Data labeling method and device, computer readable storage medium and electronic equipment

Abstract

The invention provides a data annotation method and device, a computer readable storage medium and electronic equipment, and belongs to the technical field of automatic driving. According to the embodiment of the invention, the driving intention of at least one to-be-labeled obstacle in the driving data packet is classified and labeled through the preset labeling rule, so that the driving intention of the to-be-labeled obstacle can be efficiently labeled, the efficiency and accuracy of data labeling are effectively improved, and the consistency of labeling results is ensured; meanwhile, by designing a set of complete data verification and recovery mechanism, when the labeling rule is updated, effective data which is wrongly labeled can be recycled, incremental accumulation of qualified data is realized, and the data utilization rate is improved.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving technology, and particularly to a data annotation method, apparatus, computer-readable storage medium, and electronic device. Background Art

[0002] In the related art, driving intention prediction is an important function in the field of autonomous driving. It mainly analyzes driving data to obtain driving intention labels, so as to predict driving intentions and avoid safety accidents such as vehicle collisions.

[0003] Currently, the conventional driving intention recognition method mainly uses a large amount of driving data with lane-changing behaviors to train a prediction model, and then uses the trained prediction model to predict whether there is a lane-changing behavior of other vehicles around the vehicle. Therefore, to ensure the training effect of the prediction model, there is a large annotation requirement for a large amount of driving data. However, currently, the manual annotation method is mainly used to annotate the driving intention of the above driving data. This method is not only time-consuming and laborious, but also easily affected by human subjectivity, resulting in poor annotation efficiency, accuracy, and consistency. At the same time, the mis-annotated driving data is usually discarded and cannot be effectively recovered, resulting in low utilization rate of effective data. Summary of the Invention

[0004] The present application provides a data annotation method, apparatus, computer-readable storage medium, and electronic device to solve the problems of low efficiency, low accuracy, poor consistency, and low data utilization rate existing in the traditional driving intention annotation method.

[0005] To solve the above problems, the present application adopts the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a data annotation method, and the method includes:

[0007] Obtain a driving data packet collected by a target vehicle;

[0008] Based on a preset annotation rule, classify and annotate the driving intention of at least one obstacle to be annotated in the driving data packet to obtain a first annotation result file;

[0009] Obtain the first qualified annotation data and the first unqualified annotation data returned by the manual verification module for the first annotation result file, and store the first qualified annotation data in the qualified data set;

[0010] When it is detected that the annotation rule is updated, update the driving intention corresponding to the first unqualified annotation data based on the updated annotation rule to obtain a second annotation result file;

[0011] Obtain the second qualified annotation data returned by the manual verification module for the second annotation result file, and store the second qualified annotation data in the qualified data set.

[0012] In an embodiment of the present application, the annotation rules include an effective trajectory determination sub-rule, an effective lane line determination sub-rule, and a driving intention determination sub-rule;

[0013] The step of classifying and annotating the driving intention of at least one obstacle to be annotated in the driving data packet based on a preset annotation rule to obtain a first annotation result file includes:

[0014] Based on the effective trajectory determination sub-rule, determine the effective driving trajectory of the obstacle to be annotated in the driving data packet;

[0015] Based on the effective lane line determination sub-rule, determine at least one effective lane line in the driving data packet;

[0016] Based on the driving intention determination sub-rule, determine at least one interaction relationship sequence between the effective driving trajectory of each obstacle to be annotated and at least one of the effective lane lines, and based on at least one of the interaction relationship sequences, determine the respective driving intention corresponding to each obstacle to be annotated;

[0017] Generate a driving intention label corresponding to each obstacle to be annotated to obtain the first annotation result file.

[0018] In an embodiment of the present application, the step of determining the effective driving trajectory of the obstacle to be annotated in the driving data packet based on the effective trajectory determination sub-rule includes:

[0019] Based on the effective trajectory determination sub-rule, determine a trajectory point number threshold, a trajectory curvature threshold, and a continuity parameter threshold;

[0020] For any obstacle to be annotated, determine a plurality of trajectory points of the obstacle to be annotated in a plurality of consecutive image frames of the driving data packet, and a trajectory curve and a continuity parameter corresponding to the plurality of trajectory points; wherein, the continuity parameter characterizes the continuity degree between the plurality of trajectory points;

[0021] When the number of trajectory points is greater than the trajectory point number threshold, and the curvature of the trajectory curve is less than the trajectory curvature threshold, and the continuity parameter is greater than the continuity parameter threshold, determine the trajectory curve of the obstacle to be annotated as the effective driving trajectory.

[0022] In an embodiment of the present application, the step of determining the effective lane line in the driving data packet based on the effective lane line determination sub-rule includes:

[0023] Based on the effective lane line determination sub - rule, determine the lane line length threshold, the lane line curvature threshold, and the lane line distance threshold;

[0024] For any lane line in the driving data packet, when the lane line length corresponding to the lane line is greater than the lane line length threshold, and the lane line curvature is less than the lane line curvature threshold, and the lane line distance is less than the lane line distance threshold, determine that the lane line is an effective lane line.

[0025] In an embodiment of the present application, the step of determining at least one interaction relationship sequence between the effective driving trajectory of each to - be - labeled obstacle and at least one of the effective lane lines based on the driving intention determination sub - rule, and determining the driving intention corresponding to each to - be - labeled obstacle based on at least one of the interaction relationship sequences includes:

[0026] Based on the driving intention determination sub - rule, determine the interaction type definition rule and the driving intention mapping rule; the driving intention mapping rule represents different driving intentions corresponding to different combinations of interaction types;

[0027] Based on the interaction type definition rule, determine the interaction relationship sequence between the effective driving trajectory of each to - be - labeled obstacle and each of the effective lane lines in multiple consecutive image frames of the driving data packet; the interaction relationship sequence represents the combination of interaction types corresponding to multiple consecutive image frames;

[0028] Based on the driving intention mapping rule and the combination of interaction types, determine the driving intention corresponding to each to - be - labeled obstacle.

[0029] In an embodiment of the present application, the interaction types include driving on the left, driving on the line, and / or driving on the right;

[0030] The step of determining the driving intention corresponding to each to - be - labeled obstacle based on the driving intention mapping rule and the combination of interaction types includes:

[0031] For any effective lane line, when the effective lane line is the left lane line of the target vehicle and the combination of interaction types corresponding to the effective lane line sequentially includes driving on the left and driving on the right, determine that the driving intention of the to - be - labeled obstacle is to cut in from the left;; and / or,

[0032] For any effective lane line, when the effective lane line is not the left lane line of the target vehicle and the combination of interaction types corresponding to the effective lane line sequentially includes driving on the left and driving on the right, determine that the driving intention of the to - be - labeled obstacle is to change lanes to the left; and / or,

[0033] For any valid lane line, when the valid lane line is the right lane line of the target vehicle and the corresponding interaction type combination of the valid lane line sequentially includes the right-side driving and the left-side driving, determine that the driving intention of the obstacle to be labeled is a right cut-in;; and / or,

[0034] For any valid lane line, when the valid lane line is not the right lane line of the target vehicle and the corresponding interaction type combination of the valid lane line sequentially includes the right-side driving and the left-side driving, determine that the driving intention of the obstacle to be labeled is a right lane change; and / or,

[0035] For any valid lane line, when the corresponding interaction type combination of the valid lane line only includes the line-pressing driving, or when the interaction type combination only includes the line-pressing driving and the left-side driving, or when the interaction type combination only includes the line-pressing driving and the right-side driving, determine that the driving intention of the obstacle to be labeled is line-pressing; and / or,

[0036] For any valid lane line, when the corresponding interaction type combination of the valid lane line only includes the left-side driving or the right-side driving, determine that the driving intention of the obstacle to be labeled is cruise straight.

[0037] In an embodiment of the present application, before the step of classifying and labeling the first unqualified labeling data based on the updated labeling rule to obtain a second labeling result file when it is detected that the labeling rule is updated, the method further includes:

[0038] In response to an update instruction for the valid trajectory determination sub-rule, the valid lane line determination sub-rule, and / or the driving intention determination sub-rule, update the valid trajectory determination sub-rule, the valid lane line determination sub-rule, and / or the driving intention determination sub-rule.

[0039] In a second aspect, based on the same inventive concept, an embodiment of the present application provides a data labeling device, and the device includes;

[0040] A data acquisition module, configured to acquire a driving data packet collected by a target vehicle;;

[0041] A first labeling module, configured to classify and label the driving intention of at least one obstacle to be labeled in the driving data packet based on a preset labeling rule to obtain a first labeling result file;

[0042] The first storage module is used to obtain the first qualified annotation data and the first unqualified annotation data returned by the manual verification module for the first annotation result file, and store the first qualified annotation data into the qualified data set;

[0043] The second annotation module is used to update the driving intention corresponding to the first unqualified annotation data based on the updated annotation rules when it is detected that the annotation rules are updated, so as to obtain a second annotation result file;

[0044] The second storage module is used to obtain the second qualified annotation data returned by the manual verification module for the second annotation result file, and store the second qualified annotation data into the qualified data set.

[0045] In an embodiment of the present application, the annotation rules include an effective trajectory determination sub-rule, an effective lane line determination sub-rule, and a driving intention determination sub-rule; the first annotation module includes:

[0046] The trajectory determination sub-module is used to determine the effective driving trajectory of the obstacle to be annotated in the driving data packet based on the effective trajectory determination sub-rule;

[0047] The lane line determination sub-module is used to determine at least one effective lane line in the driving data packet based on the effective lane line determination sub-rule;

[0048] The driving intention determination sub-module is used to determine at least one interaction relationship sequence between the effective driving trajectory of each obstacle to be annotated and at least one of the effective lane lines based on the driving intention determination sub-rule, and determine the driving intention corresponding to each obstacle to be annotated based on at least one of the interaction relationship sequences;

[0049] The label generation sub-module is used to generate a driving intention label corresponding to each obstacle to be annotated, so as to obtain the first annotation result file.

[0050] In an embodiment of the present application, the trajectory determination sub-module includes:

[0051] The first threshold parameter determination unit is used to determine a trajectory point number threshold, a trajectory curvature threshold, and a continuity parameter threshold based on the effective trajectory determination sub-rule;

[0052] The trajectory parameter determination unit is used to determine, for any obstacle to be annotated, a plurality of trajectory points of the obstacle to be annotated in a plurality of consecutive image frames of the driving data packet, and a trajectory curve and a continuity parameter corresponding to the plurality of trajectory points; wherein, the continuity parameter characterizes the continuity degree between the plurality of trajectory points;

[0053] An effective driving trajectory determination unit is configured to determine the trajectory curve of the obstacle to be labeled as an effective driving trajectory when the number of trajectory points is greater than the trajectory point number threshold, the curvature of the trajectory curve is less than the trajectory curvature threshold, and the continuity parameter is greater than the continuity parameter threshold.

[0054] In an embodiment of the present application, the lane line determination sub-module includes:

[0055] A second threshold parameter determination unit is configured to determine a lane line length threshold, a lane line curvature threshold, and a lane line distance threshold based on the effective lane line determination sub-rule.

[0056] An effective lane line determination unit is configured to determine that a lane line in the driving data packet is an effective lane line when the lane line length corresponding to the lane line is greater than the lane line length threshold, the lane line curvature is less than the lane line curvature threshold, and the lane line distance is less than the lane line distance threshold.

[0057] In an embodiment of the present application, the driving intention determination sub-module includes:

[0058] A rule determination unit is configured to determine an interaction type definition rule and a driving intention mapping rule based on the driving intention determination sub-rule; the driving intention mapping rule represents different driving intentions corresponding to different combinations of interaction types.

[0059] An interaction relationship sequence determination unit is configured to determine an interaction relationship sequence between the effective driving trajectory of each obstacle to be labeled and each of the effective lane lines in multiple consecutive image frames of the driving data packet based on the interaction type definition rule; the interaction relationship sequence represents a combination of interaction types corresponding to multiple consecutive image frames.

[0060] A driving intention determination unit is configured to determine the driving intention corresponding to each obstacle to be labeled based on the driving intention mapping rule and the combination of interaction types.

[0061] In an embodiment of the present application, the interaction types include driving on the left, driving on the line, and / or driving on the right; the driving intention determination unit includes:

[0062] A first driving intention determination sub-unit is configured to determine that the driving intention of the obstacle to be labeled is a left cut-in when the effective lane line is the left lane line of the target vehicle and the combination of interaction types corresponding to the effective lane line sequentially includes driving on the left and driving on the right for any effective lane line.

[0063] A second driving intention determination subunit, configured to determine, for any valid lane line, that the driving intention of the obstacle to be labeled is a left lane change when the valid lane line is not the left lane line of the target vehicle and the interaction type combination corresponding to the valid lane line sequentially includes the left driving and the right driving;

[0064] A third driving intention determination subunit, configured to determine, for any valid lane line, that the driving intention of the obstacle to be labeled is a right cut-in when the valid lane line is the right lane line of the target vehicle and the interaction type combination corresponding to the valid lane line sequentially includes the right driving and the left driving

[0065] A fourth driving intention determination subunit, configured to determine, for any valid lane line, that the driving intention of the obstacle to be labeled is a right lane change when the valid lane line is not the right lane line of the target vehicle and the interaction type combination corresponding to the valid lane line sequentially includes the right driving and the left driving;

[0066] A fifth driving intention determination subunit, configured to determine, for any valid lane line, that the driving intention of the obstacle to be labeled is line pressing when the interaction type combination corresponding to the valid lane line only includes line pressing driving, or when the interaction type combination only includes line pressing driving and left driving, or when the interaction type combination only includes line pressing driving and right driving;

[0067] A sixth driving intention determination subunit, configured to determine, for any valid lane line, that the driving intention of the obstacle to be labeled is cruise straight when the interaction type combination corresponding to the valid lane line only includes left driving or right driving.

[0068] In an embodiment of the present application, the data annotation device further includes:

[0069] An annotation rule update module, configured to update the valid trajectory determination sub-rule, the valid lane line determination sub-rule, and / or the driving intention determination sub-rule in response to an update instruction for the valid trajectory determination sub-rule, the valid lane line determination sub-rule, and / or the driving intention determination sub-rule.

[0070] In a third aspect, based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the data annotation method proposed in the first aspect of the present application.

[0071] Fourthly, based on the same inventive concept, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the data annotation method proposed in the first aspect of the present application is implemented.

[0072] Compared with the prior art, the present application has the following advantages:

[0073] A data annotation method provided by an embodiment of the present application can obtain a driving data packet collected by a target vehicle, and classify and annotate the driving intention of at least one obstacle to be annotated in the driving data packet based on a preset annotation rule, so as to obtain a first annotation result file. Then, by obtaining the first qualified annotation data and the first unqualified annotation data returned by the manual verification module for the first annotation result file, not only can the first qualified annotation data be stored in the qualified data set, but also when it is detected that the annotation rule is updated, based on the updated annotation rule, the driving intention corresponding to the first unqualified annotation data can be updated to obtain a second annotation result file. Then, the second qualified annotation data returned by the manual verification module for the second annotation result file is obtained, and the second qualified annotation data is stored in the qualified data set. By means of the preset annotation rule, the embodiment of the present application can efficiently annotate the driving intention of the obstacle to be annotated, effectively improve the efficiency and accuracy of data annotation, and ensure the consistency of the annotation result. At the same time, by designing a complete data verification and recycling mechanism, not only can the version iteration efficiency and convenience of the annotation rule be improved, but also the incremental accumulation of qualified data can be realized, and the data utilization rate can be improved. Description of the Drawings

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0075] Figure 1 It is a flowchart of the steps of a data annotation method in an embodiment of the present application.

[0076] Figure 2 It is a schematic diagram of the functional modules of a data annotation device in an embodiment of the present application.

[0077] Figure 3 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application. Detailed Embodiments

[0078] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0079] It should be noted that different from the fact that the perception data can identify obstacles through a single-frame image, the identification of the driving intention of a vehicle requires multiple-frame images, resulting in a higher labeling difficulty for the driving intention than the perception data. Therefore, the traditional labeling method relies on manual labeling. However, the labeling result is easily affected by human subjectivity, and there are still problems of poor labeling efficiency, accuracy, and consistency. For example, different personnel have different labeling experiences and cognitions, so that different personnel are likely to label different results even when labeling the driving data in the same scene within the same time period. At the same time, due to the difficulty in guaranteeing the accuracy of manual labeling, a large amount of qualified driving data that is mislabeled or considered unqualified by humans is discarded, making this part of the qualified driving data unable to be effectively utilized, resulting in low data utilization.

[0080] Aiming at the problems of low efficiency, low accuracy, poor consistency, and low data utilization rate existing in the prior art when performing driving intention labeling, the present application aims to provide a data labeling method supporting a data recycling mechanism. Through preset labeling rules, it can achieve efficient labeling of the driving intention of the obstacle to be labeled, effectively improve the efficiency and accuracy of data labeling, and ensure the consistency of the labeling result; at the same time, by designing a complete set of data verification and recycling mechanisms, it can not only improve the version iteration efficiency and convenience of the labeling rules, but also achieve incremental accumulation of qualified data and improve data utilization rate.

[0081] Referring to Figure 1 , a data labeling method of the present application is shown. The method may include the following steps:

[0082] S101: Obtain a driving data packet collected by a target vehicle.

[0083] It should be noted that all or part of the steps of this embodiment may be executed in the local electronic device of the target vehicle according to actual needs, or may also be executed in a server or other remote devices having communication capabilities and / or data processing capabilities with the target vehicle. Among them, the server may be a central server, a cluster server, or a distributed server, or may also be a cloud server for implementing cloud computing and / or cloud storage, etc. It should be noted that this embodiment does not make specific restrictions on the execution subject.

[0084] Specifically, when the execution entity is a server, the server communicates with at least one target vehicle and obtains the driving data packets transmitted back by each target vehicle to implement the driving intention annotation of the obstacles to be annotated in the driving data packets on the server side; when the execution entity is a local electronic device, the local electronic device obtains the driving data packets collected by the target vehicle, implements the driving intention annotation of the obstacles to be annotated in the driving data packets locally, and uploads the qualified annotation data stored in the qualified data set to the server.

[0085] In a specific implementation, the server can obtain the driving data packets collected by the target vehicle based on the actual road conditions. The target vehicle can collect the driving data packets at a fixed sampling period, or automatically upload the collected driving data packets to the server after completing a round of driving. Among them, the target vehicle can be a dedicated data collection vehicle or an authorized mass-produced vehicle.

[0086] It should be further noted that the driving data packet is composed of multiple consecutive image frames collected by the target vehicle according to time sequence. There is a fixed sampling interval between every two adjacent image frames, and this sampling interval can be set according to actual needs.

[0087] S102: Classify and annotate the driving intention of at least one obstacle to be annotated in the driving data packet based on a preset annotation rule to obtain a first annotation result file.

[0088] In this embodiment, to reduce the computational load brought by vehicles that are far away and have no annotation significance in each image frame of the driving data packet, a preset position range is established around the vehicle itself (i.e., the target vehicle). During the driving process of the vehicle itself, the preset position range and the vehicle itself are in a relatively static relationship. Therefore, the preset position range can be called a static area. Specifically, the obstacles outside the static area are far away from the vehicle itself, and such obstacles can be determined as invalid obstacles that do not need to be annotated; while the obstacles within the static area, because they are relatively close to the vehicle itself, such obstacles are determined as obstacles to be annotated that need to have their driving intentions annotated.

[0089] It should be noted that the obstacles to be evaluated can specifically include one or more of pedestrians, non-motor vehicles, and motor vehicles. Among them, non-motor vehicles can specifically include non-motor vehicles such as bicycles traveling on non-motor vehicle lanes, and motor vehicles can specifically include motor vehicles such as motorcycles, cars, minibuses, trucks, and fire trucks traveling on motor vehicle lanes.

[0090] In this embodiment, based on the preset annotation rule, the driving trajectory and lane lines of the obstacles to be annotated can be automatically recognized. Furthermore, based on the interaction relationship between the driving trajectory and lane lines of the obstacles to be annotated, the driving intention of the obstacles to be annotated can be automatically annotated, and then a first annotation result file can be obtained.

[0091] In this embodiment, after obtaining the first annotation result file, the first annotation result file will be sent to the manual verification module for quality inspection to determine whether the driving intention of the first annotation result file is correctly annotated or whether the image quality is qualified.

[0092] S103: Obtain the first qualified annotation data and the first unqualified annotation data returned by the manual verification module for the first annotation result file, and store the first qualified annotation data in the qualified data set.

[0093] It should be noted that after obtaining the first annotation result file, the manual verification module can display the first annotation result file through an interactive interface for the verifier to conduct quality inspection. Since the first annotation result file is data with the driving intention already annotated, the verifier does not need to perform annotation work anymore and only needs to conduct quality inspection on the first annotation result file.

[0094] In a specific implementation, the verifier can divide the first annotation result file into the first qualified annotation data, the first unqualified annotation data, and invalid data based on the screening component provided by the manual verification module. After completing the division of the first annotation result file, the first qualified annotation data and the first unqualified annotation data are returned to the server for storage, while the invalid data is not returned to the server and is deleted to avoid repeated identification of invalid data.

[0095] It should be noted that the first qualified annotation data represents data that is considered to have the driving intention correctly annotated after quality inspection; the first unqualified annotation data represents data that is considered to have the driving intention wrongly annotated after quality inspection; and the invalid data represents data with unqualified image quality that cannot meet the training requirements of the prediction model.

[0096] In this embodiment, after the server obtains the first qualified annotation data and the first unqualified annotation data, the first qualified annotation data will be stored in the qualified data set, and the unqualified annotation data will not be deleted but will be stored in the unqualified data set, so that after the annotation rules are updated, the unqualified annotation data can be recycled based on the updated annotation rules.

[0097] S104: When it is detected that the annotation rules are updated, update the driving intention corresponding to the first unqualified annotation data based on the updated annotation rules to obtain a second annotation result file.

[0098] It should be noted that the annotation rules will be updated or modified according to requirements, and different annotation rules may have differences in the recognition methods of the driving trajectories and / or lane lines of the obstacles to be annotated, which may lead to changes in the driving intention.

[0099] In this embodiment, after detecting that the annotation rule is updated, the driving intention corresponding to the first unqualified annotation data stored in the unqualified data set is updated by using the updated annotation rule, so as to realize the update of the driving intention of the to-be-annotated obstacle in the first unqualified annotation data and realize the recycling of the first unqualified annotation data.

[0100] In a specific implementation, to update the driving intention, the initially annotated driving intention of the first unqualified annotation data can be deleted before the first unqualified annotation data is stored in the unqualified data set, and then, based on the updated annotation rule, a driving intention label representing the latest driving intention is directly generated; or, after obtaining the latest driving intention of the first unqualified annotation data based on the updated annotation rule, the driving intention label is updated from the initial driving intention to the latest driving intention.

[0101] Exemplarily, the driving intention label can be represented by a number, and different numbers represent different driving intentions. For example, "0" represents "cruise straight", and "1" represents "left lane change". If the initially recognized driving intention of a to-be-annotated obstacle by the annotation rule before update is cruise straight, the corresponding driving intention label is "0", and when the latest driving intention of the to-be-annotated obstacle recognized by the updated annotation rule is left lane change, the driving intention label can be updated from "0" to "1".

[0102] S105: Obtain the second qualified annotation data returned by the manual verification module for the second annotation result file, and store the second qualified annotation data in the qualified data set.

[0103] In this embodiment, after obtaining the second annotation result file, the second annotation result file will also be sent to the manual verification module for quality inspection, and the second qualified annotation data and the second unqualified annotation data returned by the manual verification module are obtained.

[0104] In this embodiment, for the second qualified annotation data, it will be stored in the qualified data set, and then, on the basis of the first qualified annotation data, the second qualified annotation data is added, thereby realizing the incremental accumulation of qualified data.

[0105] In this embodiment, based on the type of driving intention, the qualified data set can be divided into multiple qualified data subsets, and different qualified data subsets are used to store qualified annotation data with different types of driving intentions. It should be noted that the qualified annotation data in the qualified data set is used to train a prediction model, and this prediction model is used to predict the driving intention of the obstacles around the vehicle.

[0106] In this embodiment, for the second unqualified labeled data, it will be retained in the unqualified dataset until the labeling rules are iteratively updated again. Then, the unqualified data in the unqualified dataset will be relabeled and verified to achieve the recycling of the unqualified labeled data.

[0107] In one example, the initial version of the labeling rules is V1, and the driving intention of the driving data packet is labeled only using the V1 version of the labeling rules to obtain the first labeled result file. After the quality inspection by the manual verification module, the invalid data is removed, and the first qualified labeled data and the first unqualified labeled data are obtained. Subsequently, to improve the training effect of the prediction model, the technical personnel adjust and optimize the parameters of some of the labeling rules to obtain the V2 version of the labeling rules. Since the first qualified labeled data is valid data that has passed the quality inspection, there is no need to relabel it using the V2 version of the labeling rules. Instead, the V2 version of the labeling rules is used to relabel the first unqualified labeled data to obtain the second qualified labeled data, enabling the effective recycling of the driving data that was mislabeled by the V1 version of the labeling rules. With the iterative update of the labeling rules, the valid data in the unqualified dataset will be continuously transferred to the qualified dataset, thereby achieving the incremental accumulation of the qualified labeled data in the qualified dataset and generally improving the data utilization rate of the driving data packet.

[0108] In this embodiment, on the one hand, compared with manual labeling, by constructing an efficient automated rule processing mechanism, it is possible to efficiently label the driving intention of the obstacles to be labeled based on the preset labeling rules, effectively improving the efficiency and accuracy of data labeling and ensuring the consistency of the labeling results. At the same time, considering the iterative update requirements of the labeling rules, by designing a complete data verification and recycling mechanism, not only can the version iteration efficiency and convenience of the labeling rules be improved, but also the incremental accumulation of qualified data can be achieved, the data utilization rate can be increased, and the data acquisition cost can be reduced.

[0109] In a feasible embodiment, the labeling rules may specifically include an effective trajectory determination sub-rule, an effective lane line determination sub-rule, and a driving intention determination sub-rule; S102 may specifically include the following sub-steps:

[0110] S102-1: Based on the effective trajectory determination sub-rule, determine the effective driving trajectory of the obstacle to be labeled in the driving data packet.

[0111] In this embodiment, to facilitate targeted parameter modification, update, and optimization of the annotation rules and further improve the version iteration efficiency of the annotation rules, the annotation rules are split into an effective trajectory determination sub-rule, an effective lane line determination sub-rule, and a driving intention determination sub-rule. Technicians can adjust the relevant parameters in the sub-rules through a preset interface module. Among them, the interface module includes multiple interfaces, and different interfaces correspond to different sub-rules. For example, the first interface can be set to implement parameter update for the effective trajectory determination sub-rule, the second interface can be set to implement parameter update for the effective lane line determination sub-rule, and the third interface can be set to implement parameter update for the driving intention determination sub-rule.

[0112] In this embodiment, the effective trajectory determination sub-rule is used to screen out the effective driving trajectories that meet the driving trajectory requirements from the driving trajectories of the obstacles to be annotated in the driving data packet. It should be noted that if it is detected that the driving trajectory of the obstacle to be annotated does not meet the driving trajectory requirements, it is determined that the driving trajectory is an invalid driving trajectory, and the label of the corresponding obstacle to be annotated is determined as unavailable for driving intention. Furthermore, the obstacle to be annotated will not participate in the training of the prediction model.

[0113] S102-2: Based on the effective lane line determination sub-rule, determine at least one effective lane line in the driving data packet.

[0114] In this embodiment, the effective lane line determination sub-rule is used to screen out the effective lane lines that meet the lane line requirements from the lane lines in the driving data packet. It should be noted that if it is detected that a certain lane line does not meet the lane line requirements, it is determined that the lane line is an invalid lane line. That is to say, the interaction relationship between the invalid lane line and the obstacle to be annotated will not be detected to improve the annotation accuracy.

[0115] S102-3: Based on the driving intention determination sub-rule, determine at least one interaction relationship sequence between the effective driving trajectory of each obstacle to be annotated and at least one effective lane line, and based on at least one interaction relationship sequence, determine the respective driving intention corresponding to each obstacle to be annotated.

[0116] In this embodiment, after determining the effective lane lines and the effective driving trajectories of each obstacle to be annotated, time matching will be performed on the effective driving trajectories and the effective lane lines to obtain the interaction relationship between the effective lane lines and the effective driving trajectories corresponding to each image frame, and then obtain the interaction relationship sequence corresponding to multiple consecutive image frames. The number of multiple consecutive image frames can be set according to actual needs. For example, it can be set to 15 consecutive image frames.

[0117] Exemplarily, assume that N obstacles to be labeled respectively correspond to N effective driving trajectories, and M effective lane lines are detected. Then, there will be N*M interaction relationship sequences between the N obstacles to be labeled and the M effective lane lines. For each obstacle to be labeled, there are M interaction relationship sequences between the obstacle to be labeled and the M effective lane lines. Herein, both N and M are positive integers greater than or equal to 1.

[0118] In this embodiment, for any obstacle to be labeled, one or more interaction relationship sequences corresponding to the obstacle to be labeled can be used to determine the driving intention corresponding to the obstacle to be labeled.

[0119] S102-4: Generate a driving intention label corresponding to each obstacle to be labeled, and obtain a first annotation result file.

[0120] In this embodiment, after determining the driving intention corresponding to each obstacle to be labeled, a driving intention label corresponding to each obstacle to be labeled will be generated.

[0121] In a specific implementation, for the convenience of data management and subsequent quality inspection, the first annotation result file will be classified and managed based on the type of driving intention. Exemplarily, assume that there are four types of driving intentions including intention A, intention B, intention C, and intention D. Then, the first annotation result file can be divided into a first annotation result sub-file for labeling intention A, a second annotation result sub-file for labeling intention B, a third annotation result sub-file for labeling intention C, and a fourth annotation result sub-file for labeling intention D.

[0122] In a feasible embodiment, S102-1 may specifically include the following sub-steps:

[0123] S102-1-1: Based on the effective trajectory determination sub-rule, determine the trajectory point quantity threshold, the trajectory curvature threshold, and the continuity parameter threshold.

[0124] In this embodiment, a technician can set parameters such as the trajectory point quantity threshold, the trajectory curvature threshold, and the continuity parameter threshold through the first interface of the interface module to implement the setting of the effective trajectory determination sub-rule.

[0125] S102-1-2: For any obstacle to be labeled, determine multiple trajectory points of the obstacle to be labeled within multiple consecutive image frames of the driving data packet, as well as the trajectory curve and continuity parameter corresponding to the multiple trajectory points.

[0126] It should be noted that the continuity parameter characterizes the continuity degree between multiple trajectory points. Specifically, the continuity parameter can be calculated based on the relative distance between every two adjacent trajectory points. The larger the continuity parameter, the higher the continuity degree between multiple trajectory points.

[0127] In this embodiment, based on multiple trajectory points, a complete trajectory curve can be fitted, and then the curvature of the trajectory curve can be determined. In the case where there are multiple curves in the trajectory curve, the maximum curvature among the multiple curves is taken as the curvature of the trajectory curve.

[0128] S102-1-3: When the number of trajectory points is greater than the trajectory point number threshold, the curvature of the trajectory curve is less than the trajectory curvature threshold, and the continuity parameter is greater than the continuity parameter threshold, determine that the trajectory curve of the obstacle to be labeled is a valid driving trajectory.

[0129] In this embodiment, by setting the trajectory point number threshold, abnormal trajectories with too short trajectories can be effectively filtered out; by setting the trajectory curvature threshold, abnormal trajectories with too large trajectory curvatures can be effectively filtered out; by setting the continuity parameter threshold, abnormal trajectories with unqualified continuity can be effectively filtered out.

[0130] In this embodiment, by comprehensively considering the number of multiple trajectory points, the continuity parameter, and the curvature of the trajectory curve corresponding to the multiple trajectory points, abnormal trajectories can be effectively filtered out, and then valid driving trajectories can be retained, improving the accuracy of driving intention labeling.

[0131] In a feasible embodiment, S102-2 may specifically include the following sub-steps:

[0132] S102-2-1: Based on the sub-rules for determining valid lane lines, determine the lane line length threshold, the lane line curvature threshold, and the lane line distance threshold.

[0133] In this embodiment, a technician can set parameters such as the lane line length threshold, the lane line curvature threshold, and the lane line distance threshold through the second interface of the interface module to implement the setting of the sub-rules for determining valid lane lines.

[0134] S102-2-2: For any lane line in the driving data packet, when the lane line length corresponding to the lane line is greater than the lane line length threshold, the lane line curvature is less than the lane line curvature threshold, and the lane line distance is less than the lane line distance threshold, determine that the lane line is a valid lane line.

[0135] In this embodiment, by setting the lane line length threshold, abnormal lane lines with too short lane line lengths can be effectively filtered out; by setting the lane line curvature threshold, abnormal lane lines with too large lane line curvatures can be effectively filtered out; by setting the lane line distance threshold, abnormal lane lines with too far lane line distances can be effectively filtered out.

[0136] In this embodiment, by comprehensively considering the length, curvature, and distance of lane lines, abnormal lane lines can be effectively filtered out, thereby retaining valid lane lines and improving the accuracy of driving intention annotation.

[0137] In a feasible embodiment, S102-3 may specifically include the following sub-steps:

[0138] S102-3-1: Based on the sub-rules for driving intention determination, determine the interaction type definition rule and the driving intention mapping rule.

[0139] It should be noted that the interaction type represents the interaction type between the obstacle to be annotated and the valid lane line. For example, based on the positional relationship between the obstacle to be annotated and the valid lane line, the interaction type may include driving on the left, driving on the line, and / or driving on the right. The driving intention mapping rule characterizes different driving intentions corresponding to different combinations of interaction types. That is to say, different driving intentions correspond to specific combinations of interaction types.

[0140] In this embodiment, technicians can customize the interaction type definition rule and the driving intention mapping rule through the third interface of the interface module. For example, operations such as adding, deleting, and / or modifying the interaction type, the combination of interaction types, and the driving intention can be performed according to actual needs.

[0141] S102-3-2: Based on the interaction type definition rule, determine the sequence of interaction relationships between the effective driving trajectory of each obstacle to be annotated and each valid lane line in multiple consecutive image frames of the driving packet.

[0142] It should be noted that the sequence of interaction relationships characterizes the combination of interaction types corresponding to multiple consecutive image frames. Among them, for the convenience of recognition, multiple identical and consecutive interaction types can be combined into one interaction type.

[0143] Exemplarily, define driving on the left as L, driving on the line as S, and driving on the right as R. If the sequence of interaction relationships between a certain obstacle to be annotated and a certain valid lane line in 10 consecutive image frames is LLLLSSRRR in turn, the combination of interaction types corresponding to this sequence of interaction relationships can be expressed as LSR.

[0144] S102-3-3: Based on the driving intention mapping rule and the combination of interaction types, determine the driving intention corresponding to each obstacle to be annotated.

[0145] In this embodiment, after obtaining the combination of interaction types between the obstacle to be annotated and the valid lane line, the driving intention corresponding to each obstacle to be annotated can be determined based on the driving intention mapping rule.

[0146] In this embodiment, when only considering the positional relationship between the obstacle to be labeled and the valid lane line, the driving intention can be specifically divided into four different types of driving intentions: left lane change, right lane change, cruise straight, and lane pressing; while when comprehensively considering the positional relationships among the obstacle to be labeled, the target vehicle, and the valid lane line, the driving intention can be further divided into six different types of driving intentions: left lane change, left cut-in, right lane change, right cut-in, cruise straight, and lane pressing.

[0147] Specifically, define driving on the left as L, driving on the lane line as S, and driving on the right as R. The above six different types of driving intentions respectively correspond to the following six situations:

[0148] Situation 1: For any valid lane line, when the valid lane line is the left lane line of the target vehicle and the corresponding interaction type combination of the valid lane line sequentially includes driving on the left and driving on the right, determine the driving intention of the obstacle to be labeled as left cut-in.

[0149] It should be noted that the situation where the interaction type combination sequentially includes L and R specifically includes two situations: LR and LSR. That is to say, when it is detected that the valid lane line is the left lane line of the target vehicle and the interaction type combination is LR or LSR, determine the driving intention of the obstacle to be labeled as left cut-in, that is, the obstacle to be labeled cuts into the ego lane where the target vehicle is located from the left lane of the target vehicle.

[0150] Situation 2: For any valid lane line, when the valid lane line is not the left lane line of the target vehicle and the corresponding interaction type combination of the valid lane line sequentially includes driving on the left and driving on the right, determine the driving intention of the obstacle to be labeled as left lane change.

[0151] It should be noted that the situation where the interaction type combination sequentially includes L and R specifically includes two situations: LR and LSR. That is to say, when it is detected that the valid lane line is not the left lane line of the target vehicle and the interaction type combination is LR or LSR, determine the driving intention of the obstacle to be labeled as left lane change, that is, the obstacle to be labeled cuts into the right lane of the valid lane line from the left lane of the valid lane line.

[0152] Situation 3: For any valid lane line, when the valid lane line is the right lane line of the target vehicle and the corresponding interaction type combination of the valid lane line sequentially includes driving on the right and driving on the left, determine the driving intention of the obstacle to be labeled as right cut-in.

[0153] It should be noted that the case where the interaction type combination includes R and L in sequence specifically includes two cases: RL and RSL. That is to say, when the detected valid lane line is the right lane line of the target vehicle and the interaction type combination is RL or RSL, it is determined that the driving intention of the obstacle to be labeled is to change lanes to the right, that is, the obstacle to be labeled cuts into the ego lane where the target vehicle is located from the right lane of the target vehicle.

[0154] Case 4: For any valid lane line, when the valid lane line is not the right lane line of the target vehicle and the interaction type combination corresponding to the valid lane line includes right driving and left driving in sequence, it is determined that the driving intention of the obstacle to be labeled is to change lanes to the right.

[0155] It should be noted that the case where the interaction type combination includes L and R in sequence specifically includes two cases: RL and RSL. That is to say, when the detected valid lane line is not the right lane line of the target vehicle and the interaction type combination is RL or RSL, it is determined that the driving intention of the obstacle to be labeled is to change lanes to the right, that is, the obstacle to be labeled cuts into the left lane of the valid lane line from the right lane of the valid lane line.

[0156] Case 5: For any valid lane line, when the interaction type combination corresponding to the valid lane line only includes driving on the line, or when the interaction type combination only includes driving on the line and left driving, or when the interaction type combination only includes driving on the line and right driving, it is determined that the driving intention of the obstacle to be labeled is to drive on the line.

[0157] It should be noted that the case where the interaction type combination only includes driving on the line and left driving specifically includes two cases: LS and SL; the case where the interaction type combination only includes driving on the line and right driving specifically includes two cases: RS and SR. That is to say, when the detected interaction type combination is any one of the combinations of RS, SR, LS, SL, and S, it is determined that the driving intention of the obstacle to be labeled is to drive on the line, that is, the obstacle to be labeled is in the state of driving on the line.

[0158] Case 6: For any valid lane line, when the interaction type combination corresponding to the valid lane line only includes left driving or right driving, it is determined that the driving intention of the obstacle to be labeled is to cruise straight.

[0159] In this embodiment, if it is detected that the interaction type combination is S or L, then it is determined that the driving intention of the obstacle to be labeled is to cruise straight.

[0160] It should be noted that if the driving intention of the obstacle to be labeled does not belong to the above six cases, then the driving intention of the obstacle to be labeled can be labeled as other.

[0161] In this embodiment, by comprehensively considering the positional relationship between the obstacle to be labeled, the target vehicle, and the effective lane lines, the driving intention of the obstacle to be labeled can be accurately labeled, thereby effectively improving the training effect and practical application effect of the prediction model.

[0162] In a feasible embodiment, between S104, the data annotation method may further include the following steps:

[0163] S201: In response to an update instruction for the effective trajectory determination sub-rule, the effective lane line determination sub-rule, and / or the driving intention determination sub-rule, update the effective trajectory determination sub-rule, the effective lane line determination sub-rule, and / or the driving intention determination sub-rule.

[0164] In this embodiment, when technicians update the annotation rules, they can specifically update the effective trajectory determination sub-rule, the effective lane line determination sub-rule, and / or the driving intention determination sub-rule through the interface module, thereby effectively improving the data version iteration efficiency.

[0165] In specific implementation, for the effective trajectory determination sub-rule, parameters such as the trajectory point number threshold, the trajectory curvature threshold, and the continuity parameter threshold can be updated through the first interface of the interface module; for the effective lane line determination sub-rule, parameters such as the lane line length threshold, the lane line curvature threshold, and the lane line distance threshold can be updated through the second interface of the interface module; for the driving intention determination sub-rule, the interaction type definition rule and the driving intention mapping rule can be updated through the third interface of the interface module to update the interaction type and the mapping relationship between the interaction type combination and the driving intention.

[0166] In this embodiment, by dividing the entire annotation process into multiple process nodes, and different sub-rules are correspondingly set for different process nodes, technicians can, for any process node in the annotation process, call the corresponding interface of the interface module to achieve targeted modification and optimization of the sub-rule corresponding to this process node, thereby improving the version iteration efficiency of the annotation rules and reducing the update cost.

[0167] Second aspect, based on the same inventive concept, referring to Figure 2 , the embodiment of the present application provides a data annotation device 200, and the data annotation device 200 includes:

[0168] A data acquisition module 201, configured to acquire a driving data packet collected by a target vehicle;

[0169] A first annotation module 202, configured to classify and annotate the driving intention of at least one obstacle to be labeled in the driving data packet based on a preset annotation rule, and obtain a first annotation result file;

[0170] The first storage module 203 is configured to obtain the first qualified annotation data and the first unqualified annotation data returned by the manual verification module for the first annotation result file, and store the first qualified annotation data into the qualified data set;

[0171] The second annotation module 204 is configured to, when detecting that the annotation rule is updated, update the driving intention corresponding to the first unqualified annotation data based on the updated annotation rule to obtain a second annotation result file;

[0172] The second storage module 205 is configured to obtain the second qualified annotation data returned by the manual verification module for the second annotation result file, and store the second qualified annotation data into the qualified data set.

[0173] In an embodiment of the present application, the annotation rule includes a valid trajectory determination sub-rule, a valid lane line determination sub-rule, and a driving intention determination sub-rule; the first annotation module 202 includes:

[0174] A trajectory determination sub-module, configured to determine a valid driving trajectory of an obstacle to be annotated in the driving data packet based on the valid trajectory determination sub-rule;

[0175] A lane line determination sub-module, configured to determine at least one valid lane line in the driving data packet based on the valid lane line determination sub-rule;

[0176] A driving intention determination sub-module, configured to determine at least one interaction relationship sequence between the valid driving trajectory of each obstacle to be annotated and at least one valid lane line based on the driving intention determination sub-rule, and determine the respective driving intention of each obstacle to be annotated based on at least one interaction relationship sequence;

[0177] A label generation sub-module, configured to generate a driving intention label corresponding to each obstacle to be annotated to obtain a first annotation result file.

[0178] In an embodiment of the present application, the trajectory determination sub-module includes:

[0179] A first threshold parameter determination unit, configured to determine a trajectory point number threshold, a trajectory curvature threshold, and a continuity parameter threshold based on the valid trajectory determination sub-rule;

[0180] A trajectory parameter determination unit, configured to, for any obstacle to be annotated, determine a plurality of trajectory points of the obstacle to be annotated in a plurality of consecutive image frames of the driving data packet, and a trajectory curve and a continuity parameter corresponding to the plurality of trajectory points; wherein, the continuity parameter characterizes the continuity degree between the plurality of trajectory points;

[0181] An effective driving trajectory determination unit is configured to determine the trajectory curve of an obstacle to be labeled as an effective driving trajectory when the number of trajectory points is greater than the trajectory point number threshold, the curvature of the trajectory curve is less than the trajectory curvature threshold, and the continuity parameter is greater than the continuity parameter threshold.

[0182] In an embodiment of the present application, the lane line determination sub-module includes:

[0183] A second threshold parameter determination unit is configured to determine a lane line length threshold, a lane line curvature threshold, and a lane line distance threshold based on an effective lane line determination sub-rule.

[0184] An effective lane line determination unit is configured to determine a lane line as an effective lane line for any lane line in the driving data packet when the lane line length corresponding to the lane line is greater than the lane line length threshold, the lane line curvature is less than the lane line curvature threshold, and the lane line distance is less than the lane line distance threshold.

[0185] In an embodiment of the present application, the driving intention determination sub-module includes:

[0186] A rule determination unit is configured to determine an interaction type definition rule and a driving intention mapping rule based on a driving intention determination sub-rule; the driving intention mapping rule represents different driving intentions corresponding to different combinations of interaction types.

[0187] An interaction relationship sequence determination unit is configured to determine an interaction relationship sequence between the effective driving trajectory of each obstacle to be labeled and each effective lane line in multiple consecutive image frames of the driving data packet based on the interaction type definition rule; the interaction relationship sequence represents a combination of interaction types corresponding to multiple consecutive image frames.

[0188] A driving intention determination unit is configured to determine the driving intention corresponding to each obstacle to be labeled based on the driving intention mapping rule and the combination of interaction types.

[0189] In an embodiment of the present application, the interaction types include driving on the left, driving on the line, and / or driving on the right; the driving intention determination unit includes:

[0190] A first driving intention determination sub-unit is configured to determine the driving intention of the obstacle to be labeled as a left cut-in when, for any effective lane line, the effective lane line is the left lane line of the target vehicle and the combination of interaction types corresponding to the effective lane line sequentially includes driving on the left and driving on the right.

[0191] A second driving intention determination sub-unit is configured to determine the driving intention of the obstacle to be labeled as a left lane change when, for any effective lane line, the effective lane line is not the left lane line of the target vehicle and the combination of interaction types corresponding to the effective lane line sequentially includes driving on the left and driving on the right.

[0192] The third driving intention determination subunit is configured to, for any valid lane line, when the valid lane line is the right lane line of the target vehicle and the interaction type combination corresponding to the valid lane line sequentially includes right driving and left driving, determine that the driving intention of the obstacle to be marked is right cut-in.

[0193] The fourth driving intention determination subunit is configured to, for any valid lane line, when the valid lane line is not the right lane line of the target vehicle and the interaction type combination corresponding to the valid lane line sequentially includes right driving and left driving, determine that the driving intention of the obstacle to be marked is right lane change;

[0194] The fifth driving intention determination subunit is configured to, for any valid lane line, when the interaction type combination corresponding to the valid lane line only includes pressing the line to drive, or when the interaction type combination only includes pressing the line to drive and left driving, or when the interaction type combination only includes pressing the line to drive and right driving, determine that the driving intention of the obstacle to be marked is pressing the line;

[0195] The sixth driving intention determination subunit is configured to, for any valid lane line, when the interaction type combination corresponding to the valid lane line only includes left driving or right driving, determine that the driving intention of the obstacle to be marked is cruising straight.

[0196] In an embodiment of the present application, the data annotation device 200 further includes:

[0197] The annotation rule update module is configured to, in response to an update instruction for the valid trajectory determination sub-rule, the valid lane line determination sub-rule, and / or the driving intention determination sub-rule, update the valid trajectory determination sub-rule, the valid lane line determination sub-rule, and / or the driving intention determination sub-rule.

[0198] It should be noted that the specific implementation manner of the data annotation device 200 in the embodiment of the present application refers to the specific implementation manner of the data annotation method proposed in the first aspect of the embodiment of the present application, which will not be elaborated here.

[0199] In a third aspect, based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the data annotation method proposed in the first aspect of the present application.

[0200] It should be noted that the specific implementation manner of the computer-readable storage medium in the embodiment of the present application refers to the specific implementation manner of the data annotation method proposed in the first aspect of the embodiment of the present application, which will not be elaborated here.

[0201] Fourthly, based on the same inventive concept, an embodiment of the present application provides an electronic device 300, including a memory 301, a processor 302, and a computer program stored in the memory 301 and executable on the processor 302. When the processor 302 executes the computer program, the data annotation method proposed in the first aspect of the present application is implemented.

[0202] It should be noted that for the specific implementation manner of the electronic device 300 in the embodiment of the present application, refer to the specific implementation manner of the data annotation method proposed in the first aspect of the embodiment of the present application above, which will not be elaborated here.

[0203] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a device, or a computer program product. Therefore, the embodiments of the present invention can take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0204] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of methods, terminal devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the processes and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal devices generate a device for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0205] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing terminal device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured product including an instruction device, and the instruction device implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0206] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device, so that a series of operation steps are executed on the computer or other programmable terminal device to generate a computer-implemented process. Therefore, the instructions executed on the computer or other programmable terminal device provide for implementing the functions in the processFigure 1 one process or multiple processes and / or boxes Figure 1 steps of functions specified in one box or multiple boxes.

[0207] Although the preferred embodiments of the embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications to these embodiments once they know the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

[0208] Finally, it should also be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or terminal device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or terminal device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of additional identical elements in the process, method, article or terminal device comprising the element.

[0209] The above has introduced in detail a data annotation method, device, computer-readable storage medium and electronic device provided by the present invention. Specific examples are used in this text to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A data annotation method, characterized in that, The method includes: Obtaining a driving data packet collected by a target vehicle; Based on a preset annotation rule, classifying and annotating the driving intention of at least one obstacle to be annotated in the driving data packet to obtain a first annotation result file; Obtaining first qualified annotation data and first unqualified annotation data returned by an artificial verification module for the first annotation result file, and storing the first qualified annotation data in a qualified data set; When it is detected that the annotation rule is updated, based on the updated annotation rule, updating the driving intention corresponding to the first unqualified annotation data to obtain a second annotation result file; Obtaining second qualified annotation data returned by the artificial verification module for the second annotation result file, and storing the second qualified annotation data in the qualified data set.

2. The data annotation method according to claim 1, characterized in that, The annotation rule includes an effective trajectory determination sub-rule, an effective lane line determination sub-rule, and a driving intention determination sub-rule; The step of classifying and annotating the driving intention of at least one obstacle to be annotated in the driving data packet based on a preset annotation rule to obtain a first annotation result file includes: Based on the effective trajectory determination sub-rule, determining the effective driving trajectory of the obstacle to be annotated in the driving data packet; Based on the effective lane line determination sub-rule, determining at least one effective lane line in the driving data packet; Based on the driving intention determination sub-rule, determining at least one interaction relationship sequence between the effective driving trajectory of each obstacle to be annotated and at least one of the effective lane lines, and based on at least one of the interaction relationship sequences, determining the driving intention corresponding to each obstacle to be annotated; Generating a driving intention label corresponding to each obstacle to be annotated to obtain the first annotation result file.

3. The data annotation method according to claim 2, characterized in that, The step of determining the effective driving trajectory of the obstacle to be annotated in the driving data packet based on the effective trajectory determination sub-rule includes: Based on the effective trajectory determination sub-rule, determining a trajectory point number threshold, a trajectory curvature threshold, and a continuity parameter threshold; For any one of the obstacles to be annotated, determining a plurality of trajectory points of the obstacle to be annotated in a plurality of consecutive image frames of the driving data packet and a trajectory curve and a continuity parameter corresponding to the plurality of trajectory points; wherein, the continuity parameter characterizes the continuity degree between the plurality of trajectory points; When the number of trajectory points is greater than the trajectory point number threshold, and the curvature of the trajectory curve is less than the trajectory curvature threshold, and the continuity parameter is greater than the continuity parameter threshold, determining the trajectory curve of the obstacle to be annotated as an effective driving trajectory.

4. The data annotation method according to claim 2, characterized in that, The step of determining the effective lane line in the driving data packet based on the effective lane line determination sub-rule includes: Based on the effective lane line determination sub-rule, determining a lane line length threshold, a lane line curvature threshold, and a lane line distance threshold; For any lane line in the driving data packet, when the lane line length corresponding to the lane line is greater than the lane line length threshold, and the lane line curvature is less than the lane line curvature threshold, and the lane line distance is less than the lane line distance threshold, determine that the lane line is a valid lane line.

5. The data annotation method according to claim 2, characterized in that, The step of determining at least one interaction relationship sequence between the effective driving trajectory of each to-be-annotated obstacle and at least one of the valid lane lines based on the driving intention determination sub-rule, and determining the driving intention corresponding to each to-be-annotated obstacle based on at least one of the interaction relationship sequences includes: Based on the driving intention determination sub-rule, determine the interaction type definition rule and the driving intention mapping rule; the driving intention mapping rule represents different driving intentions corresponding to different combinations of interaction types; Based on the interaction type definition rule, determine the interaction relationship sequence between the effective driving trajectory of each to-be-annotated obstacle and each of the valid lane lines in multiple consecutive image frames of the driving data packet; the interaction relationship sequence represents the combination of interaction types corresponding to the multiple consecutive image frames; Based on the driving intention mapping rule and the combination of interaction types, determine the driving intention corresponding to each to-be-annotated obstacle.

6. The data annotation method according to claim 5, characterized in that, The interaction types include driving on the left, driving on the line, and / or driving on the right; The step of determining the driving intention corresponding to each to-be-annotated obstacle based on the driving intention mapping rule and the combination of interaction types includes: For any valid lane line, when the valid lane line is the left lane line of the target vehicle and the combination of interaction types corresponding to the valid lane line sequentially includes driving on the left and driving on the right, determine that the driving intention of the to-be-annotated obstacle is to cut in from the left; and / or, For any valid lane line, when the valid lane line is not the left lane line of the target vehicle and the combination of interaction types corresponding to the valid lane line sequentially includes driving on the left and driving on the right, determine that the driving intention of the to-be-annotated obstacle is to change lanes to the left; and / or, For any valid lane line, when the valid lane line is the right lane line of the target vehicle and the combination of interaction types corresponding to the valid lane line sequentially includes driving on the right and driving on the left, determine that the driving intention of the to-be-annotated obstacle is to cut in from the right; and / or, For any valid lane line, when the valid lane line is not the right lane line of the target vehicle and the combination of interaction types corresponding to the valid lane line sequentially includes driving on the right and driving on the left, determine that the driving intention of the to-be-annotated obstacle is to change lanes to the right; and / or, For any valid lane line, when the combination of interaction types corresponding to the valid lane line only includes driving on the line, or when the combination of interaction types only includes driving on the line and driving on the left, or when the combination of interaction types only includes driving on the line and driving on the right, determine that the driving intention of the to-be-annotated obstacle is to drive on the line; and / or, For any valid lane line, when the combination of interaction types corresponding to the valid lane line only includes driving on the left or driving on the right, determine that the driving intention of the obstacle to be labeled is cruising straight ahead.

7. The data annotation method according to claim 2, characterized in that, Before the step of classifying and labeling the first unqualified labeled data based on the updated labeling rule when it is detected that the labeling rule is updated, the method further includes: In response to an update instruction for the valid trajectory determination sub-rule, the valid lane line determination sub-rule, and / or the driving intention determination sub-rule, update the valid trajectory determination sub-rule, the valid lane line determination sub-rule, and / or the driving intention determination sub-rule.

8. A data annotation device, characterized in that, The device includes; A data acquisition module for acquiring a driving data packet collected by a target vehicle;; A first labeling module for classifying and labeling the driving intention of at least one obstacle to be labeled in the driving data packet based on a preset labeling rule to obtain a first labeled result file; A first storage module for obtaining the first qualified labeled data and the first unqualified labeled data returned by the manual verification module for the first labeled result file, and storing the first qualified labeled data in a qualified data set; A second labeling module for updating the driving intention corresponding to the first unqualified labeled data based on the updated labeling rule when it is detected that the labeling rule is updated, to obtain a second labeled result file; A second storage module for obtaining the second qualified labeled data returned by the manual verification module for the second labeled result file, and storing the second qualified labeled data in the qualified data set.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the data labeling method according to any one of claims 1 to 7.

10. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the data labeling method according to any one of claims 1 to 7.

Images