Data screening method and equipment for intelligent driving
By inputting the unlabeled test data sets into different lane line models and matching them based on specific indicators to determine the corresponding data for false detection and missed detection, the problem of inefficient performance improvement of lane line detection models in the shadow scene in the existing technology is solved, and efficient model retraining and performance improvement are achieved.
Patent Information
- Application Number
- CN202311204700.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-18
- Publication Date
- 2025-03-18
AI Technical Summary
The prior art is inefficient when improving the performance of lane line detection models in shadow scenarios, and the labeling of useless data increases unnecessary costs.
By entering the unlabeled test data set to different lane line models, multiple detection results are obtained, and matching the lane line gradient, cross-parameter IOU and confidence, the corresponding test data for false detection and missed detection are determined so that the model can be retrained after labeling.
It improves the performance efficiency of lane line detection models, reduces dependence on useless data, reduces labeling costs, and provides lightweight models with key data for performance improvement.
Smart Images
Figure CN119649322A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent driving, and more specifically, to a data screening method and device for intelligent driving, a training method for a business model, a computer storage medium, and a computer program product. Background Art
[0002] Lane detection is a basic task in autonomous driving or intelligent driving, and has an important impact on various autonomous driving functions (such as lane keeping, adaptive cruise control, automatic lane change, etc.) and subsequent lane departure or trajectory planning decisions.
[0003] In order to get a good lane detection model that meets the project requirements, it is necessary to find valid data for the detection model. For example, if we want to improve the performance of the lane detection model in shadow scenes, the usual method is to select the data with shadows manually or through shadow filters. The two usual methods can help select data for specific scenes, but it is understandable that the lane detection model does not perform poorly in all shadow scenes, that is, a lot of data is later proved to be useless to the model, and only a few data are helpful to improve the model, so it is inefficient; moreover, the annotation of useless data will also increase unnecessary costs. Summary of the invention
[0004] One or more embodiments of the present application are intended to provide an improved data screening solution for intelligent driving, which can help find data that is effective for improving model performance.
[0005] According to one aspect of the present application, a data screening method is provided, the method comprising: inputting an unlabeled test data set into a first object model to obtain a first detection result; inputting the test data set into one or more other object models different from the first object model to obtain a second detection result; and matching the first detection result with the second detection result to determine test data in the test data set corresponding to false detections and missed detections in the first detection result, wherein the test data corresponding to false detections and missed detections in the first detection result are used to retrain the first object model after labeling.
[0006] As a supplement or replacement for the above solution, in the above method, the data screening method is a lane line data screening method, and the first object model and the other one or more object models are different lane line models.
[0007] As a supplement or replacement for the above-mentioned scheme, in the above-mentioned method, the test data set is input into one or more object models other than the first object model, and obtaining the second detection result includes: inputting the test data set into a third lane line model to obtain a third detection result; inputting the test data set into a fourth lane line model to obtain a fourth detection result; and matching the third detection result and the fourth detection result based on the lane line gradient, the lane line intersection-union ratio (IOU) and the lane line confidence, so as to determine the second detection result.
[0008] As a supplement or replacement for the above-mentioned scheme, in the above-mentioned method, matching the third detection result and the fourth detection result based on the lane line gradient, the lane line intersection-and-union ratio (IOU) and the lane line confidence includes: determining the first lane line that matches each other and the second lane line that does not match each other in the third detection result and the fourth detection result based on the lane line gradient and the lane line intersection-and-union ratio (IOU), and the matched first lane line is included in the second detection result; and when the confidence of the unmatched second lane line is greater than a first threshold, the second lane line is also included in the second detection result; otherwise, the second lane line is moved out of the second detection result.
[0009] As a supplement or replacement for the above-mentioned scheme, in the above-mentioned method, the first detection result is matched with the second detection result to determine the test data in the test data set corresponding to the false detection and missed detection in the first detection result, including: based on the lane line gradient and the lane line intersection-union ratio (IOU), using the Hungarian matching algorithm to determine the unmatched third lane line in the first detection result and the second detection result; and determining the corresponding test data based on the unmatched third lane line.
[0010] According to another aspect of the present application, a data screening device for intelligent driving is provided, the device comprising: a first object model, for receiving an unlabeled test data set to generate a first detection result; one or more other object models different from the first object model, for receiving the test data set to generate a second detection result; and a matching device, for matching the first detection result with the second detection result to determine test data in the test data set corresponding to false detections and missed detections in the first detection result, wherein the test data corresponding to false detections and missed detections in the first detection result are used to retrain the first object model after labeling.
[0011] As a supplement or replacement for the above solution, in the above device, the data screening device is a lane line data screening device, and the first object model and the other one or more object models are different lane line models.
[0012] As a supplement or alternative to the above-mentioned scheme, in the above-mentioned device, the other one or more object models include: a third lane line model, used to receive the test data set to generate a third detection result; and a fourth lane line model, used to receive the test data set to generate a fourth detection result; wherein the matching device is configured to determine the second detection result by matching the third detection result and the fourth detection result based on lane line gradient, lane line intersection-union ratio (IOU) and lane line confidence.
[0013] As a supplement or alternative to the above-mentioned scheme, in the above-mentioned device, the matching device is configured to: determine the first lane line that matches each other and the second lane line that does not match each other in the third detection result and the fourth detection result based on the lane line gradient and the lane line intersection-union ratio (IOU), and the matched first lane line is included in the second detection result; and when the confidence of the unmatched second lane line is greater than a first threshold, the second lane line is also included in the second detection result; otherwise, the second lane line is moved out of the second detection result.
[0014] As a supplement or replacement for the above-mentioned scheme, in the above-mentioned device, the matching device is configured to: based on the lane line gradient and the lane line intersection-union ratio (IOU), use the Hungarian matching algorithm to determine the unmatched third lane line in the first detection result and the second detection result; and determine the corresponding test data based on the unmatched third lane line.
[0015] According to another aspect of the present application, a training method for a business model is provided, the method comprising: inputting an unlabeled test data set into the business model to obtain a first detection result; inputting the test data set into other large models different from the business model to obtain a second detection result; matching the first detection result with the second detection result to determine the test data in the test data set corresponding to the false detections and missed detections in the first detection result; marking the test data corresponding to the false detections and missed detections in the first detection result; and retraining the business model with the marked test data.
[0016] As a supplement or replacement for the above solution, in the above method, the business model is a lane line detection model in the camera.
[0017] As a supplement or replacement for the above-mentioned scheme, in the above-mentioned method, the first detection result is matched with the second detection result to determine the test data in the test data set corresponding to the false detection and missed detection in the first detection result, including: at least based on the lane line gradient and the lane line intersection-union ratio (IOU), using the Hungarian matching algorithm to determine the unmatched third lane line in the first detection result and the second detection result; and determining the corresponding test data based on the unmatched third lane line.
[0018] According to another aspect of the present application, a computer-readable storage medium is provided, wherein the medium includes instructions, and the instructions execute the above method when executed.
[0019] According to another aspect of the present application, a computer program product is provided, comprising a computer program, and when the computer program is executed by a processor, the method described above is implemented.
[0020] The data screening scheme of the embodiment of the present application obtains the first detection result and the second detection result respectively by inputting the unlabeled test data set into the first object model and one or more other object models, and matches the first detection result with the second detection result, thereby determining the test data in the test data set corresponding to the false detection and missed detection (FP, FN) in the first detection result, that is, finding the test data that is effective for improving the performance of the first object model (the test data can be used to retrain the first object model after marking). The above data screening helps to select accurate and difficult-to-detect sample examples for the first object model. In addition to improving efficiency by dozens of times, the data screening scheme of the embodiment of the present application can also obtain huge performance improvements with a small amount of data, which is especially necessary for lightweight models (models with limited capacity). BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The above and other objects and advantages of the present application will become more fully apparent from the following detailed description in conjunction with the accompanying drawings, wherein the same or similar elements are represented by the same reference numerals.
[0022] Figure 1 A schematic diagram of a process of a data screening method according to an embodiment of the present application is shown;
[0023] Figure 2 A schematic diagram of the structure of a data screening device according to an embodiment of the present application is shown;
[0024] Figure 3 A schematic diagram showing a flow chart of a training method for a business model according to an embodiment of the present application; and
[0025] Figure 4A schematic diagram showing the principle of lane line intersection-over-union (IOU) matching according to an embodiment of the present application is shown. DETAILED DESCRIPTION
[0026] Hereinafter, a data screening scheme according to various exemplary embodiments of the present application will be described in detail with reference to the accompanying drawings.
[0027] Figure 1 FIG. 1 is a flow chart of a data screening method 1000 according to an embodiment of the present application. Figure 1 As shown, the data screening method 1000 includes:
[0028] In step S110, an unlabeled test data set is input into a first object model to obtain a first detection result;
[0029] In step S120, the test data set is input into one or more other object models different from the first object model to obtain a second detection result; and
[0030] In step S130, the first detection result is matched with the second detection result to determine the test data in the test data set corresponding to the false detections and missed detections in the first detection result, wherein the test data corresponding to the false detections and missed detections in the first detection result are used to retrain the first object model after labeling.
[0031] In the context of the present application, the term "data screening method" means screening the data in the test data set that is effective for improving the object model. In one or more embodiments, the data screening method is a lane line data screening method, and the first object model and the other one or more object models are different lane line models. For example, the first object model is a lightweight lane line model, while the other object models are large models (with large model parameters, strong robustness, and high precision), including Anchor-based detection models, Rawclassification detection models, and key point-based detection models. Because the lightweight lane line model is subject to the computing power of the business chip, the model is generally small and its robustness is not as good as that of the large model. Therefore, by comparing the results of these models, test data that helps to improve the performance of the lightweight lane line model can be selected.
[0032] The unlabeled test data set can be input into multiple object models (e.g., multiple lane line models), which can solve the limitations of a single solution (object model). In one embodiment, step S120 includes: inputting the test data set into a third lane line model to obtain a third detection result; inputting the test data set into a fourth lane line model to obtain a fourth detection result; and matching the third detection result and the fourth detection result (e.g., using the Hungarian matching algorithm) based on the lane line gradient, the lane line intersection-over-union ratio (IOU), and the lane line confidence, thereby determining the second detection result. In the above embodiment, matching the third detection result and the fourth detection result based on the lane gradient, the lane intersection-and-union ratio (IOU) and the lane confidence may further include: determining the first lane line that matches each other and the second lane line that does not match each other in the third detection result and the fourth detection result based on the lane gradient and the lane intersection-and-union ratio (IOU), and the matched first lane line is included in the second detection result; and when the confidence of the unmatched second lane line is greater than a first threshold (e.g., 0.9), the second lane line is also included in the second detection result; otherwise, the second lane line is removed from the second detection result. That is, the third detection result and the fourth detection result can serve as the true value after fusion. If there is inconsistency in the detection results, for example, if an independent lane line does not match any lane line and its confidence is greater than the first threshold (e.g., 0.9), it can be considered as a new lane line, otherwise it is considered as a false positive FP and needs to be discarded. In this way, all other lane lines detected by one or more object models (e.g., large models) can be screened.
[0033] In one embodiment, matching based on lane line gradient includes: calculating the gradient average of the bottom 10 points of the lane line for different detection results; comparing whether the difference between the gradient averages of different detection results is within a threshold range, if so, the lane lines in the detection results are considered to be matched, otherwise they are considered to be mismatched.
[0034] Figure 4 FIG. 2 shows a schematic diagram of the principle of matching based on lane intersection-over-union (IOU) according to an embodiment of the present application. Figure 4 As shown, 410 shows the detected first lane line (wherein the black pixel value is 0 and the white pixel value is 1), and 415 shows the first extended lane line, which is obtained by extending the width of the detected first lane line (for example, a preset empirical value). Figure 4 , 420 shows the detected second lane line (wherein the black pixel value is 0 and the white pixel value is 1), and 425 shows the second extended lane line, which is obtained by expanding the width of the detected second lane line (for example, a preset empirical value).
[0035] In one embodiment, matching based on lane line intersection-and-merge ratio (IOU) includes: calculating a first overlap ratio between a detected first lane line (e.g., as shown in 410) and a second extended lane line (e.g., as shown in 425); calculating a second overlap ratio between a detected second lane line (e.g., as shown in 420) and a first extended lane line (e.g., as shown in 415), wherein the first extended lane line is obtained by extending the width of the detected first lane line (e.g., a preset empirical value); averaging the first overlap ratio and the second overlap ratio to obtain the lane line intersection-and-merge ratio (IOU); when the obtained lane line intersection-and-merge ratio (IOU) is greater than a preset threshold, determining that the first lane line matches the second lane line, otherwise the two lane lines are considered to be mismatched.
[0036] For example, the first overlap ratio pred_overlap may be determined by the following formula:
[0037]
[0038] The numerator is used to count the number of pixels whose product of the pixel value matrix pred_line representing the detected first lane line (for example, as shown in 410) and the pixel value matrix gt_expand representing the second extended lane line is greater than or equal to 1 (i.e., the number of matching pixels), and the denominator is used to count the number of white pixels (pixel value is 1) in the pixel value matrix pred_line.
[0039] For another example, the second overlap ratio gt_overlap may be determined by the following formula:
[0040]
[0041] The numerator is used to count the number of pixels whose product of the pixel value matrix gt_line representing the detected second lane line (for example, as shown in 420) and the pixel value matrix pred_expand representing the first extended lane line is greater than or equal to 1 (i.e., the number of matching pixels), and the denominator is used to count the number of white pixels (pixel value is 1) in the pixel value matrix gt_line.
[0042] The lane line intersection-overlap ratio (IOU) can be calculated using the formula: IOU = (pred_overlap + gt_overlap) / 2.
[0043] In one embodiment, step S130 includes: based on the lane gradient and the lane intersection-union ratio (IOU), using the Hungarian matching algorithm to determine the third lane line that is not matched in the first detection result and the second detection result; and determining the corresponding test data based on the third lane line that is not matched. In other words, after matching, it is necessary to determine whether there are false positives FP and missed detections FN. The data corresponding to these false positives FP and missed detections FN are sample data (or test data) that need to be re-input into the object model for iteration.
[0044] In addition, in this embodiment, the use of the Hungarian matching algorithm for matching (for example, determining the first detection result and the third lane line that is not matched in the second detection result) can reduce the complexity of the calculation and find the best matching solution between the lane lines. For example, the first object model outputs a, b, c, d, e, a total of 5 lane lines (that is, the first detection result includes 5 lane lines), and the other one or more object models (such as the large model) output A, B, C, D, E, F, a total of 6 lane lines (that is, the second detection result includes 6 lane lines). The matching method of the first detection result and the second detection result can be based on gradient, intersection-over-union (IOU), etc.
[0045] Figure 3 A flow chart of a training method 3000 for a business model according to an embodiment of the present application is shown. The training method 3000 includes the following steps:
[0046] In step S310, an unlabeled test data set is input into the business model to obtain a first detection result;
[0047] In step S320, the test data set is input into another large model different from the business model to obtain a second detection result;
[0048] In step S330, the first detection result is matched with the second detection result to determine the test data in the test data set corresponding to the false detection and missed detection in the first detection result;
[0049] In step S340, the test data corresponding to the false detection and missed detection in the first detection result are marked; and
[0050] In step S350, the business model is retrained using the labeled test data.
[0051] The above-mentioned training method 3000 for the business model can be applied to various types of models for intelligent driving, including but not limited to lane line detection models in cameras (such as multi-functional monocular camera mpc3).
[0052] In one embodiment, step S330 may include: using the Hungarian matching algorithm to determine the unmatched third lane line in the first detection result and the second detection result based at least on the lane line gradient and the lane line intersection-union ratio (IOU); and determining the corresponding test data based on the unmatched third lane line.
[0053] Using the Hungarian matching algorithm for matching can reduce the complexity of calculation. For example, the first object model outputs a, b, c, d, e, a total of 5 lane lines (i.e., the first detection result includes 5 lane lines), and one or more other object models (such as the large model) output A, B, C, D, E, F, a total of 6 lane lines (i.e., the second detection result includes 6 lane lines). The best matching solution between lane lines can be found based on the Hungarian matching algorithm (for example, A matches b, B matches c, C matches d, D matches e, E fails to match, and F matches a). The matching method of the first detection result and the second detection result can be based on gradient, intersection-over-union (IOU), etc.
[0054] In addition, it is easy for a person skilled in the art to understand that the data screening method 1000 and the training method 3000 for a business model provided in the above one or more embodiments of the present application can be implemented by a computer program. For example, the computer program is included in a computer program product, and when the computer program is executed by a processor, the data screening method 1000 and / or the training method 3000 for a business model of one or more embodiments of the present application are implemented. For another example, when a computer-readable storage medium (such as a USB flash drive) storing the computer program is connected to a computer, the data screening method 1000 and / or the training method 3000 for a business model of one or more embodiments of the present application can be executed by running the computer program.
[0055] refer to Figure 2 , Figure 2 FIG. 2 shows a schematic diagram of the structure of a data screening device 2000 according to an embodiment of the present application. Figure 2 As shown, the data screening device 2000 includes: a first object model 210, one or more other object models 220 different from the first object model 210, and a matching device 230. The first object model 210 is used to receive an unlabeled test data set to generate a first detection result; the one or more other object models 220 are used to receive the test data set to generate a second detection result; the matching device 230 is used to match the first detection result with the second detection result to determine the test data in the test data set corresponding to the false detection and missed detection in the first detection result, wherein the test data corresponding to the false detection and missed detection in the first detection result is used to retrain the first object model 210 after labeling.
[0056] In one embodiment, the data screening device 2000 is a lane line data screening device, and the first object model 210 and the other one or more object models 220 are different lane line models. For example, the first object model 210 is a lightweight lane line model, and the other one or more object models 220 are large models (large model parameters, strong robustness, high precision), including Anchor-based detection model, Raw classification detection model, and key point-based detection model. Since the lightweight lane line model is limited by the computing power of the business chip, the model is generally small and its robustness is not as good as that of the large model.
[0057] although Figure 2 Not shown, in one embodiment, the other one or more object models 220 may include multiple models, for example, a third lane line model and a fourth lane line model (those skilled in the art will appreciate that the other one or more object models 220 may include more lane line models, not limited to 2). The third lane line model is used to receive the test data set to generate a third detection result; the fourth lane line model is used to receive the test data set to generate a fourth detection result. In this embodiment, the matching device 230 is configured to determine the second detection result by matching the third detection result and the fourth detection result based on the lane line gradient, the lane line intersection-union ratio (IOU) and the lane line confidence.
[0058] In one embodiment, the matching device 230 is configured to: determine the first lane line that matches each other and the second lane line that does not match each other in the third detection result and the fourth detection result based on the lane line gradient and the lane line intersection-union ratio (IOU), and the matched first lane line is included in the second detection result; and when the confidence of the unmatched second lane line is greater than a first threshold, the second lane line is also included in the second detection result; otherwise, the second lane line is moved out of the second detection result.
[0059] In one embodiment, the matching device 230 is further configured to: determine the unmatched third lane line in the first detection result and the second detection result using the Hungarian matching algorithm based on the lane line gradient and the lane line intersection-union ratio (IOU); and determine the corresponding test data based on the unmatched third lane line.
[0060] In one or more embodiments, the above-mentioned data screening device 2000 can be integrated into various types of image processing devices, including but not limited to the multi-functional monocular camera mpc3.
[0061] In summary, the data screening scheme of the embodiment of the present application obtains the first detection result and the second detection result respectively by inputting the unlabeled test data set into the first object model and one or more other object models, and matches the first detection result with the second detection result, so as to determine the test data corresponding to the false detection and missed detection (FP, FN) in the test data set, that is, to find the test data that is effective for improving the performance of the first object model (the test data can be used to retrain the first object model after marking). The above data screening helps to select accurate and difficult to detect sample examples for the first object model. In addition to improving the efficiency by dozens of times, the data screening scheme of the embodiment of the present application can also obtain huge performance improvements with a small amount of data, which is especially necessary for lightweight models (models with limited capacity).
[0062] The above examples mainly illustrate the data screening scheme of the embodiments of the present application. Although only some of the embodiments of the present application are described, it should be understood by those skilled in the art that the present application can be implemented in many other forms without departing from its subject matter and scope. Therefore, the examples and embodiments shown are considered to be illustrative rather than restrictive, and the present application may cover various modifications and substitutions without departing from the spirit and scope of the present application as defined in the claims.
Claims
1. A data screening method for intelligent driving, characterized in that: The method comprises: Inputting an unlabeled test data set into a first object model to obtain a first detection result; Inputting the test data set into one or more other object models different from the first object model to obtain a second detection result; and Matching the first detection result with the second detection result to determine test data in the test data set corresponding to false detections and missed detections in the first detection result, The test data corresponding to the false detections and missed detections in the first detection result are used to retrain the first object model after labeling.
2. The method of claim 1, wherein: The data screening method is a lane line data screening method, and the first object model and the other one or more object models are different lane line models.
3. The method of claim 2, wherein: Inputting the test data set into one or more other object models different from the first object model to obtain a second detection result comprises: Inputting the test data set into a third lane line model to obtain a third detection result; Inputting the test data set into a fourth lane line model to obtain a fourth detection result; and Based on the lane line gradient, the lane line intersection-over-union (IOU) and the lane line confidence, the third detection result and the fourth detection result are matched to determine the second detection result.
4. The method of claim 3, wherein: Based on the lane line gradient, the lane line intersection-over-union ratio (IOU), and the lane line confidence, matching the third detection result and the fourth detection result includes: Determine a first lane line that matches each other and a second lane line that does not match each other in the third detection result and the fourth detection result based on the lane line gradient and the lane line intersection-over-union ratio (IOU), and include the matched first lane line in the second detection result; and When the confidence of the unmatched second lane line is greater than a first threshold, the second lane line is also included in the second detection result; otherwise, the second lane line is removed from the second detection result.
5. The method of claim 1, wherein: Matching the first detection result with the second detection result to determine test data in the test data set corresponding to false detections and missed detections in the first detection result includes: Based on the lane line gradient and the lane line intersection-over-union ratio (IOU), using the Hungarian matching algorithm to determine a third lane line that is not matched in the first detection result and the second detection result; and Corresponding test data is determined based on the unmatched third lane line.
6. A data screening device for intelligent driving, characterized in that: The device comprises: A first object model, configured to receive an unlabeled test data set to generate a first detection result; one or more other object models different from the first object model, for receiving the test data set to generate a second detection result; and A matching device is used to match the first detection result with the second detection result so as to determine the test data in the test data set corresponding to the false detections and missed detections in the first detection result, wherein the test data corresponding to the false detections and missed detections in the first detection result are used to retrain the first object model after labeling.
7. The device of claim 6, wherein: The data screening device is a lane line data screening device, and the first object model and the other one or more object models are different lane line models.
8. The device of claim 7, wherein: The other one or more object models include: a third lane line model, configured to receive the test data set so as to generate a third detection result; and a fourth lane line model, used to receive the test data set to generate a fourth detection result; Wherein, the matching device is configured to determine the second detection result by matching the third detection result and the fourth detection result based on the lane line gradient, the lane line intersection-over-union ratio (IOU) and the lane line confidence.
9. The device of claim 8, wherein: The matching device is configured as follows: Determine a first lane line that matches each other and a second lane line that does not match each other in the third detection result and the fourth detection result based on the lane line gradient and the lane line intersection-over-union ratio (IOU), and include the matched first lane line in the second detection result; and When the confidence of the unmatched second lane line is greater than a first threshold, the second lane line is also included in the second detection result; otherwise, the second lane line is removed from the second detection result.
10. The device of claim 6, wherein: The matching device is configured as follows: Based on the lane line gradient and the lane line intersection-over-union ratio (IOU), a Hungarian matching algorithm is used to determine a third lane line that is not matched in the first detection result and the second detection result; as well as Corresponding test data is determined based on the unmatched third lane line.
11. A training method for a business model, characterized in that: The method comprises: Inputting an unlabeled test data set into the business model to obtain a first detection result; Inputting the test data set into another large model different from the business model to obtain a second detection result; Matching the first detection result with the second detection result to determine test data in the test data set corresponding to false detections and missed detections in the first detection result; Marking the test data corresponding to the false detection and missed detection in the first detection result; and The business model is retrained using the labeled test data.
12. The method of claim 11, wherein: The business model is a lane line detection model in the camera.
13. The method of claim 11, wherein: Matching the first detection result with the second detection result to determine test data in the test data set corresponding to false detections and missed detections in the first detection result includes: Determine a third lane line that is not matched in the first detection result and the second detection result using a Hungarian matching algorithm based at least on a lane line gradient and a lane line intersection-over-union ratio (IOU); and Corresponding test data is determined based on the unmatched third lane line.
14. A computer-readable storage medium, characterized in that: The medium includes instructions that, when executed, perform the method of any one of claims 1 to 5, 11 to 13.
15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 5 and 11 to 13 is implemented.