Map matching result processing method and device, storage medium, equipment and vehicle
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- MOMENTA (SUZHOU) TECHNOLOGY CO LTD
- Filing Date
- 2022-11-04
- Publication Date
- 2026-07-07
AI Technical Summary
In existing technologies, road element matching errors occur during map matching, resulting in low accuracy of matching results and the final semantic map.
By determining the optimal relative transformation coordinates from each sub-map pair output by the target map matching model, setting the target error range, filtering out abnormal road element pairs, and adopting different filtering methods for different types of road elements, including directly filtering dashed lane lines and filtering solid lane lines through transformation points, combined with global consistency verification and targeted model training, the matching accuracy is improved.
It improves the accuracy of map matching results, enhances the accuracy of the semantic map obtained by final clustering, solves the long-tail problem, and ensures the effective application of the model in various road scenarios.
Smart Images
Figure CN118035753B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of map processing technology, and more specifically, to a method, apparatus, storage medium, device, and vehicle for processing map matching results. Background Technology
[0002] To improve the accuracy of semantic maps when generating them, a map matching model can be used to match road elements of multiple initial semantic maps generated from multiple crowdsourced data trips. Then, based on the relative positional relationship of the same road element, the two initial semantic maps that have successfully matched are shifted to reduce the positional offset between them. Finally, the multiple initial semantic maps that have completed matching and shifting are clustered to obtain the final semantic map.
[0003] However, in the matching results obtained by matching the initial semantic map according to the map matching model, there may be some road elements that are not matched correctly, which will result in low accuracy of the matching results and thus low accuracy of the semantic map obtained by the final clustering. Summary of the Invention
[0004] This application provides a method, apparatus, storage medium, device, and vehicle for processing map matching results, which can solve the problem of incorrect matching of certain road elements in map matching, resulting in low map matching accuracy.
[0005] The specific technical solution is as follows:
[0006] In a first aspect, embodiments of this application provide a method for processing map matching results, the method comprising:
[0007] For each submap pair output by the target map matching model, the relative transformation coordinates of each first road element pair contained in the submap pair are obtained. The target map matching model is used to divide the two maps to be matched into multiple submaps of the same number and match the submaps in the two maps to obtain the submap pair. The first road element pair includes road element matching pairs that are not lane lines.
[0008] The optimal relative transformation coordinates are determined based on the relative transformation coordinates of each first road element pair in the subgraph pair, and the coordinate transformation range within the target error range and including the optimal relative transformation coordinates is determined.
[0009] Filter out the first road element pairs and the second road element pairs whose relative transformation coordinates are not within the range of the coordinate transformation, wherein the second road element pair includes each dashed line segment matching pair in the lane line where the line presentation mode is dashed;
[0010] Select at least one road element point from any road element in the third road element pair, and transform each road element point according to the optimal relative transformation coordinates to obtain the transformation point corresponding to each road element point. The third road element pair includes lane line matching pairs with solid lines as the line presentation mode.
[0011] Based on the distance from each of the transformation points to the other road element in the third road element pair and the target error range, abnormal third road element pairs are filtered out.
[0012] As can be seen from the above scheme, the embodiments of this application first determine the optimal relative transformation coordinates from the relative transformation coordinates of each first road element pair in each sub-map pair output by the target map matching model, and determine the coordinate transformation range within the target error range and including the optimal relative transformation coordinates. For road element pairs that are not lane lines and each dashed line segment pair in the dashed lane lines, anomaly filtering can be performed directly through this coordinate transformation range. For solid lane line pairs (i.e., third road element pairs), at least one road element point can be selected from any road element in the third road element pair, and each road element point can be transformed according to the optimal relative transformation coordinates to obtain the transformation point corresponding to each road element point. Then, based on the distance from each transformation point to the other road element in the third road element pair and the target error range, abnormal third road element pairs are filtered. Thus, the embodiments of this application can use different methods for anomaly filtering according to different road elements, thereby improving the accuracy of the matching results and thus improving the accuracy of the semantic map obtained by the final clustering.
[0013] In a first possible implementation of the first aspect, after performing anomaly filtering for road element pairs on each of the sub-map pairs output by the target map matching model, the method further includes:
[0014] The subgraph pairs corresponding to the optimal relative transformation coordinates that do not satisfy the linear distribution are filtered.
[0015] As can be seen from the above scheme, after performing road element anomaly filtering on each sub-map pair, the embodiments of this application can achieve global consistency verification of the complete map by judging whether the optimal relative transformation coordinates corresponding to each sub-map pair conform to a linear distribution, thereby further improving the accuracy of the matching results and thus improving the accuracy of the semantic map obtained by the final clustering.
[0016] In a second possible implementation of the first aspect, filtering out anomalous third road element pairs based on the distance from each of the transformation points to the other road element in the third road element pair and the target error range includes:
[0017] Calculate the distance from each of the transition points to the other road element;
[0018] If any of the distances corresponding to at least one road element point is greater than or equal to the target ratio and is outside the target error range, the third road element pair is filtered.
[0019] As can be seen from the above scheme, the embodiments of this application only determine that the third road element pair is abnormal and filter it when the distance greater than or equal to the target ratio is outside the target error range. This can avoid the third road element pair being mistakenly considered abnormal because the distance corresponding to a small number of road element points is outside the target error range, thereby improving the accuracy of the third road element pair anomaly detection.
[0020] In a third possible implementation of the first aspect, obtaining the relative transformation coordinates of each first road element pair contained in the subgraph pair includes:
[0021] For each pair of first road elements contained in the subgraph pair, the midpoint of each road element in the first road element pair is obtained, and the coordinates of the difference between the midpoints are determined as the relative transformation coordinates.
[0022] In the fourth possible implementation of the first aspect, the training method for the target map matching model includes:
[0023] The target processing operation is executed repeatedly until there are no abnormal sub-map pairs in the sub-map pairs output by the second map matching model obtained in the target processing operation. Then, the obtained second map matching model is determined as the target map matching model.
[0024] The target processing operation includes:
[0025] The first map matching model is trained based on the training set to obtain the second map matching model and the sub-map pairs output by the second map matching model. The training set includes multiple maps, each map includes multiple sub-maps and each map has the same number of sub-maps. The sub-map pairs output by the second map matching model include multiple road element matching pairs.
[0026] Based on the precision and / or recall of each road element matching pair contained in the sub-map pair output by the second map matching model, determine whether the sub-map pair output by the second map matching model is an anomalous sub-map pair.
[0027] In the presence of the abnormal subgraph pairs, the second map matching model is trained based on the abnormal subgraph pairs and the target training round number to obtain a third map matching model, and the third map matching model is determined as the new first map matching model.
[0028] As can be seen from the above scheme, the embodiments of this application can first perform conventional training on the first map matching model using a complete training set to obtain the second map matching model and the sub-map pairs output by the second map matching model. Then, based on the precision and / or recall of each road element matching pair contained in the sub-map pairs, abnormal sub-map pairs are filtered out. The second map matching model is then trained specifically based on the abnormal sub-map pairs and the target training round number. The obtained third map matching model is then used as the new first map matching model for the next round of conventional training. Compared with conventional training using only a complete training set, the embodiments of this application, by alternating and iterating between conventional and targeted training, can ensure good results for most road scenarios while strengthening targeted training for a minority of road scenarios where conventional training is ineffective. This reduces the number of abnormal sub-map pairs, enabling even these minority road scenarios to achieve good results. In other words, the final target map matching model can be applied to any road scenario, thus avoiding the long-tail problem.
[0029] In a fifth possible implementation of the first aspect, determining whether a sub-map pair output by the second map matching model is an anomalous sub-map pair based on the precision and / or recall of each road element matching pair contained in the sub-map pair output by the second map matching model includes:
[0030] For each sub-map pair output by the second map matching model, based on the road element matching pairs and their corresponding ground truth values in each sub-map pair output by the second map matching model, calculate the precision and / or recall of each type of road element matching in each sub-map pair output by the second map matching model.
[0031] The precision of matching road elements of the first target type is determined as the precision of the sub-map pairs output by the second map matching model, and / or the recall of matching road elements of the second target type is determined as the recall of the sub-map pairs output by the second map matching model, wherein the first target type includes the road element type with the lowest precision in the sub-map pairs output by the second map matching model, and the second target type includes the road element type with the lowest recall in the sub-map pairs output by the second map matching model;
[0032] Based on the precision and / or recall of the sub-map pairs output by the second map matching model, determine whether the sub-map pairs output by the second map matching model are the abnormal sub-map pairs.
[0033] As can be seen from the above scheme, the embodiments of this application determine whether a sub-map pair is abnormal based on the precision of the road element type with the lowest precision rate in each sub-map pair, and / or the recall rate of the road element type with the lowest recall rate in the sub-map pair, rather than directly determining whether a sub-map pair is abnormal based on the mean of the precision rate and / or the mean of the recall rate of each road element type in the sub-map pair. This allows for targeted training on the road elements with the lowest precision rate and the road elements with the lowest recall rate, thereby reducing the number of abnormal sub-map pairs and thus obtaining a target map matching model with higher matching accuracy.
[0034] Secondly, embodiments of this application provide a map matching result processing apparatus, the apparatus comprising:
[0035] The acquisition unit is used to acquire the relative transformation coordinates of each first road element pair contained in each sub-map pair output by the target map matching model, wherein the target map matching model is used to divide the two maps to be matched into multiple sub-maps of the same number, and to match the sub-maps in the two maps to obtain the sub-map pair, wherein the first road element pair includes road element matching pairs that are not lane lines.
[0036] The determining unit is configured to determine the optimal relative transformation coordinates based on the relative transformation coordinates of each of the first road element pairs in the sub-graph pair, and to determine the coordinate transformation range within the target error range and including the optimal relative transformation coordinates;
[0037] The first filtering unit is used to filter the first road element pair and the second road element pair whose relative transformation coordinates are not within the range of the coordinate transformation, wherein the second road element pair includes each dashed line segment matching pair in the lane line where the line presentation mode is dashed;
[0038] A selection unit is used to select at least one road element point from any road element in a third road element pair, wherein the third road element pair includes lane line matching pairs whose line presentation mode is solid lines;
[0039] A conversion unit is used to convert each of the road element points according to the optimal relative conversion coordinates to obtain the conversion point corresponding to each of the road element points;
[0040] The second filtering unit is used to filter out abnormal third road element pairs based on the distance from each conversion point to another road element in the third road element pair and the target error range.
[0041] In a first possible implementation of the second aspect, the device further includes:
[0042] The third filtering unit is used to filter the sub-map pairs corresponding to the optimal relative transformation coordinates that do not satisfy the linear distribution after performing anomaly filtering for road element pairs on each sub-map pair output by the target map matching model.
[0043] In a second possible implementation of the second aspect, the second filtering unit includes:
[0044] A calculation module is used to calculate the distance from each of the transformation points to the other road element;
[0045] A filtering module is used to filter the third road element pair when there is a distance greater than or equal to the target ratio outside the target error range among the distances corresponding to the at least one road element point.
[0046] In a third possible implementation of the second aspect, the acquiring unit includes:
[0047] The acquisition module is used to acquire the midpoint of each road element in each of the first road element pairs contained in the subgraph pair;
[0048] The first determining module is used to determine the coordinates of the difference between the midpoints as the relative transformation coordinates.
[0049] In a fourth possible implementation of the second aspect, the device further includes:
[0050] The loop unit is used to repeatedly execute the target processing operation until there are no abnormal sub-map pairs in the sub-map pairs output by the second map matching model obtained in the target processing operation, and then the obtained second map matching model is determined as the target map matching model.
[0051] The target processing operation performed by the loop unit includes:
[0052] The first training module is used to train the first map matching model based on the training set to obtain the second map matching model and the sub-map pairs output by the second map matching model. The training set includes multiple maps, each map includes multiple sub-maps and each map has the same number of sub-maps. The sub-map pairs output by the second map matching model include multiple road element matching pairs.
[0053] The judgment module is used to determine whether the sub-map pair output by the second map matching model is an abnormal sub-map pair based on the precision and / or recall of each road element matching pair contained in the sub-map pair output by the second map matching model.
[0054] The second training module is used to train the second map matching model based on the abnormal subgraph pairs and the target training round number when the abnormal subgraph pairs exist, so as to obtain the third map matching model.
[0055] The second determining module is used to determine the third map matching model as the new first map matching model.
[0056] In a fifth possible implementation of the second aspect, the judgment module is configured to, for each sub-map pair output by the second map matching model, calculate the precision and / or recall of each type of road element matching in each sub-map pair output by the second map matching model, based on the road element matching pairs and their corresponding ground truth values in each sub-map pair output by the second map matching model; determine the precision of road element matching of the first target type as the precision of the sub-map pair output by the second map matching model, and / or determine the recall of road element matching of the second target type as the recall of the sub-map pair output by the second map matching model, wherein the first target type includes the road element type with the lowest precision in the sub-map pair output by the second map matching model, and the second target type includes the road element type with the lowest recall in the sub-map pair output by the second map matching model; and determine whether the sub-map pair output by the second map matching model is the abnormal sub-map pair based on the precision and / or recall of the sub-map pair output by the second map matching model.
[0057] As can be seen from the above scheme, the embodiments of this application first determine the optimal relative transformation coordinates from the relative transformation coordinates of each first road element pair in each sub-map pair output by the target map matching model, and determine the coordinate transformation range within the target error range and including the optimal relative transformation coordinates. For road element pairs that are not lane lines and each dashed line segment pair in the dashed lane lines, anomaly filtering can be performed directly through this coordinate transformation range. For solid lane line pairs (i.e., third road element pairs), at least one road element point can be selected from any road element in the third road element pair, and each road element point can be transformed according to the optimal relative transformation coordinates to obtain the transformation point corresponding to each road element point. Then, based on the distance from each transformation point to the other road element in the third road element pair and the target error range, abnormal third road element pairs are filtered. Thus, the embodiments of this application can use different methods for anomaly filtering according to different road elements, thereby improving the accuracy of the matching results and thus improving the accuracy of the semantic map obtained by the final clustering.
[0058] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method as described in any possible implementation of the first aspect.
[0059] Fourthly, embodiments of this application provide an electronic device, which includes:
[0060] One or more processors;
[0061] The processor is coupled to a storage device for storing one or more programs;
[0062] When one or more programs are executed by one or more processors, the electronic device performs the method as described in any possible implementation of the first aspect.
[0063] Fifthly, embodiments of this application provide a vehicle that includes the means as described in any possible implementation of the second aspect, or includes electronic equipment as described in the fourth aspect.
[0064] In a sixth aspect, embodiments of this application provide a computer program product containing instructions that, when executed on a computer or processor, cause the computer or processor to perform the method described in any possible implementation of the first aspect. Attached Figure Description
[0065] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.
[0066] Figure 1 A flowchart illustrating a method for processing map matching results provided in an embodiment of this application;
[0067] Figure 2 A block diagram illustrating the composition of a map matching result processing apparatus provided in an embodiment of this application;
[0068] Figure 3 This is a structural schematic diagram of a vehicle provided in an embodiment of this application. Detailed Implementation
[0069] 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 a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0070] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The terms "comprising" and "having," and any variations thereof, in the embodiments and drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.
[0071] Figure 1 This is a flowchart illustrating a method for processing map matching results. This method can be applied to electronic devices or computer equipment, specifically vehicles or servers, and may include the following steps:
[0072] S110: For each sub-map pair output by the target map matching model, obtain the relative transformation coordinates of each first road element pair contained in the sub-map pair.
[0073] The target map matching model is used to divide two maps to be matched into multiple sub-maps of equal number and to match the sub-maps in the two maps to obtain sub-map pairs. When dividing the two maps into multiple sub-maps, each target map can be directly divided into multiple sub-maps, or keyframes of each map can be obtained first (e.g., image frames extracted at certain distance intervals), and then each road element can be assigned to the nearest keyframe, so that multiple road elements near a keyframe form a sub-map. The distance between a road element and a keyframe is mainly determined by the road element's position on the map and the camera's position when the keyframe was captured, and is unrelated to the image content of the keyframe itself. Any map matching model mentioned in the embodiments of this application can be a GAT (Graph Attention Network) or other types of neural networks.
[0074] The first road element pair includes matching pairs of road elements that are not lane lines. The road elements that are not lane lines include traffic lights, road signs, streetlights, and markings on the road surface that are not lane lines (such as text markings). The relative transformation coordinates of the first road element pair include the transformation matrix between the first road element pairs, which includes a translation matrix.
[0075] To improve the calculation efficiency of relative transformation coordinates, each road element in the first road element pair can be treated as a point, and the coordinate difference between the two points can be used as the relative transformation coordinates. Specifically, for each first road element pair contained in the sub-graph pair, the midpoint of each road element in the first road element pair can be obtained, and the coordinate difference between the midpoints can be determined as the relative transformation coordinates. For traffic lights, road signs, streetlights, etc., which consist of poles and sign areas, when obtaining the midpoint of each road element in the first road element pair, the poles can be ignored, and the midpoint of the sign area can be directly selected as the midpoint of the road element.
[0076] S120: Determine the optimal relative transformation coordinates based on the relative transformation coordinates of each first road element pair in the subgraph pair, and determine the coordinate transformation range within the target error range that includes the optimal relative transformation coordinates.
[0077] The specific implementation method for determining the optimal relative transformation coordinates based on the relative transformation coordinates of each first road element pair in the subgraph pair includes: using the RANSAC (Random Sample Consensus) algorithm or the least squares method to calculate the relative transformation coordinates of each first road element pair in the subgraph pair to obtain the optimal relative transformation coordinates.
[0078] The target error range can be (-X, +X), the optimal relative transformation coordinate can be the midpoint of the coordinate transformation range, and the difference between the optimal relative transformation coordinate and the start and end points of the coordinate transformation range is Xm, for example, 2m.
[0079] S130: Filter the first and second road element pairs whose relative transformation coordinates are not within the coordinate transformation range.
[0080] The second road element pair includes each dashed line segment matching pair in the lane lines where the lines are presented as dashed lines. The relative transformation coordinates of the dashed line segment matching pair can be calculated as follows: first, take the midpoint of each dashed line segment in the dashed line segment matching pair, and then calculate the difference between the two midpoints as the relative transformation coordinates of the dashed line segment matching pair.
[0081] When the relative transformation coordinates of the first road element pair or the second road element pair are not within the coordinate transformation range, it means that the positions of the two road elements contained in the first road element pair or the second road element pair are significantly different and do not match. Therefore, the first road element pair or the second road element pair is an abnormal road element pair and can be filtered out.
[0082] S140: Select at least one road element point from any road element in the third road element pair, and transform each road element point according to the optimal relative transformation coordinates to obtain the transformation point corresponding to each road element point.
[0083] The third road element pair includes lane line matching pairs where the lines are presented as solid lines. The process of calculating the transformation point coordinates is the reverse of the aforementioned process of calculating relative transformation coordinates. For example, when the aforementioned method for calculating relative transformation coordinates belongs to the... Figure 1 The midpoint coordinates of the road element minus the coordinates of the land Figure 2 The coordinates of the midpoint of the road element in the map, and at least one selected road element point is located on the ground. Figure 1 In this case, the coordinates of the transformation point = the coordinates of the road element point - the optimal relative transformation coordinates; when the aforementioned method for calculating the relative transformation coordinates belongs to the area... Figure 1 The midpoint coordinates of the road element minus the coordinates of the land Figure 2 The coordinates of the midpoint of the road element in the map, and at least one selected road element point is located on the ground. Figure 2 In the middle, the coordinates of the transformation point = the coordinates of the road element point + the optimal relative transformation coordinates.
[0084] S150: Filter out abnormal third road element pairs based on the distance from each conversion point to the other road element in the third road element pair and the target error range.
[0085] Specifically, the distance from each conversion point to another road element can be calculated; if there is a distance greater than or equal to the target proportion in the distances corresponding to at least one road element point that is outside the target error range, the third road element pair is filtered out.
[0086] When at least one road element point has a distance greater than or equal to the target proportion outside the target error range, it indicates that most road element points have not been mapped to another road element through the optimal relative transformation coordinates. For example, most points on one lane line have not been mapped to another lane line through the optimal relative transformation coordinates. This means that the two road elements in the third road element pair output by the target map matching model are far apart and do not match, so they can be filtered out. The target proportion can be determined based on practical experience, for example, it can be 50%.
[0087] The above scheme determines that the third road element pair is abnormal and filters it only when the distance greater than or equal to the target ratio is outside the target error range. This avoids mistaking the third road element pair for anomaly due to the distance corresponding to a small number of road element points being outside the target error range, thereby improving the accuracy of anomaly detection for the third road element pair.
[0088] It should be added that the above method mainly considers whether the matching result is abnormal from the range of the subgraph pair. Therefore, the above process can be called local consistency check.
[0089] The map matching result processing method provided in this application first determines the optimal relative transformation coordinates from the relative transformation coordinates of each first road element pair in each sub-map pair output by the target map matching model, and determines the coordinate transformation range within the target error range and including the optimal relative transformation coordinates. For non-lane line road element pairs and each dashed line segment pair in dashed lane lines, anomaly filtering can be performed directly through this coordinate transformation range. For solid line lane line pairs (i.e., third road element pairs), at least one road element point can be selected from any road element in the third road element pair, and each road element point can be transformed according to the optimal relative transformation coordinates to obtain the transformation point corresponding to each road element point. Then, based on the distance from each transformation point to another road element in the third road element pair and the target error range, abnormal third road element pairs are filtered. Therefore, this application embodiment can use different methods for anomaly filtering according to different road elements, thereby improving the accuracy of the matching results and ultimately improving the accuracy of the semantic map obtained by clustering.
[0090] In one implementation, to further improve the accuracy of the matching results, this application embodiment can filter out the sub-map pairs corresponding to the optimal relative transformation coordinates that do not satisfy the linear distribution after performing anomaly filtering for road element pairs on each sub-map pair output by the target map matching model.
[0091] Because the trajectory is continuous, all optimal relative transformation coordinates between two maps have a certain linear distribution pattern. Therefore, outliers with jumps can be eliminated, that is, submap pairs corresponding to optimal relative transformation coordinates that do not satisfy the linear distribution can be filtered out.
[0092] In this embodiment of the application, after performing road element anomaly filtering on each sub-map pair, the global consistency verification of the complete map can be achieved by judging whether the optimal relative transformation coordinates corresponding to each sub-map pair conform to a linear distribution. This further improves the accuracy of the matching results and thus improves the accuracy of the semantic map obtained by the final clustering.
[0093] In one implementation, in related technologies, a map matching model is typically trained using only the complete training set until convergence, yielding the final desired map matching model. However, a map matching model trained using only the complete training set may perform well for most road scenes but poorly for a subset of road scenes; this is known as the long-tail problem. The subset of road scenes where the model performs poorly can also be understood as challenging scenes. To address the long-tail problem, embodiments of this application can employ the following method to train the target map matching model:
[0094] The target processing operation is executed repeatedly until there are no abnormal submap pairs in the submap pairs output by the second map matching model obtained in the target processing operation. Then, the obtained second map matching model is determined as the target map matching model.
[0095] The target processing operation includes steps A1-A3:
[0096] A1. Train the first map matching model based on the training set to obtain the second map matching model and the sub-map pairs output by the second map matching model.
[0097] The training set includes multiple maps, each containing multiple submaps of the same number. The submap pairs output by the second map matching model include multiple road element matching pairs. The training set can also contain ground truth values for road element matching pairs, indicating which road elements actually refer to the same road element in the real world. Although the training set includes multiple maps, during training, each pair of maps is matched separately. "Training the first map matching model based on the training set" can be understood as performing regular training on the first map matching model; that is, the training set used for training the first map matching model includes all training samples.
[0098] When training for the first time based on the training set, the first map matching model used is the initial map matching model. After training the first map matching model multiple times based on the training set and reaching the convergence condition, a second map matching model and the sub-map pairs output by the second map matching model are obtained. The convergence condition includes, but is not limited to: the number of training rounds reaches a preset number of training rounds, or the loss value calculated based on the positional distance difference between the currently output sub-map pairs is less than or equal to a preset loss threshold.
[0099] To improve the matching accuracy of the second map matching model, this embodiment of the application may first divide the training set into multiple training subsets, and then iteratively train the first map matching model based on the multiple training subsets to obtain the second map matching model.
[0100] Iterative training of the first map matching model based on multiple training subsets includes: training the first map matching model based on the first training subset to obtain the fourth map matching model, training the fourth map matching model based on the second training subset to obtain the fifth map matching model, training the fourth map matching model based on the third training subset to obtain the sixth map matching model, and so on, until the second map matching model is obtained by training based on the last training subset.
[0101] This batch-based iterative training method allows the matching accuracy of the map matching model obtained in the later training to be higher than that of the map matching model obtained in the earlier training, and the convergence speed of the later training is higher than that of the earlier training. As a result, the final second map matching model has a higher matching accuracy than the second map matching model obtained by directly using the training set.
[0102] A2. Based on the precision and / or recall of each road element matching pair contained in the sub-map pair output by the second map matching model, determine whether the sub-map pair output by the second map matching model is an anomalous sub-map pair.
[0103] Precision is relative to the prediction result; it represents the percentage of samples predicted as positive that are actually correct. Recall is relative to the number of samples; it represents the percentage of positive samples that are correctly predicted. Anomaly subgraph pairs can be understood as subgraph pairs representing challenging scenarios.
[0104] Specifically, for each sub-map pair output by the second map matching model, based on the road element matching pairs and their corresponding ground truth values in each sub-map pair output by the second map matching model, the precision and / or recall of each type of road element matching in each sub-map pair output by the second map matching model are calculated; the precision of the road element matching of the first target type is determined as the precision of the sub-map pair output by the second map matching model, and / or the recall of the road element matching of the second target type is determined as the recall of the sub-map pair output by the second map matching model, wherein the first target type includes the road element type with the lowest precision in the sub-map pair output by the second map matching model, and the second target type includes the road element type with the lowest recall in the sub-map pair output by the second map matching model; based on the precision and / or recall of the sub-map pair output by the second map matching model, it is determined whether the sub-map pair output by the second map matching model is an anomalous sub-map pair.
[0105] Road elements include road signs, traffic lights, streetlights, lane markings, and non-lane markings on the road surface. A subgraph pair may contain multiple types of road elements, such as road signs, traffic lights, streetlights, lane markings, and non-lane markings on the road surface. Precision and / or recall can be calculated separately for each type of road element. The first target type and the second target type may be the same or different.
[0106] It should be added that when the precision of matching road elements of the first target type is determined as the precision of the subgraph pair, the recall of matching road elements of the first target type can be determined as the recall of the subgraph pair; when the recall of matching road elements of the second target type is determined as the recall of the subgraph pair, the precision of matching road elements of the second target type can be determined as the precision of the subgraph pair.
[0107] When the quality requirements for the target map matching model are relatively low, this step can be implemented as follows: if the precision of the sub-map pair is less than or equal to the precision threshold and the recall of the sub-map pair is less than or equal to the recall threshold, determine that the sub-map pair is an anomalous sub-map pair; if the precision of the sub-map pair is greater than the precision threshold or the recall of the sub-map pair is greater than the recall threshold, determine that the sub-map pair is not an anomalous sub-map pair.
[0108] When the quality requirements for the target map matching model are high, this step can be implemented as follows: if the precision of the sub-map pair is less than or equal to the precision threshold, or the recall of the sub-map pair is less than or equal to the recall threshold, determine that the sub-map pair is an anomalous sub-map pair; if the precision of the sub-map pair is greater than the precision threshold and the recall of the sub-map pair is greater than the recall threshold, determine that the sub-map pair is not an anomalous sub-map pair.
[0109] A3. In the case of abnormal subgraph pairs, the second map matching model is trained according to the abnormal subgraph pairs and the target training round number to obtain the third map matching model, and the third map matching model is determined as the new first map matching model.
[0110] "Training the second map matching model based on the abnormal subgraph pairs and the target training rounds" can be understood as conducting targeted training on the second map matching model. That is, the training set used when training the second map matching model is a training set that only includes subgraph matching pairs of difficult scenes.
[0111] When conducting targeted training to obtain the third map matching model, training can be performed either on the abnormal road elements within the abnormal sub-map pairs or on each type of road element contained in the abnormal sub-map pairs. Specifically, the third map matching model can be obtained by training the second map matching model based on the first and second target type road elements contained in the abnormal sub-map pairs, along with the target number of training rounds; or, the third map matching model can be obtained by training the second map matching model based on each type of road element contained in the abnormal sub-map pairs, along with the target number of training rounds. The target number of training rounds is the ratio of the number of sub-maps in each map data package to the number of abnormal sub-map pairs. Fewer abnormal sub-map pairs require a larger target number of training rounds, and more abnormal sub-map pairs require a smaller target number of training rounds, thus minimizing the total number of target training rounds while achieving the desired effect through targeted training.
[0112] This embodiment first performs conventional training on a first map matching model using a complete training set to obtain a second map matching model and its output sub-map pairs. Then, based on the precision and / or recall of each road element matching pair within the sub-map pairs, abnormal sub-map pairs are identified. The second map matching model is then trained specifically based on these abnormal sub-map pairs and the target training round number. The resulting third map matching model is then used as the new first map matching model for the next round of conventional training. Compared to conventional training using only a complete training set, this embodiment, by alternating and iterating between conventional and targeted training, ensures good results for most road scenarios while strengthening targeted training for a minority of road scenarios where conventional training is ineffective. This reduces the number of abnormal sub-map pairs, allowing even these minority road scenarios to achieve good results. In other words, the final target map matching model can be applied to any road scenario, thus avoiding the long-tail problem. Furthermore, the anomaly of a sub-map pair can be determined based on the precision of the road element type with the lowest precision in each sub-map pair, and / or the recall of the road element type with the lowest recall in that sub-map pair, rather than directly based on the mean of the precision and / or the mean of the recall of each road element type in the sub-map pair. This allows for targeted training on the road elements with the lowest precision and the lowest recall, thereby reducing the number of anomaly sub-map pairs and resulting in a target map matching model with higher matching accuracy.
[0113] Corresponding to the above method embodiments, another embodiment of this application provides a map matching result processing apparatus, such as... Figure 2 As shown, the device includes:
[0114] The acquisition unit 210 is used to acquire the relative transformation coordinates of each first road element pair contained in each sub-map pair output by the target map matching model. The target map matching model is used to divide the two maps to be matched into multiple sub-maps of the same number and match the sub-maps in the two maps to obtain the sub-map pair. The first road element pair includes road element matching pairs that are not lane lines.
[0115] The determining unit 220 is configured to determine the optimal relative transformation coordinates based on the relative transformation coordinates of each of the first road element pairs in the sub-graph pair, and to determine the coordinate transformation range within the target error range and including the optimal relative transformation coordinates.
[0116] The first filtering unit 230 is used to filter the first road element pair and the second road element pair whose relative transformation coordinates are not within the range of the coordinate transformation, wherein the second road element pair includes each dashed line segment matching pair in the lane line where the line presentation mode is dashed line;
[0117] The selection unit 240 is used to select at least one road element point from any road element in the third road element pair, wherein the third road element pair includes lane line matching pairs whose line presentation mode is solid lines;
[0118] The conversion unit 250 is used to convert each of the road element points according to the optimal relative conversion coordinates to obtain the conversion point corresponding to each of the road element points.
[0119] The second filtering unit 260 is used to filter out abnormal third road element pairs based on the distance from each conversion point to another road element in the third road element pair and the target error range.
[0120] In one possible implementation, the device further includes:
[0121] The third filtering unit is used to filter the sub-map pairs corresponding to the optimal relative transformation coordinates that do not satisfy the linear distribution after performing anomaly filtering for road element pairs on each sub-map pair output by the target map matching model.
[0122] In a second possible implementation of the second aspect, the second filtering unit 260 includes:
[0123] A calculation module is used to calculate the distance from each of the transformation points to the other road element;
[0124] A filtering module is used to filter the third road element pair when there is a distance greater than or equal to the target ratio outside the target error range among the distances corresponding to the at least one road element point.
[0125] In one possible implementation, the acquisition unit 210 includes:
[0126] The acquisition module is used to acquire the midpoint of each road element in each of the first road element pairs contained in the subgraph pair;
[0127] The first determining module is used to determine the coordinates of the difference between the midpoints as the relative transformation coordinates.
[0128] In one possible implementation, the device further includes:
[0129] The loop unit is used to repeatedly execute the target processing operation until there are no abnormal sub-map pairs in the sub-map pairs output by the second map matching model obtained in the target processing operation, and then the obtained second map matching model is determined as the target map matching model.
[0130] The target processing operation performed by the loop unit includes:
[0131] The first training module is used to train the first map matching model based on the training set to obtain the second map matching model and the sub-map pairs output by the second map matching model. The training set includes multiple maps, each map includes multiple sub-maps and each map has the same number of sub-maps. The sub-map pairs output by the second map matching model include multiple road element matching pairs.
[0132] The judgment module is used to determine whether the sub-map pair output by the second map matching model is an abnormal sub-map pair based on the precision and / or recall of each road element matching pair contained in the sub-map pair output by the second map matching model.
[0133] The second training module is used to train the second map matching model based on the abnormal subgraph pairs and the target training round number when the abnormal subgraph pairs exist, so as to obtain the third map matching model.
[0134] The second determining module is used to determine the third map matching model as the new first map matching model.
[0135] In one possible implementation, the judgment module is configured to, for each sub-map pair output by the second map matching model, calculate the precision and / or recall of each type of road element matching in each sub-map pair output by the second map matching model, based on the road element matching pairs and their corresponding ground truth values in each sub-map pair output by the second map matching model; determine the precision of road element matching of a first target type as the precision of the sub-map pair output by the second map matching model, and / or determine the recall of road element matching of a second target type as the recall of the sub-map pair output by the second map matching model, wherein the first target type includes the road element type with the lowest precision in the sub-map pair output by the second map matching model, and the second target type includes the road element type with the lowest recall in the sub-map pair output by the second map matching model; and determine whether the sub-map pair output by the second map matching model is the abnormal sub-map pair based on the precision and / or recall of the sub-map pair output by the second map matching model.
[0136] The map matching result processing apparatus provided in this application first determines the optimal relative transformation coordinates from the relative transformation coordinates of each first road element pair in each sub-map pair output by the target map matching model, and determines the coordinate transformation range within the target error range and including the optimal relative transformation coordinates. For non-lane line road element pairs and each dashed line segment pair in dashed lane lines, anomaly filtering can be performed directly through this coordinate transformation range. For solid line lane line pairs (i.e., third road element pairs), at least one road element point can be selected from any road element in the third road element pair, and each road element point can be transformed according to the optimal relative transformation coordinates to obtain the transformation point corresponding to each road element point. Then, based on the distance from each transformation point to another road element in the third road element pair and the target error range, abnormal third road element pairs are filtered. Therefore, this application embodiment can use different methods for anomaly filtering according to different road elements, thereby improving the accuracy of the matching results and ultimately improving the accuracy of the semantic map obtained by clustering.
[0137] Based on the above method embodiments, another embodiment of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any of the above embodiments.
[0138] Based on the above method embodiments, another embodiment of this application provides an electronic device or computer device, including:
[0139] One or more processors;
[0140] The processor is coupled to a storage device for storing one or more programs;
[0141] When the one or more programs are executed by the one or more processors, the electronic device or computer device performs the method as described in any of the above embodiments.
[0142] Based on the above method embodiments, another embodiment of this application provides a vehicle that includes the apparatus as described in any of the above embodiments, or includes electronic devices as described above.
[0143] like Figure 3As shown, the vehicle includes a CPU (Central Processing Unit) 310, a T-Box (Telematics Box) 320, a camera 330, and a positioning module 340. The T-Box 320 can act as a gateway to communicate with the server. The CPU 310 can obtain a target map matching model without long-tail problems by executing the training method described above for the map matching model. Alternatively, the CPU 310 can also send the training set to the server via the T-Box 320, allowing the server to execute the training method and obtain a target map matching model without long-tail problems. Or, technicians can directly store the training set on the server, allowing the server to execute the training method and obtain a target map matching model without long-tail problems. The camera 330 is used to acquire road images, and the positioning module 340 is used to obtain the vehicle's geographical location. When the vehicle is a data acquisition vehicle used to generate a map, the CPU 310 is also used to acquire information such as the road images acquired by the camera 330 and the vehicle's geographical location acquired by the positioning module 340, and generate a map based on this information, or send this information to the server via the T-Box 320 to generate a map.
[0144] Based on the above embodiments, another embodiment of this application provides a computer program product, which includes instructions that, when executed on a computer or processor, cause the computer or processor to perform the method described in any of the above embodiments.
[0145] The above-described apparatus embodiments correspond to the method embodiments and have the same technical effects. For detailed descriptions, please refer to the method embodiments. The apparatus embodiments are derived from the method embodiments; detailed descriptions can be found in the method embodiments section, and will not be repeated here. Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application.
[0146] Those skilled in the art will understand that the modules in the apparatus of the embodiments can be distributed in the apparatus of the embodiments as described in the embodiments, or they can be located in one or more devices different from this embodiment with corresponding changes. The modules of the above embodiments can be combined into one module, or they can be further divided into multiple sub-modules.
[0147] 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A method for processing map matching results, characterized in that, The method includes: For each submap pair output by the target map matching model, the relative transformation coordinates of each first road element pair contained in the submap pair are obtained. The target map matching model is used to divide the two maps to be matched into multiple submaps of the same number and match the submaps in the two maps to obtain the submap pair. The first road element pair includes road element matching pairs that are not lane lines. The optimal relative transformation coordinates are determined based on the relative transformation coordinates of each first road element pair in the subgraph pair, and the coordinate transformation range within the target error range and including the optimal relative transformation coordinates is determined. Filter out the first road element pairs and the second road element pairs whose relative transformation coordinates are not within the range of the coordinate transformation, wherein the second road element pair includes each dashed line segment matching pair in the lane line where the line presentation mode is dashed; At least one road element point is selected from any road element in the third road element pair, and each road element point is transformed according to the optimal relative transformation coordinates to obtain the transformation point corresponding to each road element point. The third road element pair includes lane line matching pairs with solid lines. Based on the distance from each of the transformation points to the other road element in the third road element pair and the target error range, abnormal third road element pairs are filtered out.
2. The method according to claim 1, characterized in that, After performing anomaly filtering for road element pairs on each sub-map pair output by the target map matching model, the method further includes: The subgraph pairs corresponding to the optimal relative transformation coordinates that do not satisfy the linear distribution are filtered.
3. The method according to claim 1, characterized in that, The step of filtering out anomalous third road element pairs based on the distance from each conversion point to another road element in the third road element pair and the target error range includes: Calculate the distance from each of the transition points to the other road element; If any of the distances corresponding to at least one road element point is greater than or equal to the target ratio and is outside the target error range, the third road element pair is filtered.
4. The method according to claim 1, characterized in that, The step of obtaining the relative transformation coordinates of each first road element pair contained in the subgraph pair includes: For each pair of first road elements contained in the subgraph pair, the midpoint of each road element in the first road element pair is obtained, and the coordinates of the difference between the midpoints are determined as the relative transformation coordinates.
5. The method according to any one of claims 1-4, characterized in that, The training method for the target map matching model includes: The target processing operation is executed repeatedly until there are no abnormal sub-map pairs in the sub-map pairs output by the second map matching model obtained in the target processing operation. Then, the obtained second map matching model is determined as the target map matching model. The target processing operation includes: The first map matching model is trained based on the training set to obtain the second map matching model and the sub-map pairs output by the second map matching model. The training set includes multiple maps, each map includes multiple sub-maps and each map has the same number of sub-maps. The sub-map pairs output by the second map matching model include multiple road element matching pairs. Based on the precision and / or recall of each road element matching pair contained in the sub-map pair output by the second map matching model, determine whether the sub-map pair output by the second map matching model is an anomalous sub-map pair. In the presence of the abnormal subgraph pairs, the second map matching model is trained based on the abnormal subgraph pairs and the target training round number to obtain a third map matching model, and the third map matching model is determined as the new first map matching model.
6. The method according to claim 5, characterized in that, The step of determining whether a sub-map pair output by the second map matching model is an anomalous sub-map pair based on the precision and / or recall of each road element matching pair contained in the sub-map pair output by the second map matching model includes: For each sub-map pair output by the second map matching model, based on the road element matching pairs and their corresponding ground truth values in each sub-map pair output by the second map matching model, calculate the precision and / or recall of each type of road element matching in each sub-map pair output by the second map matching model. The precision of matching road elements of the first target type is determined as the precision of the sub-map pairs output by the second map matching model, and / or the recall of matching road elements of the second target type is determined as the recall of the sub-map pairs output by the second map matching model, wherein the first target type includes the road element type with the lowest precision in the sub-map pairs output by the second map matching model, and the second target type includes the road element type with the lowest recall in the sub-map pairs output by the second map matching model; Based on the precision and / or recall of the sub-map pairs output by the second map matching model, determine whether the sub-map pairs output by the second map matching model are the abnormal sub-map pairs.
7. A processing device for map matching results, characterized in that, The device includes: The acquisition unit is used to acquire the relative transformation coordinates of each first road element pair contained in each sub-map pair output by the target map matching model, wherein the target map matching model is used to divide the two maps to be matched into multiple sub-maps of the same number, and to match the sub-maps in the two maps to obtain the sub-map pair, wherein the first road element pair includes road element matching pairs that are not lane lines. The determining unit is configured to determine the optimal relative transformation coordinates based on the relative transformation coordinates of each of the first road element pairs in the sub-graph pair, and to determine the coordinate transformation range within the target error range and including the optimal relative transformation coordinates; The first filtering unit is used to filter the first road element pair and the second road element pair whose relative transformation coordinates are not within the range of the coordinate transformation, wherein the second road element pair includes each dashed line segment matching pair in the lane line where the line presentation mode is dashed; A selection unit is used to select at least one road element point from any road element in a third road element pair, wherein the third road element pair includes lane line matching pairs whose line presentation mode is solid lines; A conversion unit is used to convert each of the road element points according to the optimal relative conversion coordinates to obtain the conversion point corresponding to each of the road element points; The second filtering unit is used to filter out abnormal third road element pairs based on the distance from each conversion point to another road element in the third road element pair and the target error range.
8. The apparatus according to claim 7, characterized in that, The device further includes: The third filtering unit is used to filter the sub-map pairs corresponding to the optimal relative transformation coordinates that do not satisfy the linear distribution after performing anomaly filtering for road element pairs on each sub-map pair output by the target map matching model.
9. The apparatus according to claim 7, characterized in that, The second filtering unit includes: A calculation module is used to calculate the distance from each of the transformation points to the other road element; A filtering module is used to filter the third road element pair when there is a distance greater than or equal to the target ratio outside the target error range among the distances corresponding to the at least one road element point.
10. The apparatus according to claim 7, characterized in that, The acquisition unit includes: The acquisition module is used to acquire the midpoint of each road element in each of the first road element pairs contained in the subgraph pair; The first determining module is used to determine the coordinates of the difference between the midpoints as the relative transformation coordinates.
11. The apparatus according to any one of claims 7-10, characterized in that, The device further includes: The loop unit is used to repeatedly execute the target processing operation until there are no abnormal sub-map pairs in the sub-map pairs output by the second map matching model obtained in the target processing operation, and then the obtained second map matching model is determined as the target map matching model. The target processing operation performed by the loop unit includes: The first training module is used to train the first map matching model based on the training set to obtain the second map matching model and the sub-map pairs output by the second map matching model. The training set includes multiple maps, each map includes multiple sub-maps and each map has the same number of sub-maps. The sub-map pairs output by the second map matching model include multiple road element matching pairs. The judgment module is used to determine whether the sub-map pair output by the second map matching model is an abnormal sub-map pair based on the precision and / or recall of each road element matching pair contained in the sub-map pair output by the second map matching model. The second training module is used to train the second map matching model based on the abnormal subgraph pairs and the target training round number when the abnormal subgraph pairs exist, so as to obtain the third map matching model. The second determining module is used to determine the third map matching model as the new first map matching model.
12. The apparatus according to claim 11, characterized in that, The judgment module is configured to, for each sub-map pair output by the second map matching model, calculate the precision and / or recall of each type of road element matching in each sub-map pair output by the second map matching model, based on the road element matching pairs and their corresponding ground truth values in each sub-map pair output by the second map matching model; determine the precision of road element matching of the first target type as the precision of the sub-map pair output by the second map matching model, and / or determine the recall of road element matching of the second target type as the recall of the sub-map pair output by the second map matching model, wherein the first target type includes the road element type with the lowest precision in the sub-map pair output by the second map matching model, and the second target type includes the road element type with the lowest recall in the sub-map pair output by the second map matching model; and determine whether the sub-map pair output by the second map matching model is the abnormal sub-map pair based on the precision and / or recall of the sub-map pair output by the second map matching model.
13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.
14. An electronic device, characterized in that, The electronic device includes: One or more processors; The processor is coupled to a storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the electronic device performs the method as described in any one of claims 1-6.
15. A vehicle, characterized in that, The vehicle includes the device as described in any one of claims 7-12, or the electronic device as described in claim 14.