may be inconsistent.
[0003] To ensure the accuracy of high-precision map data, it is necessary to identify and deduplicate redundant and noisy lane lines. Currently, in the existing technology, when identifying redundant and noisy lane lines, staff usually manually define an identification strategy based on the characteristics of lane line noise, and then analyze each lane line according to the identification strategy to identify abnormal lane lines.
[0004] However, traditional technical solutions rely on manually defined rules to identify noisy lane lines, which suffers from low recognition accuracy. Summary of the Invention
[0005] This application provides a lane line recognition method and apparatus to overcome the problem of low accuracy in recognizing abnormal lane lines.
[0006] In a first aspect, embodiments of this application provide a lane line recognition method, including:
[0007] Obtain multiple lane lines to be identified;
[0008] Each lane line to be identified is divided into multiple line segments;
[0009] For any given line segment, an initial feature vector is determined based on the lane line attribute of the lane to which the line segment belongs. The initial feature vector does not reflect the direction of the lane line.
[0010] Based on the initial feature vectors corresponding to each line segment, the multiple line segments are clustered to obtain at least one cluster.
[0011] For any given cluster, each line segment is classified according to the direction attribute of the lane line to which it belongs, to obtain at least one category.
[0012] Identify abnormal lane lines based on the number of line segments corresponding to at least one category.
[0013] Secondly, embodiments of this application provide a lane line recognition device, comprising:
[0014] The acquisition module is used to acquire multiple lane lines to be identified;
[0015] The segmentation module is used to divide each lane line to be identified into multiple segments;
[0016] The determination module is used to determine an initial feature vector corresponding to any given line segment based on the lane line attribute of the lane line to which the line segment belongs. The initial feature vector does not reflect the direction of the lane line.
[0017] The processing module is used to perform clustering processing on the multiple line segments according to the initial feature vectors corresponding to each line segment, so as to obtain at least one cluster.
[0018] The classification module is used to classify each line segment according to the direction attribute of the lane line to which each line segment in the cluster belongs, for any given cluster, to obtain at least one category;
[0019] The identification module is used to identify abnormal lane lines based on the number of line segments corresponding to the at least one category.
[0020] Thirdly, embodiments of this application provide an electronic device, including:
[0021] Memory, used to store programs;
[0022] A processor for executing the program stored in the memory, wherein, when the program is executed, the processor is configured to perform the method described in the first aspect above and any of the various possible designs of the first aspect.
[0023] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the methods described in the first aspect above and any of the various possible designs of the first aspect.
[0024] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect above and any of the various possible designs of the first aspect.
[0025] This application provides a lane line recognition method and apparatus. The method includes: acquiring multiple lane lines to be recognized; dividing each lane line to be recognized into multiple segments; for any given segment, determining an initial feature vector corresponding to the segment based on the lane line attribute of the segment, wherein the initial feature vector does not reflect the direction of the lane line; clustering the multiple segments based on the initial feature vectors corresponding to each segment to obtain at least one cluster; for any given cluster, classifying each segment based on the direction attribute of the lane line to which each segment belongs in the cluster to obtain at least one category; and identifying abnormal lane lines based on the number of segments corresponding to the at least one category. By dividing multiple lane lines into multiple segments, and considering that normal lane lines and redundant lane lines are similar in all aspects except for their direction, clustering is performed on the initial feature vector of each segment that does not reflect the direction attribute of the lane line. This clusters normal lane lines and redundant lane lines into the same cluster. Then, the direction attribute is added to classify the segments in the cluster, so that the segments in the cluster are divided into at least two categories, where the lane lines to which the segments in different categories belong are different. This distinguishes the normal lane lines and redundant lane lines in the cluster. Finally, based on the number of segments corresponding to each category, abnormal lane lines can be effectively and accurately identified. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 A schematic diagram of redundant lane lines provided in the embodiments of this application;
[0028] Figure 2 A flowchart illustrating the lane line recognition method provided in this application embodiment;
[0029] Figure 3 This is a schematic diagram of lane lines with similar features provided in the embodiments of this application;
[0030] Figure 4 The flow chart of the lane line recognition method provided in the embodiments of this application Figure 2 ;
[0031] Figure 5 A schematic diagram illustrating the implementation of dividing lane lines into line segments, provided in an embodiment of this application;
[0032] Figure 6A schematic diagram of the hyperplane provided for an embodiment of this application;
[0033] Figure 7 This is a schematic diagram illustrating the process of identifying abnormal lane lines provided in an embodiment of this application;
[0034] Figure 8 This is a schematic diagram of the lane line recognition device provided in the embodiments of this application;
[0035] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation
[0036] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0037] To better understand the technical solution of this application, the relevant technologies involved in this application will be further described in detail below.
[0038] During the production of high-precision map data, a large number of redundant and noisy lane lines may be generated. For example, multiple lane lines may be output for the same data collection object, and these lane lines may have inconsistent directions. In this case, there will inevitably be lane lines with recognition errors and redundancy. The data collection object introduced here can be understood as lane lines in a real-world scene.
[0039] For example, it can be combined Figure 1 To understand the application scenarios of this application, Figure 1 A schematic diagram of redundant lane lines provided in an embodiment of this application.
[0040] like Figure 1 As shown, assuming that a high-precision map is produced based on point cloud data of a certain road segment, the output is... Figure 1 The diagram shows six lane lines, a through f, each with its own direction. Figure 1 The arrows in the middle illustrate this.
[0041] Reference Figure 1 It is certain that lane line c and lane line d are very close in distance, but opposite in direction. This is because in the actual scenario, there is only one lane line at that location. In other words, at least one of lane lines c and d is a redundant lane line, and the direction of this redundant lane line is also incorrect.
[0042] The reason for this situation is that during lane line generation, lane lines are typically generated based on a point cloud dataset collected along a certain driving direction. To ensure data completeness, as many features as possible are collected, and the direction of the lane line collected in a single pass is consistent with the driving direction corresponding to the current point cloud data. This leads to redundancy at the intersection of two driving directions, and even at non-intersection locations. In summary, in a real-world scenario, there may only be one lane line at a certain location, but the map data may show multiple lane lines at that location. Furthermore, the recognized direction of some lane lines may not match the actual direction, resulting in noise and redundancy in the high-precision map.
[0043] To ensure the accuracy of high-precision map data, it is necessary to identify and deduplicate redundant and noisy lane lines. Currently, in the existing technology, when identifying redundant and noisy lane lines, staff usually manually check and mark abnormal lane lines, or staff manually define identification strategies based on the characteristics of lane line noise, and then analyze each lane line according to the identification strategy to identify abnormal lane lines.
[0044] However, manually checking and marking abnormal lane lines suffers from low automation efficiency. The implementation of manually defined recognition strategies typically involves first manually analyzing the characteristics of noisy lane lines, and then setting a targeted recognition strategy. For the scenario described above, a possible strategy is as follows: among two lane lines that are close in distance but in opposite directions, there might be a redundant lane line.
[0045] However, this strategy cannot distinguish which lane lines are redundant and which are normal. For example, additional information could be added to differentiate them, such as determining the distance and direction of each lane line from its neighboring lane lines. However, it is still impossible to define a universal and accurate identification strategy to identify the redundant lane lines described in the above scenario. Therefore, the accuracy of manually defined identification strategies is not high.
[0046] Therefore, traditional solutions rely on manual operation or manually defined rules, resulting in low recognition accuracy and low automation efficiency.
[0047] To address the aforementioned technical problems, this application proposes the following technical concept: The redundant lane lines and normal lane lines identified in the above scenario share the characteristic of being close in distance but facing opposite directions. Therefore, lane lines can be divided into multiple segments. Then, these segments are clustered based on lane line features without direction, thereby grouping the redundant and normal lane lines together. Finally, the clustered results are classified based on lane line features with direction to identify abnormal lane lines. The lane line recognition method provided by this application is described below with reference to specific embodiments. The execution entity in each embodiment of this application can be a server, processor, chip, or other device with data processing capabilities. This embodiment does not limit the specific execution entity.
[0048] First, combine Figure 2 and Figure 3 To explain, Figure 2 This is a flowchart of the lane line recognition method provided in the embodiments of this application. Figure 3 This is a schematic diagram of lane lines with similar features provided in an embodiment of this application.
[0049] like Figure 2 As shown, the method includes:
[0050] S201. Obtain multiple lane lines to be identified.
[0051] In this embodiment, redundant lane lines need to be identified. Therefore, there will be at least two lane lines in the dataset to be processed, and each lane line contains multiple shape points.
[0052] S202. Divide each lane line to be identified into multiple line segments.
[0053] In this embodiment, each lane line to be identified can be divided separately, thereby dividing each lane line into multiple segments. For example, for any lane line, this embodiment can divide the lane line into multiple segments based on the multiple shape points contained in the lane line, where the endpoints of the segments are the shape points.
[0054] S203. For any line segment, determine the initial feature vector corresponding to the line segment based on the lane line attribute of the lane line to which the line segment belongs. The initial feature vector does not reflect the direction of the lane line.
[0055] After dividing the lane into segments, the correspondence between each segment and the lane lines remains unchanged; that is, each segment belongs to a specific lane line. In this embodiment, the same processing is applied to all lane lines and segments. The following explanation uses any one segment as an example.
[0056] In this embodiment, each lane line to which a line segment belongs has lane line attributes, which may include, for example, the position of the lane line, the identification of the lane line, the type of the lane line, the identification of the road segment to which the lane line belongs, etc. This embodiment does not limit the specific implementation of the lane line attributes.
[0057] Then, based on the lane line attributes of the lane to which the line segment belongs, the initial feature vector corresponding to the line segment can be determined. In this embodiment, the initial feature vector does not reflect the direction of the lane line. For example, the lane line attributes can be set to not include the direction of the lane line, because the initial feature vector in this embodiment is obtained based on the lane line attributes and is used to reflect the various attribute information contained in the lane line attributes. Therefore, the initial feature vector determined based on the lane line attributes that do not include the direction of the lane line will naturally not reflect the direction of the lane line.
[0058] Therefore, in actual implementation, any attribute information related to lane lines other than the direction of the lane lines can be used as lane line attributes in this embodiment.
[0059] S204. Based on the initial feature vectors corresponding to each line segment, perform clustering processing on multiple line segments to obtain at least one cluster.
[0060] Then, clustering can be performed on multiple line segments based on their respective initial feature vectors. It's understandable that the initial feature vectors of line segments within the same lane and those of adjacent lane segments are quite similar. Here, adjacent lane segments can include the aforementioned... Figure 1 The scenario described, where the lane lines are very close to each other on the left and right, can also include... Figure 3 This illustration shows a situation where multiple lane markings are consecutive on the same section of road. Figure 3 The lane lines a and b shown in the diagram are two consecutive lane lines on the same road segment, and the initial feature vectors corresponding to each segment of these two lane lines are also quite similar.
[0061] In this embodiment, after clustering, at least one cluster can be obtained. Each cluster contains multiple line segments with similar initial feature vectors. Based on the above description, it can be understood that in real-world scenarios where redundant lane lines are identified from a single lane line, the correct lane line and the redundant lane line are often very similar in all aspects, differing only in direction.
[0062] In this embodiment, the initial feature vector does not reflect the direction attribute of the lane line. Excluding the direction attribute, the remaining attributes of the correct lane line and the redundant lane line are very similar. Therefore, the initial feature vectors of the line segments of the correct lane line and the redundant lane line described above are also very similar, and thus will be clustered into the same cluster. In this embodiment, the correct lane line and the redundant lane line identified for the same lane line are first classified into one cluster. Of course, this cluster may also include other neighboring lane lines.
[0063] Here, we will further explain the reasons for dividing the lane lines into segments as described above. One reason for dividing the lane lines into multiple segments is to facilitate clustering. Compared to clustering based on a relatively small number of long lane lines, clustering based on a large number of short segments makes it easier to group similar lane line segments into a single cluster, thus ensuring the accuracy of the clustering results.
[0064] S205. For any given cluster, classify each line segment according to the direction attribute of the lane line to which each line segment belongs in the cluster, and obtain at least one category.
[0065] After clustering the correct lane lines and redundant lane lines identified for the same lane line into the same cluster, since the lane lines of the correct lane lines and redundant lane lines are different, we can further classify the line segments in the cluster according to the lane line direction attribute to distinguish between the correct lane lines and redundant lane lines. The processing method for each cluster is similar in this embodiment, so we will take any cluster as an example for introduction.
[0066] In this embodiment, each line segment can be classified according to the direction attribute of the lane line to which each line segment belongs in the cluster, thereby obtaining at least one category, wherein the direction of the lane line corresponding to the line segment of different categories is different.
[0067] In this embodiment, it is not necessary to determine the specific direction of the lane line corresponding to each line segment. It is only necessary to classify the line segments corresponding to different lane line directions into different categories. For example, the line segments in the cluster are divided into a first category and a second category. In fact, it is not necessary to know which lane line direction the first category actually corresponds to. Similarly, it is not necessary to know that the lane line directions corresponding to the line segments of the first category and the line segments of the second category are different. That is to say, in this embodiment, it is only necessary to classify the line segments, and it is not necessary to know the specific meaning of the category. This saves computing resources and avoids unnecessary operations while achieving the purpose of the solution.
[0068] S206. Identify abnormal lane lines based on the number of line segments corresponding to at least one category.
[0069] After classifying the line segments in the clusters, abnormal lane lines can be identified based on the number of line segments corresponding to at least one category.
[0070] For example, the presence of abnormal lane lines can be determined first by the number of categories. If there is only one category, it means that all line segments in the cluster are classified into the same category, which means that the lane lines corresponding to the line segments in that category are all in the same direction, and thus it can be determined that there are no abnormal lane lines.
[0071] However, if the number of categories is greater than or equal to 2, it means that the line segments in the cluster are divided into different categories, and the direction of the lane lines corresponding to the line segments of different categories is different, which can also determine the existence of abnormal lane lines.
[0072] When abnormal lane lines are identified, since noisy and redundant lane lines are always a small part of the overall lane line data, the line segments corresponding to redundant lane lines are also a small part of the data in a cluster. Therefore, the number of line segments corresponding to each category can reflect the correct direction of the lane lines corresponding to the current cluster. So in this embodiment, abnormal lane lines can be identified based on the number of line segments corresponding to each category.
[0073] The lane line recognition method provided in this application includes: acquiring multiple lane lines to be recognized; dividing each lane line to be recognized into multiple segments; for any given segment, determining an initial feature vector corresponding to the segment based on the lane line attribute of the segment to which it belongs, wherein the initial feature vector does not reflect the direction of the lane line; clustering the multiple segments based on the initial feature vectors corresponding to each segment to obtain at least one cluster; classifying each segment based on the direction attribute of the lane line to which each segment in the cluster belongs to each cluster to obtain at least one category; and identifying abnormal lane lines based on the number of segments corresponding to at least one category. In this embodiment, multiple lane lines are divided into multiple segments. Then, based on the characteristic that normal lane lines and redundant lane lines are similar in all aspects except for their direction, clustering is performed on the initial feature vector of each segment that does not reflect the direction attribute of the lane line. This clusters normal lane lines and redundant lane lines into the same cluster. Next, the segments in the cluster are classified according to their direction attribute, dividing the segments in the cluster into at least two categories. The different categories correspond to lane lines with different directions, thus distinguishing normal lane lines from redundant lane lines in the cluster. Finally, based on the number of segments corresponding to each category, abnormal lane lines can be effectively and accurately identified.
[0074] Based on the above introduction, the following will further combine... Figures 4 to 7 The lane line recognition method provided in this application will be further described in detail. Figure 4 The flow chart of the lane line recognition method provided in the embodiments of this application Figure 2 , Figure 5 This is a schematic diagram illustrating the implementation of dividing lane lines into line segments according to an embodiment of this application. Figure 6 This is a schematic diagram of the hyperplane provided in an embodiment of this application. Figure 7 This is a schematic diagram illustrating the process of identifying abnormal lane lines provided in an embodiment of this application.
[0075] like Figure 4 As shown, the method includes:
[0076] S401. For any lane line, connect every two adjacent shape points in the lane line in sequence to obtain multiple line segments after the lane line is divided.
[0077] In this embodiment, when dividing the lane line into multiple segments, in one possible implementation, every two adjacent shape points of the lane line can be connected sequentially, with each connection resulting in a segment, thereby obtaining multiple segments after the lane line is divided.
[0078] For example, it can be combined Figure 5 To understand, such as Figure 5As shown, suppose there is currently a lane line, which contains Figure 5 The six shape points shown (shape points 1 through 6) are then connected sequentially between every two adjacent shape points in the lane line to obtain five line segments. (Refer to...) Figure 5 Connecting adjacent points 1 and 2 yields a line segment consisting of point 1 and point 2; connecting adjacent points 2 and 3 yields a line segment consisting of point 2 and point 3; and so on, to obtain... Figure 5 The five line segments are shown in the image.
[0079] In this embodiment, by sequentially connecting every two adjacent shape points in the lane line, multiple line segments are obtained after dividing the lane line. This effectively ensures that each line segment is as short as possible and that the number of line segments is as large as possible, so as to provide a large number of simple clustering samples for subsequent clustering processing. Generally, the more clustering samples there are, the more accurate the clustering analysis results will be. Therefore, this embodiment can effectively improve the accuracy of the clustering results.
[0080] In practical implementation, for example, a line segment can be obtained by dividing a continuous set of shape points (e.g., 3 or 4). The two endpoints of the line segment are the first and last of these continuous shape points. Alternatively, as described above, a fixed number of shape points can be used to divide the line segment, ensuring that all line segments have the same length. Or, the length of each line segment can be flexible, meaning the number of shape points covered by each line segment can vary. This application does not limit the specific implementation method for dividing the lane line into multiple segments; it can be selected and set according to actual needs, as long as at least one division of the lane line is achieved.
[0081] S402. For any line segment, determine the initial feature vector corresponding to the line segment based on the lane line attribute of the lane line to which the line segment belongs. The initial feature vector does not reflect the direction of the lane line.
[0082] The implementation of S402 is similar to that of S202, and will not be described in detail here.
[0083] S403. Map the initial feature vectors corresponding to each line segment to a high-dimensional feature space to obtain the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector.
[0084] After obtaining the initial feature vectors corresponding to each line segment, this embodiment can perform clustering processing on each line segment based on the initial feature vectors. In one possible implementation, this embodiment can first map the initial feature vectors corresponding to each line segment to a high-dimensional feature space to obtain the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector. Specifically, the vector distance here is the vector distance between the high-dimensional feature vectors of each line segment in the high-dimensional feature space.
[0085] It is understandable that the initial feature vector, relative to the corresponding high-dimensional feature vector, can be considered a low-dimensional feature vector. Here, high-dimensional and low-dimensional are relative concepts; therefore, this embodiment does not limit the specific dimensions of the high-dimensional feature vector or the initial feature vector. The feature representation of a high-dimensional feature vector is richer and more complete than that of a low-dimensional feature vector. Therefore, mapping the initial feature vector to a high-dimensional feature space, and then performing clustering based on the vector distances between the high-dimensional feature vectors, can further improve the accuracy of the clustering results.
[0086] When determining the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector, one approach is to map the initial feature vectors corresponding to each line segment to a high-dimensional feature space. This first yields the high-dimensional feature vectors corresponding to each initial feature vector, and then the vector distance between them is determined based on these high-dimensional feature vectors. However, this approach requires first obtaining the high-dimensional feature vectors based on the mapping relationship, and then processing them. Because high-dimensional feature vectors have many dimensions, the computational complexity is also very high.
[0087] However, mapping feature vectors to a high-dimensional feature space and then calculating the vector distance in that high-dimensional feature space yields the same result as directly calculating the vector distance between feature vectors and then mapping to a high-dimensional feature space. Therefore, to address this issue, this embodiment employs a kernel function. The kernel function directly outputs the inner product of two feature vectors mapped to a high-dimensional feature space. This inner product reflects the vector distance between the two feature vectors, eliminating the need for calculation within the high-dimensional feature space. In other words, the kernel function allows obtaining the vector distance between feature vectors in a high-dimensional feature space without needing to determine the specific mapping function or even select a mapping function.
[0088] The kernel function can be expressed as follows: Assume there exist eigenvectors x and x' in the low-dimensional feature space, and that there exists a function K(x, x') in the low-dimensional feature space, which is exactly equal to the inner product of the mapping eigenvectors of eigenvectors x and x' in the high-dimensional feature space, i.e. For mapping functions, This means that the eigenvector x is mapped to a high-dimensional eigenvector in the high-dimensional feature space, and φ(x`) represents the mapping of the eigenvector x` to a high-dimensional eigenvector in the high-dimensional feature space. Therefore, in this embodiment, there is no need to calculate complex nonlinear transformations. The vector distance in the high-dimensional feature space can be directly obtained from this kernel function K(x,x`), which greatly simplifies the calculation. Such a K(x,x`) function is called a kernel function.
[0089] Taking any two initial feature vectors from the initial feature vectors corresponding to each line segment as an example, in this embodiment, these two initial feature vectors can be input into a kernel function to obtain the inner product between the high-dimensional feature vectors corresponding to each of the two initial feature vectors output by the kernel function. Then, the inner product of the two high-dimensional feature vectors can reflect the vector distance between the two high-dimensional feature vectors. The above processing is performed on any two initial feature vectors from multiple initial feature vectors (which can also be understood as every two), thereby obtaining the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector.
[0090] For example, there are n line segments, and the initial feature vectors corresponding to these n line segments are initial feature vector 1 to initial feature vector n, where n is an integer greater than or equal to 1.
[0091] Then, we can input initial feature vector 1 and initial feature vector 2 into the kernel function to obtain the inner product of the corresponding high-dimensional feature vectors of each of the two initial feature vectors 1. Then, we can input initial feature vector 1 and initial feature vector 3 into the kernel function to obtain the inner product of these two vectors in the high-dimensional feature space. Then, we can input initial feature vector 1 and initial feature vector 4, and so on, until for any two initial feature vectors, we obtain the inner product of their corresponding high-dimensional feature vectors.
[0092] S404. Based on the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector, perform clustering on multiple line segments to obtain at least one cluster.
[0093] After obtaining the vector distances between the high-dimensional feature vectors corresponding to each initial feature vector, clustering can be performed based on the vector distances. It can be understood that clustering is to group line segments with similar feature vectors into a cluster. The vector distances between feature vectors can directly reflect whether the vectors are similar. Therefore, clustering of multiple line segments can be performed efficiently and accurately to obtain at least one cluster.
[0094] S405. For any line segment in any cluster, obtain the advanced feature vector corresponding to the line segment based on the direction attribute of the lane line to which the line segment belongs.
[0095] Each cluster contains multiple line segments. Line segments within a cluster typically share similar characteristics; their corresponding lane lines may be those with similar left and right distances as described above, or they may be continuous within the same road segment. This embodiment will perform the same processing on each cluster; therefore, the following explanation uses any single cluster as an example.
[0096] In this embodiment, for any line segment in a cluster, an advanced feature vector is obtained based on the direction attribute of the lane line to which the line segment belongs. This advanced feature vector reflects the direction of the lane line to which the line segment belongs. It can be understood that when generating a high-precision map, the direction of each lane line is marked; the direction attribute here refers to the direction marked for the lane line.
[0097] It is understandable that, compared with the initial feature vector described above, the advanced feature vector here will include the directional features of the lane lines corresponding to the line segments, while the initial feature vector does not include this feature.
[0098] For example, an advanced feature vector can be obtained based on the lane line attributes and direction attributes of the lane to which the line segment belongs, where the lane line attributes are similar to those described in the previous embodiment. In this case, the advanced feature vector can reflect not only the direction of the lane line corresponding to the line segment, but also the position, identification, type, road segment identification, and other attribute information of the lane line. In this implementation, because the advanced feature vector contains richer feature information, the accuracy of classification can be improved to a certain extent.
[0099] S406. Map the advanced feature vectors corresponding to each line segment to a high-dimensional feature space to obtain the vector distance between the high-dimensional feature vectors corresponding to each advanced feature vector.
[0100] After obtaining the advanced feature vectors corresponding to each line segment, this embodiment will also map the advanced feature vectors corresponding to each line segment to a high-dimensional feature space to obtain the vector distance between the high-dimensional feature vectors corresponding to each advanced feature vector. The implementation method here is similar to the implementation method of S403 above.
[0101] Therefore, taking any two advanced feature vectors from the advanced feature vectors corresponding to each line segment as an example, in this embodiment, these two advanced feature vectors can be input into a kernel function to obtain the inner product between the high-dimensional feature vectors corresponding to each of the two advanced feature vectors output by the kernel function. Then, the inner product of the two high-dimensional feature vectors can reflect the vector distance between the two high-dimensional feature vectors. The above processing is performed on any two advanced feature vectors from multiple advanced feature vectors (which can also be understood as every two), thereby obtaining the vector distance between the high-dimensional feature vectors corresponding to each advanced feature vector.
[0102] S407. The support vector machine is used to classify the line segments according to the vector distance between each high-dimensional feature vector to obtain the category corresponding to each line segment output by the support vector machine.
[0103] After obtaining the vector distance between the high-dimensional feature vectors corresponding to each line segment, the line segments in the cluster can be classified. In one possible implementation, this embodiment can use a support vector machine for classification.
[0104] Support Vector Machine (SVM) is a type of generalized linear classifier that performs binary classification of data using supervised learning. Its decision boundary is a hyperplane solved by the learning samples. In other words, the SVM will find a hyperplane as the classification standard. All data points on the same side of this hyperplane belong to the first class, and all data points on the other side belong to the second class.
[0105] The decision boundary that separates the dataset is called the hyperplane. If the dataset is N-dimensional, then the hyperplane is N-1-dimensional. The hyperplane can be simply understood as a linear expression. In two dimensions, the hyperplane is a straight line, and in three dimensions, the hyperplane is a plane.
[0106] Based on the above introduction, it is understandable that to achieve data classification, a hyperplane must be found to separate the two classes of data. However, for classification, some datasets are linearly separable, meaning that different classes of data in the dataset can be separated by a straight line, while some datasets are linearly inseparable, meaning that different classes of data cannot be separated by a straight line.
[0107] For example, refer to Figure 6 In (a), the data in the dataset can be divided into two categories, represented by stars and circles respectively. Figure 6 In the low-dimensional space shown in (a), no straight line can be found to separate the two types of data. However, referring to... Figure 6 (b) If the dataset is mapped to a high-dimensional space, a hyperplane can be found in the high-dimensional space to separate the two different categories of data.
[0108] Therefore, in this embodiment, in addition to the reason that vectors in high-dimensional feature space have richer feature representations as mentioned above, another reason for determining the vector distance of the advanced feature vector in high-dimensional feature space is that the advanced feature vector corresponding to the line segment can be classified in high-dimensional feature space.
[0109] When performing classification, Support Vector Machines (SVMs) only care about the inner product between data points, not the vector representation of the advanced feature vectors in the high-dimensional feature space. Therefore, in this embodiment, the advanced feature vectors are processed by a kernel function to obtain the vector distance between each pair of high-dimensional feature vectors of each advanced feature vector. Then, the SVM is used to perform classification based on the vector distance between each high-dimensional feature vector, which can effectively obtain the category corresponding to each line segment output by the SVM. Therefore, the technical solution of this application can accurately and effectively classify line segments in the cluster.
[0110] The support vector machine introduced in this embodiment is a binary classification machine. When the number of categories is greater than 1, the category corresponding to the line segment in this embodiment can be represented as the first category or the second category. It should be noted that in this embodiment, it is not necessary to determine what direction the first category and the second category specifically correspond to. That is to say, we do not care about the meaning of the category. We only need to divide the line segments corresponding to different lane line directions into two categories to achieve the purpose of the subsequent solution. Therefore, the technical solution of this application avoids unnecessary analysis and processing and effectively saves computing resources.
[0111] Alternatively, in actual implementation, other multi-class classification models can be used for the classification process described above. In this case, the categories of line segments obtained by the classification are not limited to the first and second categories described above.
[0112] S408. Based on the category corresponding to each line segment in the cluster, classify each line segment to obtain at least one category.
[0113] In this embodiment, each line segment in the cluster will be assigned a corresponding category. After determining the category of each line segment in the cluster, the line segments can be classified to obtain at least one category.
[0114] For example, if each line segment corresponds to the same category, then the classification process can result in a category that includes all line segments in the current cluster.
[0115] If the categories corresponding to each line segment are not exactly the same, for example, some line segments correspond to the first category as described above, and some line segments correspond to the second category as described above, then it can be determined that the classification process has yielded the first category and the second category.
[0116] S409. If the number of categories obtained after classification is greater than 1, then it is determined that there are abnormal lane lines.
[0117] After classification, it can be first determined whether the number of categories after classification is greater than 1. Only if the number of categories is greater than 1 can it be determined that there are abnormal lane lines, and then subsequent processing can be performed.
[0118] If the number of categories is equal to 1, it can be determined that there are no abnormal lane lines, so the current process can be terminated directly.
[0119] S410. Based on the number of line segments in the first category and the total number of line segments in the cluster, determine the first proportion corresponding to the number of line segments in the first category.
[0120] Then, abnormal lane lines can be identified based on the number of line segments in the first category and the number of line segments in the second category. It should be noted that during clustering, line segments with similar features will be clustered into one cluster. The features of the correct lane lines and redundant lane lines identified for the same lane line are similar, and the features of multiple consecutive lane lines in the same road segment may also be quite similar. Therefore, a cluster may contain the line segments of the lane lines described above.
[0121] In other words, the number of lane line segments in a cluster is very large. Because redundant lane lines are always a small part of the overall lane line data, the number of segments corresponding to redundant lane lines in a cluster is also a small part. Therefore, in this embodiment, the proportion of each type of line segment in the cluster can be determined, thereby identifying abnormal line segments.
[0122] Therefore, in this embodiment, the first proportion corresponding to the number of line segments in the first category can be determined based on the number of line segments in the first category and the total number of line segments in the cluster.
[0123] For example, you can refer to Figure 7To understand this, assuming the total number of line segments in the current cluster is 2000, and the number of line segments in the first category is 1900, then the proportion of line segments in the first category is 95%.
[0124] S411. Based on the number of line segments in the second category and the total number of line segments in the cluster, determine the second proportion corresponding to the number of line segments in the second category.
[0125] Similarly, in this embodiment, the second proportion corresponding to the number of line segments in the second category can be determined based on the number of line segments in the second category and the total number of line segments in the cluster.
[0126] For example, you can refer to Figure 7 To understand this, assuming the total number of line segments in the current cluster is 2000, and the number of line segments in the second category is 100, then the proportion of the number of line segments in the first category is determined to be 5%.
[0127] S412. Compare the target values in the first proportion and the second proportion with the preset threshold to identify abnormal lane lines.
[0128] The target value is either the larger or smaller value. That is, we can first compare the first proportion and the second proportion to determine the larger or smaller value. The larger value corresponds to the category of the majority of line segments in the current cluster, indicating that the lane lines corresponding to these line segments are in the correct direction. Conversely, the smaller value corresponds to the category of a smaller subset of line segments in the current cluster, indicating that the lane lines corresponding to these line segments are in the wrong direction.
[0129] For example in Figure 7 In the example, the first proportion is 95% and the second proportion is 5%. Therefore, we can determine that the first proportion is the larger value. Thus, we can determine that the direction of the lane lines corresponding to each line segment in the first category is the normal direction. Correspondingly, the direction of the lane lines corresponding to each line segment in the second category should be abnormal.
[0130] However, it's important to note that achieving the above analytical goals requires a prerequisite: a significantly higher percentage of lanes belonging to a particular category. In this case, the lane direction corresponding to the category with the higher percentage can be considered the normal direction. Conversely, if a significantly lower percentage of lanes belongs to a particular category, the lane direction corresponding to the category with the lower percentage can be considered the abnormal direction. If the percentages of two categories are roughly the same, the accuracy of the determined normal lane direction cannot be guaranteed.
[0131] Therefore, a preset threshold is set in this embodiment to limit the target value in the first proportion and the second proportion to be significantly larger or smaller before subsequent judgment can be made.
[0132] In one possible implementation, if the target value is large, the larger of the first proportion and the second proportion must be greater than or equal to the first preset threshold before the normal direction of the lane line can be correctly determined. Then, the category corresponding to the larger value can be determined as the normal category, and the category corresponding to the smaller of the first proportion and the second proportion can be determined as the abnormal category. Then, the lane lines to which the line segments of the abnormal category in the cluster belong can be determined as abnormal lane lines.
[0133] For example, refer to Figure 7 To understand this, the larger of the first and second percentages is 95%. Assuming a preset threshold of 80%, the larger value is greater than the preset threshold. Therefore, the first category can be classified as the normal category, and the second category as the abnormal category. Furthermore, the lane lines belonging to the second category within the cluster can be classified as abnormal lane lines. For example... Figure 7 The diagram illustrates that the lane lines belonging to the second category of line segments include lane line d and lane line n. Therefore, these two lane lines can be identified as abnormal lane lines, or redundant lane lines.
[0134] If the larger of the first and second proportions is less than a preset threshold, it indicates that the proportions of the two categories of line segments are not overwhelmingly different. For example, the first proportion is 46% and the second proportion is 54%. In this case, it is impossible to correctly analyze which category is normal and which category is abnormal, and it is difficult to distinguish which lane line direction is normal. This situation may be because there are a large number of continuous redundant lane lines, which are clustered with normal lane lines in the same cluster. Therefore, in this embodiment, the current cluster can be marked, and each line segment in the cluster and its corresponding category can be output so that the abnormal lane lines can be correctly identified later through special classification processing or manual identification.
[0135] In another possible implementation, if the target value is small, the smaller of the first proportion and the second proportion needs to be greater than or equal to the second preset threshold before the normal direction of the lane line can be correctly determined. Then, the category corresponding to the smaller value can be determined as the abnormal category, and the category corresponding to the larger of the first proportion and the second proportion can be determined as the normal category. Then, the lane lines to which the line segments of the abnormal category in the cluster belong are determined as abnormal lane lines.
[0136] The first preset threshold is set for larger values, and the second preset threshold is set for smaller values. Therefore, the first preset threshold is greater than the second preset threshold.
[0137] Therefore, in this embodiment, by first comparing the target value in the first proportion and the second proportion with a preset threshold, and only when the target value meets the size relationship with the preset threshold, the normal category and the abnormal category are determined in the first category and the second category. Then, the lane line corresponding to the line segment of the abnormal category is determined as the abnormal lane line, which can effectively ensure the accuracy of the identified abnormal lane line.
[0138] Alternatively, in an optional implementation, the ratio of the number of line segments in the first category to the number of line segments in the second category can be determined. This ratio is then compared to a corresponding threshold to determine which category of line segments constitutes the majority in the cluster. For example, the ratio of a larger value to a smaller value could be determined, or the ratio of a smaller value to a larger value could be determined. In this case, the corresponding threshold or the method of comparison with the threshold would differ.
[0139] For example, if the ratio of the larger value to the smaller value is determined, the corresponding comparison threshold is usually set relatively large. When the ratio is greater than or equal to the preset value, abnormal lane lines are determined in a similar manner as described above. When the ratio is less than the preset threshold, the clusters are marked and output.
[0140] Correspondingly, if the ratio of the smaller value to the larger value is determined, the comparison threshold is usually set relatively small. When the ratio is less than or equal to the preset threshold, abnormal lane lines are determined in a similar manner to the above. When the ratio is greater than the preset threshold, the clusters are marked and output.
[0141] In actual implementation, the specific implementation method for identifying abnormal lane lines based on the number of line segments in each category can be selected and set according to actual needs. As long as the category that accounts for the majority of the number of line segments in each category is determined based on the number of line segments in each category, the abnormal lane lines can be determined accordingly. Therefore, this embodiment does not limit the specific implementation method for determining abnormal lane lines.
[0142] In summary, the lane line recognition method provided in this application first clusters lane line segments based on lane line attributes excluding direction, thereby grouping correct and redundant lane lines identified for the same lane line into the same cluster. Then, for each segment within each cluster, the segments are further classified according to the lane line's direction attribute. Based on the number of segments in each classified category, segments of the abnormal category are determined, and the lane lines corresponding to these abnormal segments are identified as abnormal lane lines. This automatically achieves the correct identification of abnormal or redundant lane lines, improving recognition efficiency and accuracy.
[0143] It should also be noted that the lane line recognition method provided in this application can not only classify the correct lane lines and redundant lane lines obtained by the same lane line recognition as described above, but can also classify lane lines that are close in distance but have different attributes in a certain aspect in the road. This can be achieved by simply replacing the directional attribute described above with the attribute that is significantly different. The specific implementation method can be referred to the description of the above embodiments, and will not be repeated here.
[0144] Figure 8 This is a schematic diagram of the lane line recognition device provided in an embodiment of this application. Figure 8 As shown, the device 80 includes: an acquisition module 801, a division module 802, a determination module 803, a classification module 804, and an identification module 805.
[0145] The acquisition module 801 is used to acquire multiple lane lines to be identified;
[0146] The segmentation module 802 is used to divide each lane line to be identified into multiple line segments;
[0147] The determining module 803 is used to determine, for any given line segment, an initial feature vector corresponding to the line segment based on the lane line attribute of the lane line to which the line segment belongs, wherein the initial feature vector does not reflect the direction of the lane line;
[0148] The processing module is used to perform clustering processing on the multiple line segments according to the initial feature vectors corresponding to each line segment, so as to obtain at least one cluster.
[0149] The classification module 804 is used to classify each line segment according to the direction attribute of the lane line to which each line segment in the cluster belongs, for any given cluster, to obtain at least one category;
[0150] The identification module 805 is used to identify abnormal lane lines based on the number of line segments corresponding to the at least one category.
[0151] The processing module is specifically used for:
[0152] Map the initial feature vectors corresponding to each line segment to a high-dimensional feature space to obtain the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector.
[0153] Based on the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector, the multiple line segments are clustered to obtain at least one cluster.
[0154] In one possible design, the processing module is specifically used for:
[0155] For any two initial feature vectors in the initial feature vectors corresponding to each line segment, input the two initial feature vectors into the kernel function to obtain the inner product between the high-dimensional feature vectors corresponding to the two initial feature vectors output by the kernel function;
[0156] The inner product is used to reflect the vector distance between two high-dimensional feature vectors.
[0157] In one possible design, the identification module 805 is specifically used for:
[0158] If the number of categories obtained after the classification process is greater than 1, then it is determined that there is an abnormal lane line, and the at least one category includes a first category and a second category;
[0159] For any of the clusters, a first proportion of the number of line segments corresponding to the first category is determined based on the number of line segments corresponding to the first category within the cluster and the total number of line segments in the cluster.
[0160] Based on the number of line segments corresponding to the second category within the cluster and the total number of line segments in the cluster, determine the second proportion of the number of line segments corresponding to the second category;
[0161] The target values in the first proportion and the second proportion are compared with a preset threshold to identify abnormal lane lines, wherein the target value is a larger or smaller value.
[0162] In one possible design, the identification module 805 is specifically used for:
[0163] If the target value is large, then when the large value is greater than or equal to the first preset threshold, the category corresponding to the smaller value between the first proportion and the second proportion is determined as the abnormal category, and the lane line to which the line segment corresponding to the abnormal category in the cluster belongs is determined as the abnormal lane line.
[0164] If the target value is a smaller value, then when the smaller value is less than or equal to the second preset threshold, the category corresponding to the smaller value between the first proportion and the second proportion is determined as an abnormal category, and the lane line to which the line segment corresponding to the abnormal category in the cluster belongs is determined as the abnormal lane line.
[0165] The first preset threshold is greater than the second preset threshold.
[0166] In one possible design, the classification module 804 is specifically used for:
[0167] For any line segment in the cluster, the advanced feature vector corresponding to the line segment is obtained according to the direction attribute of the lane line to which the line segment belongs;
[0168] Based on the advanced feature vectors corresponding to each line segment, determine the vector distance between the high-dimensional feature vectors corresponding to each advanced feature vector;
[0169] The category of each line segment is determined based on the vector distance between the high-dimensional feature vectors corresponding to each of the advanced feature vectors.
[0170] Based on the category corresponding to each line segment in the cluster, the line segments are classified to obtain at least one category.
[0171] In one possible design, the classification module 804 is specifically used for:
[0172] The support vector machine is used to classify the line segments according to the vector distance between the high-dimensional feature vectors to obtain the category corresponding to each line segment output by the support vector machine, wherein the category corresponding to the line segment is the first category or the second category.
[0173] In one possible design, the partitioning module 802 is specifically used for:
[0174] For any lane line, connect every two adjacent shape points in the lane line in sequence to obtain multiple line segments after the lane line is divided.
[0175] The apparatus provided in this embodiment can be used to execute the technical solutions of the above method embodiments. Its implementation principle and technical effects are similar, and will not be described again here.
[0176] Figure 9 A schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application, such as... Figure 9 As shown, the electronic device 90 of this embodiment includes: a processor 901 and a memory 902; wherein
[0177] Memory 902 is used to store instructions executed by the computer;
[0178] The processor 901 is used to execute computer execution instructions stored in the memory to implement the various steps of the lane line recognition method in the above embodiments. For details, please refer to the relevant descriptions in the foregoing method embodiments.
[0179] Alternatively, the memory 902 can be either standalone or integrated with the processor 901.
[0180] When the memory 902 is set up independently, the electronic device also includes a bus 903 for connecting the memory 902 and the processor 901.
[0181] This application also provides a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the lane line recognition method performed by the above-mentioned electronic device.
[0182] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.
[0183] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.
[0184] The integrated modules implemented as software functional modules described above can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this application.
[0185] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.
[0186] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.
[0187] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.
[0188] The aforementioned storage medium can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.
[0189] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0190] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A lane line recognition method, characterized in that, include: Obtain multiple lane lines to be identified; Each lane line to be identified is divided into multiple line segments; For any given line segment, an initial feature vector is determined based on the lane line attribute of the lane to which the line segment belongs. The lane line attribute does not include the direction of the lane line, and the initial feature vector does not reflect the direction of the lane line. Based on the initial feature vectors corresponding to each line segment, the multiple line segments are clustered to obtain at least one cluster. For any given cluster, each line segment is classified according to the direction attribute of the lane line to which it belongs, to obtain at least one category. Identify abnormal lane lines based on the number of line segments corresponding to at least one category.
2. The method according to claim 1, characterized in that, The step of clustering the multiple line segments according to the initial feature vectors corresponding to each line segment to obtain at least one cluster includes: Map the initial feature vectors corresponding to each line segment to a high-dimensional feature space to obtain the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector. Based on the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector, the multiple line segments are clustered to obtain at least one cluster.
3. The method according to claim 2, characterized in that, The step of mapping the initial feature vectors corresponding to each line segment to a high-dimensional feature space to obtain the vector distance between the high-dimensional feature vectors corresponding to each initial feature vector includes: For any two initial feature vectors in the initial feature vectors corresponding to each line segment, input the two initial feature vectors into the kernel function to obtain the inner product between the high-dimensional feature vectors corresponding to the two initial feature vectors output by the kernel function; The inner product is used to reflect the vector distance between two high-dimensional feature vectors.
4. The method according to any one of claims 1-3, characterized in that, The step of identifying abnormal lane lines based on the number of line segments corresponding to at least one category includes: If the number of categories obtained after the classification process is greater than 1, then it is determined that there is an abnormal lane line, and the at least one category includes a first category and a second category; For any of the clusters, a first proportion of the number of line segments corresponding to the first category is determined based on the number of line segments corresponding to the first category within the cluster and the total number of line segments in the cluster. Based on the number of line segments corresponding to the second category within the cluster and the total number of line segments in the cluster, determine the second proportion of the number of line segments corresponding to the second category; The target values in the first proportion and the second proportion are compared with a preset threshold to identify abnormal lane lines, wherein the target value is a larger or smaller value.
5. The method according to claim 4, characterized in that, The step of comparing the target value in the first proportion and the second proportion with a preset threshold to identify abnormal lane lines includes: If the target value is large, then when the large value is greater than or equal to the first preset threshold, the category corresponding to the smaller value between the first proportion and the second proportion is determined as the abnormal category, and the lane line to which the line segment corresponding to the abnormal category in the cluster belongs is determined as the abnormal lane line. If the target value is a smaller value, then when the smaller value is less than or equal to the second preset threshold, the category corresponding to the smaller value between the first proportion and the second proportion is determined as an abnormal category, and the lane line to which the line segment corresponding to the abnormal category in the cluster belongs is determined as the abnormal lane line. The first preset threshold is greater than the second preset threshold.
6. The method according to claim 2, characterized in that, The step of classifying each line segment according to the direction attribute of the lane line to which each line segment belongs in the cluster to obtain at least one category includes: For any line segment in the cluster, the advanced feature vector corresponding to the line segment is obtained according to the direction attribute of the lane line to which the line segment belongs; Based on the advanced feature vectors corresponding to each line segment, determine the vector distance between the high-dimensional feature vectors corresponding to each advanced feature vector; The category of each line segment is determined based on the vector distance between the high-dimensional feature vectors corresponding to each of the advanced feature vectors. Based on the category corresponding to each line segment in the cluster, the line segments are classified to obtain at least one category.
7. The method according to claim 6, characterized in that, The step of determining the category of each line segment based on the vector distance between the high-dimensional feature vectors corresponding to each of the advanced feature vectors includes: The support vector machine (SVM) is used to classify the line segments based on the vector distance between the high-dimensional feature vectors to obtain the category corresponding to each line segment output by the SVM, wherein the category corresponding to the line segment is either the first category or the second category.
8. The method according to any one of claims 1-3, characterized in that, The process of dividing each lane line to be identified into multiple segments includes: For any lane line, connect every two adjacent shape points in the lane line in sequence to obtain multiple line segments after the lane line is divided.
9. A lane line recognition device, characterized in that, include: The acquisition module is used to acquire multiple lane lines to be identified; The segmentation module is used to divide each lane line to be identified into multiple segments; The determination module is used to determine an initial feature vector corresponding to any given line segment based on the lane line attribute of the lane to which the line segment belongs. The lane line attribute does not include the direction of the lane line, and the initial feature vector does not reflect the direction of the lane line. The processing module is used to perform clustering processing on the multiple line segments according to the initial feature vectors corresponding to each line segment, so as to obtain at least one cluster. The classification module is used to classify each line segment according to the direction attribute of the lane line to which each line segment in the cluster belongs, for any given cluster, to obtain at least one category; The identification module is used to identify abnormal lane lines based on the number of line segments corresponding to the at least one category.
10. An electronic device, characterized in that, include: Memory, used to store programs; A processor for executing the program stored in the memory, wherein when the program is executed, the processor is configured to perform the method as described in any one of claims 1 to 8.
11. A computer-readable storage medium, characterized in that, Includes instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1 to 8.