Driving decision data labeling method and device, equipment and storage medium

Abstract

The invention provides a driving decision data labeling method and device, equipment and a storage medium, and the method comprises the steps: obtaining the driving behavior decision data of a current vehicle, the driving behavior decision data comprising the driving environment data of the current vehicle; and marking the driving behavior decision data based on the driving behavior decision threshold to obtain marking information which is used for marking the driving behavior corresponding to the current vehicle. Therefore, the driving behavior decision threshold can be dynamically adjusted based on the driving environment data, so that the driving behavior decision threshold can adapt to the change of the current vehicle driving environment, the driving behavior decision data is labeled based on the driving behavior decision threshold, the labeling accuracy can be improved, and further, the accuracy of labeling is improved. And the labeled driving decision data is used as training data, so that the training precision of the driving behavior decision model can be ensured.

Description

Technical Field

[0001] The present application relates to the technical field of driving data analysis and processing, and specifically to a method, device, equipment and storage medium for labeling driving decision data. Background Art

[0002] With the technological transformation, compared with the traditional rule-based autonomous driving decision-making method, the learning-based decision-making method relies on a large amount of accurately labeled training data to improve performance. The manual labeling method is inefficient and costly, but accurate and reliable; although automated labeling can reduce costs and increase efficiency, it faces the challenge of a large number of abnormal samples, which can be retained to improve sample diversity or removed to ensure sample accuracy. Using an overly aggressive or overly conservative labeling strategy here will affect subsequent algorithm training.

[0003] In related technologies, a fixed threshold processing strategy is usually used to label autonomous driving decision data. However, such methods cannot adapt to changes in the vehicle driving environment, and the labeling accuracy is low. It is impossible to ensure that the learning-based decision method corresponds to the training accuracy of the driving behavior decision model, thereby reducing the decision performance of the driving behavior decision model. Summary of the invention

[0004] The present application provides a method, device, equipment and readable storage medium for labeling driving decision data. The method can adapt to changes in the vehicle driving environment and improve the labeling accuracy of driving decision data. Using the labeled driving decision data as training data can ensure the training accuracy of the driving behavior decision model.

[0005] The technical solution of this application is implemented as follows:

[0006] An embodiment of the present application provides a method for labeling driving decision data, including: obtaining driving behavior decision data of a current vehicle; the driving behavior decision data includes driving environment data of the current vehicle; labeling the driving behavior decision data based on a driving behavior decision threshold to obtain labeling information; wherein the driving behavior decision threshold is dynamically adjusted based on the driving environment data, and the labeling information is used to identify the driving behavior corresponding to the current vehicle.

[0007] According to the above technical means, the driving behavior decision data of the current vehicle is obtained, and the driving behavior decision data includes the driving environment data of the current vehicle; the driving behavior decision data is annotated based on the driving behavior decision threshold to obtain the annotated information, and the annotated information is used to identify the driving behavior corresponding to the current vehicle. In this way, since the driving behavior decision threshold can be dynamically adjusted based on the driving environment data, the driving behavior decision threshold can adapt to the changes in the driving environment of the current vehicle. Therefore, the driving behavior decision data is annotated based on the driving behavior decision threshold, which can improve the accuracy of the annotation. Furthermore, the annotated driving decision data can be used as training data to ensure the training accuracy of the driving behavior decision model.

[0008] Furthermore, the driving behavior decision data includes multiple image frames; the driving decision data is annotated based on the driving behavior decision threshold to obtain annotation information, including: determining the first lane centerline of the lane corresponding to the current vehicle in each image frame based on the position information of the current vehicle and the position information of the center line of each lane in each image frame; determining the lane width of the lane corresponding to the current vehicle in each image frame based on the position information of the first lane centerline and the position information of the current vehicle in each image frame; based on the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width and the driving behavior decision threshold, the driving decision data is annotated to obtain annotation information.

[0009] According to the above technical means, by determining the first lane centerline of the lane corresponding to the current vehicle in each image frame, determining the lane width according to the position information of the current vehicle and the position information of the first lane centerline in each image frame, and based on the position information of the first lane centerline, the position information of the current vehicle, the driving behavior decision threshold and the lane width, the driving behavior of the current vehicle in different image frames can be quickly determined, thereby realizing rapid labeling of the driving decision data.

[0010] Furthermore, the driving behavior decision threshold includes a threshold for the number of frames of continuous crossing, and the annotation information includes information on whether the current vehicle has crossed the line; based on the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width and the driving behavior decision threshold, the driving decision data is annotated to obtain the annotation information, including: if the center line of the first lane in the current image frame is different from the center line of the first lane in the next image frame, according to the position information of the current vehicle in the current image frame, the position information of the center line of the first lane in the current image frame, the position information of the center line of the first lane in the next image frame and the lane width, determining whether the next image frame meets the vehicle crossing condition; determining the number of consecutive image frames that meet the vehicle crossing condition; annotating the consecutive image frames according to the number of frames and the threshold for the number of frames of continuous crossing to obtain information on whether the current vehicle has crossed the line.

[0011] According to the above technical means, by determining the number of consecutive image frames that meet the vehicle crossing the line condition, and based on the number of consecutive image frames and the threshold of the number of continuous frames of crossing the line, it is possible to directly determine whether there is any crossing of the line behavior corresponding to the current vehicle in the continuous image frames, thereby realizing the labeling of the continuous image frames.

[0012] Furthermore, the driving environment data includes the number of lanes and road status data in the image frame; the method also includes: obtaining the number of lanes in the first reference image frame; the first reference image frame is any one of multiple consecutive image frames; based on the road status data of the lane corresponding to the current vehicle in the first reference image frame, the number of lanes in the first reference image frame, and the driving speed of the current vehicle within a preset time period before the corresponding moment of the first reference image frame, determine the threshold number of frames for continuous crossing the line.

[0013] According to the above technical means, by combining the number of lanes in the first reference image frame, the road status data and the current vehicle's driving speed within a preset time period before the corresponding moment of the first reference image frame, it is possible to determine a line crossing duration frame number threshold that changes dynamically with the driving environment, so that the driving behavior decision data can be labeled based on the dynamically changing line crossing duration frame number threshold, and more accurate labeling results can be obtained.

[0014] Furthermore, the driving behavior decision threshold includes a lane change duration threshold, and the annotation information includes information on whether the current vehicle is changing lanes; the driving decision data is annotated based on the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width, and the driving behavior decision threshold, to obtain the annotation information, including: obtaining a plurality of consecutive crossing-line image frames in which the annotation information is that the current vehicle is crossing the line in the plurality of image frames; determining the distance between the current vehicle and the center line of the corresponding first lane in the crossing-line image frame according to the position information of the current vehicle in the crossing-line image frame and the position information of the center line of the corresponding first lane in the crossing-line image frame; determining the crossing-line image frame as a lane change candidate image frame if the distance and the lane width meet the first condition; annotating the candidate lane change image frame according to the time information of the lane change candidate image frame and the lane change duration threshold to obtain information on whether the current vehicle is changing lanes.

[0015] According to the above technical means, a lane changing candidate image frame is determined from a plurality of consecutive line crossing image frames, and based on the time information of the lane changing candidate image frame and the lane changing duration threshold, it can be directly determined whether the candidate lane changing image frame is an image frame for the current vehicle to change lanes, thereby achieving the labeling of the lane changing candidate image frame.

[0016] Furthermore, the driving environment data includes the traffic flow in the image frame; the method also includes: obtaining the traffic flow of vehicles in the second reference image frame and the distance information between the current vehicle and the first vehicle in the second reference image frame; the second reference image frame is any image frame in the lane change candidate image frame; the first vehicle may be other vehicles adjacent to the current vehicle; and determining the lane change duration threshold based on the traffic flow and distance information in the second reference image frame.

[0017] According to the above technical means, since the traffic flow of vehicles in the second reference image frame and the distance information between the current vehicle and the first vehicle in the second reference image frame can change with the change of the driving environment of the current vehicle, the lane change time threshold determined according to the traffic flow of vehicles in the second reference image frame and the distance information between the current vehicle and the first vehicle in the second reference image frame can adapt to the changes in the vehicle driving environment, so that the driving behavior decision data can be labeled based on the lane change time threshold, which can improve the labeling accuracy.

[0018] Further, based on the position information of the center line of the first lane and the position information of the current vehicle in each image frame, the lane width of the lane corresponding to the current vehicle in each image frame is determined, including: based on the position information of the current vehicle in the third reference image frame and the position information of the center line of the first lane corresponding to the current vehicle in the third reference image frame, the initial lane width of the lane corresponding to the current vehicle in the third reference image frame is determined; the third reference image frame is any one of the multiple image frames; if the absolute value of the difference between the initial lane width and the lane width of the fourth reference image frame is greater than a first preset threshold, the lane width information in a preset number of image frames before the third reference image frame is obtained; the fourth reference image frame is the next image frame of the third reference image frame; based on the initial lane width and the lane width information, the target lane width of the lane corresponding to the current vehicle in the fourth reference image frame is determined.

[0019] According to the above technical means, by determining the initial width of the third reference image frame, when the absolute value of the difference between the initial width and the lane width of the next image frame (fourth reference image frame) of the third reference image is greater than a preset threshold, the lane width of the lane corresponding to the current vehicle in the fourth reference image frame is calculated according to the road section characteristics through the initial lane width and the lane width information in a preset number of image frames before the third reference image frame, so that a more accurate lane width value can be obtained, thereby ensuring the accuracy of subsequent driving behavior decision data labeling.

[0020] Furthermore, the method also includes: obtaining a manually annotated driving behavior decision data set; the driving behavior decision data set includes driving behavior decision samples and corresponding real annotation information; annotating the driving behavior decision samples based on a driving behavior decision threshold to obtain predicted annotation information; determining the annotation accuracy for the driving behavior decision samples based on the real annotation information and the predicted annotation information; if the annotation accuracy is less than a second preset threshold, updating the driving behavior decision threshold to obtain a new driving behavior decision threshold, so as to annotate the driving behavior decision data based on the new driving behavior decision threshold.

[0021] According to the above technical means, the labeling accuracy of the driving behavior decision samples based on the driving behavior decision threshold is determined by using the real labeling information of the driving behavior decision samples in the manually labeled driving behavior decision data set. When the labeling accuracy is less than the second preset threshold, the driving behavior decision threshold is updated, which can improve the accuracy of the driving behavior decision threshold and ensure the accuracy of subsequent labeling of the driving behavior decision data.

[0022] Furthermore, the method also includes: obtaining reference driving behavior decision data whose annotation information indicates abnormal driving behavior of the current vehicle; the abnormal driving behavior at least includes the current vehicle changing lanes without turning on the lights and moving away from the center line of the lane; based on the reference driving behavior decision data and the annotation information corresponding to the reference driving behavior decision data, establishing an abnormal driving behavior data set to train the driving decision model based on the abnormal driving behavior data set.

[0023] According to the above-mentioned technical means, by obtaining reference driving behavior decision data whose annotation information indicates abnormal driving behavior of the current vehicle, and establishing an abnormal driving behavior data set for training a driving behavior decision model based on the reference driving behavior decision data and the corresponding annotation information, the diversity of the driving behavior decision model training samples can be ensured, thereby improving the training accuracy of the driving behavior decision model.

[0024] The embodiment of the present application provides a driving decision data labeling device, including:

[0025] The first acquisition module is used to acquire the driving behavior decision data of the current vehicle; the driving behavior decision data includes the driving environment data of the current vehicle;

[0026] The first labeling module is used to label the driving behavior decision data based on the driving behavior decision threshold to obtain labeling information; wherein the driving behavior decision threshold is dynamically adjusted based on the driving environment data, and the labeling information is used to identify the driving behavior corresponding to the current vehicle.

[0027] The embodiment of the present application provides a driving decision data labeling device, including:

[0028] A memory for storing computer programs that can be run on the processor;

[0029] The processor is used to execute the driving decision data labeling method provided in the embodiment of the present application when running the computer program.

[0030] An embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. The computer-executable instructions are configured to execute the above-mentioned driving decision data labeling method. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A flowchart of a method for labeling driving decision data provided in an embodiment of the present application;

[0032] Figure 2 A flowchart of a method for self-labeling an autonomous driving decision data set provided in an embodiment of the present application;

[0033] Figure 3 A schematic diagram of a method flow for adaptively labeling based on a dynamic threshold provided in an embodiment of the present application;

[0034] Figure 4 A schematic diagram of an abnormal sample type provided in an embodiment of the present application;

[0035] Figure 5 A schematic flow chart of a method for post-verification of a dynamic threshold provided in an embodiment of the present application;

[0036] Figure 6 A schematic diagram of the structure of a driving decision data labeling device provided in an embodiment of the present application;

[0037] Figure 7 A schematic diagram of the composition structure of a driving decision data labeling device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0039] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.

[0040] In the following description, reference is made to “some embodiments\other embodiments”, which describe a subset of all possible embodiments, but it can be understood that “some embodiments\other embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0041] In the following description, the terms "first\second\third" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0042] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0043] In the related technologies, a decision-making labeling method for autonomous driving data uses fixed confidence intervals and behavior thresholds to perform labeling a posteriori judgments to improve obstacle avoidance safety; another method for labeling autonomous driving data focuses on achieving 4D (Dimension) scene reconstruction to obtain automatic labeling results using a neural network model. Both of the above solutions are targeted at limited driving scenarios and usually do not consider abnormal sample processing methods. At the same time, the fixed threshold processing strategy used in them cannot adapt to environmental changes, further leading to accuracy loss, especially abnormal samples are difficult to use effectively, making the labeled samples too single and losing sample diversity.

[0044] Based on the problems existing in the related art, the embodiment of the present application provides a method for labeling driving decision data, which can adapt to changes in the vehicle driving environment and improve the labeling accuracy of driving decision data. Using the labeled driving decision data as training data can ensure the training accuracy of the driving behavior decision model. Figure 1 FIG. 1 is a flow chart of a method for labeling driving decision data provided by an embodiment of the present application, the method comprising the following steps:

[0045] S101. Acquire driving behavior decision data of the current vehicle.

[0046] It should be noted that the driving behavior decision data may be data used to determine the vehicle's driving behavior, which may include changing lanes left or right, crossing the line, going straight, following a vehicle, etc. The current vehicle may be an observed autonomous driving vehicle.

[0047] In some embodiments, the driving behavior decision data includes the driving environment data of the current vehicle, which may include the slope and curvature of the road on which the current vehicle is traveling, obstacles around the current vehicle, other vehicles, etc. The driving behavior decision data may also include the driving track of the current vehicle, the vehicle driving status information, and the corresponding map segment information, and the driving status information may include the driving speed, acceleration, braking operation, etc. of the current vehicle.

[0048] In some embodiments, the collected original driving behavior decision data may be preprocessed to obtain processed driving behavior decision data. The original driving behavior decision data may be obtained in real time according to the current driving conditions of the vehicle, or the pre-collected driving behavior decision data may be obtained from a local database, a remote database, etc. The preprocessing of the collected original driving behavior decision data includes but is not limited to data parsing, screening, synchronization, slicing, etc. The original driving behavior decision data set may be divided into structured segment data so as to uniformly process different data segments.

[0049] In some embodiments, current vehicles are usually equipped with sensor devices, so online / offline software can be used to collect raw driving behavior decision data in full or triggered mode, where full collection can be to collect all driving behavior decision data of the current vehicle, and triggered collection can be to collect driving behavior decision data corresponding to a specific driving behavior, which can be determined based on the analysis scenario of the driving behavior. Depending on the collected data, raw unstructured data such as images, radars, global positioning systems (GPS), inertial measurement units (IMU), etc. can be used to form a raw data set with a time scale.

[0050] In some embodiments, the original driving behavior decision data may be obtained by the current vehicle during a real autonomous driving process or by driving experts to ensure the reliability of the data set.

[0051] S102: Annotate the driving behavior decision data based on the driving behavior decision threshold to obtain annotation information.

[0052] In some embodiments, the driving behavior decision threshold can be used to determine whether the driving behavior corresponding to the driving behavior decision data is a specific driving behavior, and the specific driving behavior can include crossing the line, changing lanes left, changing lanes right, etc. The driving behavior decision threshold can include a lane change duration threshold, a crossing line duration frame number threshold, etc.

[0053] In some embodiments, the driving behavior decision threshold can be dynamically adjusted based on driving environment data. For example, the driving behavior decision threshold can be dynamically adjusted based on data such as the curvature, slope, traffic flow, and the distance between the current vehicle and other vehicles of the road corresponding to the current vehicle, so that the driving behavior decision threshold can adapt to changes in the current vehicle driving environment.

[0054] In some embodiments, the annotation information is used to identify the driving behavior corresponding to the current vehicle. The annotation information can be identified in the form of text. For example, if the driving behavior corresponding to a certain driving behavior decision data is determined to be a left lane change based on the driving behavior decision threshold, the corresponding identification information can be that the driving behavior of the current vehicle is a left lane change.

[0055] In an embodiment of the present application, the driving behavior decision data of the current vehicle is obtained, and the driving behavior decision data includes the driving environment data of the current vehicle; the driving behavior decision data is annotated based on the driving behavior decision threshold to obtain the annotation information, and the annotation information is used to identify the driving behavior corresponding to the current vehicle. In this way, since the driving behavior decision threshold can be dynamically adjusted based on the driving environment data, the driving behavior decision threshold can adapt to the changes in the driving environment of the current vehicle. Therefore, the driving behavior decision data is annotated based on the driving behavior decision threshold, which can improve the accuracy of the annotation. Furthermore, the annotated driving decision data can be used as training data to ensure the training accuracy of the driving behavior decision model.

[0056] In some embodiments of the present application, the driving behavior decision data includes multiple image frames; the driving decision data is labeled based on the driving behavior decision threshold to obtain the labeling information, that is, the above-mentioned step S102 can be implemented by the following steps S1021 to S1023, and each step is described separately below.

[0057] S1021. Determine a first lane centerline of a lane corresponding to the current vehicle in each image frame according to the position information of the current vehicle and the position information of each lane centerline in each image frame.

[0058] In some embodiments, the position of the current vehicle in different image frames may be different, and each image frame may include one or more lane center lines. The different lane center lines in multiple image frames and the position information of the different lane center lines can be determined first, and then the first lane center line corresponding to the current vehicle in each image frame can be determined.

[0059] In some embodiments, the position information of the current vehicle in each image frame may be the position coordinates corresponding to the driving trajectory of the current vehicle in the image frame. The lane centerline coordinate system may be first aligned with the driving trajectory coordinate system of the current vehicle, and the position information of the centerline of the first lane may be converted into the Frenet coordinate system for representation, thereby obtaining the position information of the centerlines of each lane and the position information of the current vehicle in each image frame.

[0060] In some embodiments, the first lane centerline may be the lane centerline closest to the current vehicle among multiple lane centerlines in a single image frame. The first lane centerline corresponding to the current vehicle in different image frames may be different. The first lane centerline corresponding to the current vehicle in each image frame can be determined based on the minimum projection distance of the current vehicle from each lane centerline.

[0061] For example, in the process of determining the center line of the first lane corresponding to the current vehicle, the calculation method can be as follows based on the minimum projection distance principle: Let the coordinates of the current vehicle position point be P(x p ,y p ), the set of points on the lane centerline is L = {(x i ,y i )|I=1,2,..,n}, where n is the number of points. The minimum distance point P min On L, its coordinates are (x min ,y min ), satisfying formula (1):

[0062]

[0063] If P min It is not an endpoint. The projection point P is calculated by formula (2) proj To P min The distance d on the line segment proj :

[0064]

[0065] If P min is an endpoint, calculate the projection distance from P to the extension line of the endpoint, and compare it with the distance from P to the other endpoint to get the minimum value. If there is an unmatched frame in the sample segment, the sample is determined to be processed as an abnormal sample.

[0066] S1022. Determine a lane width of a lane corresponding to the current vehicle in each image frame according to the position information of the center line of the first lane and the position information of the current vehicle in each image frame.

[0067] In some embodiments, a second lane centerline that is opposite to the first lane centerline and closest to the current vehicle may be determined. The first lane centerline and the second lane centerline may be lane centerlines located on both sides of the current vehicle. Based on the position information of the first lane centerline, the position information of the second lane centerline, and the position information of the current vehicle in each image frame, the lane width of the lane corresponding to the current vehicle in each image frame is determined. The lane width may be the distance between the left and right boundary lines of the lane.

[0068] In some embodiments, the first distance between the current vehicle and the lane centerline on the left (the first lane centerline / the second lane centerline) can be calculated to obtain the left width of the lane, and the right width of the lane can be obtained according to the second distance between the current vehicle and the lane centerline on the right (the first lane centerline / the second lane centerline), and then the total width of the lane, i.e., the lane width, can be obtained based on the left width and the right width of the lane. Wherein, when the first lane centerline is the lane centerline on the left side of the current vehicle, the second lane centerline is the lane centerline on the right side of the current vehicle; when the first lane centerline is the lane centerline on the right side of the current vehicle, the second lane centerline is the lane centerline on the left side of the current vehicle.

[0069] For example, the left lane width W can be calculated according to the following formula (3): l :

[0070] W l =min P∈Cl (x p -x Cr ) 2 +(y p -y Cr ) 2 ) 1 / 2 (3);

[0071] Among them, x Cr ,y Cr Respectively represent the horizontal and vertical coordinates of the center line of the left lane of the current vehicle, and the width W on the right side r With left width W l Similar calculation, if only one side exists, the total lane width W = W l +W r .

[0072] S1023. Based on the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width, and the driving behavior decision threshold, the driving decision data is labeled to obtain the labeled information.

[0073] In some embodiments, the driving behavior of the current vehicle corresponding to different image frames can be determined based on the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width and the driving behavior decision threshold, and then the driving behavior of different image frames can be annotated to obtain the annotation information.

[0074] According to the above technical means, by determining the first lane centerline of the lane corresponding to the current vehicle in each image frame, determining the lane width according to the position information of the current vehicle and the position information of the first lane centerline in each image frame, and based on the position information of the first lane centerline, the position information of the current vehicle, the driving behavior decision threshold and the lane width, the driving behavior of the current vehicle in different image frames can be quickly determined, thereby realizing rapid labeling of the driving decision data.

[0075] In some embodiments of the present application, the driving behavior decision threshold includes a threshold for the number of frames of continuous crossing the line, and the annotation information includes information on whether the current vehicle has crossed the line. Based on this, the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width and the driving behavior decision threshold, in the process of annotating the driving decision data, if it is determined that the center line of the first lane in the current image frame and the center line of the first lane in the next image frame are different, determine whether the next image frame meets the vehicle crossing the line condition based on the position information of the current vehicle in the current image frame, the position information of the center line of the first lane in the current image frame, the position information of the center line of the first lane in the next image frame, and the lane width; determine the number of consecutive image frames that meet the vehicle crossing the line condition; and annotate the consecutive image frames based on the number of frames and the threshold for the number of frames of continuous crossing the line to obtain information on whether the current vehicle has crossed the line.

[0076] In some embodiments, it is possible to determine whether the centerline of the first lane in the current image frame is the same as the centerline of the first lane in the next image frame, that is, whether the centerline of the first lane in the next image frame is the successor of the centerline of the first lane in the current image frame, based on the Euclidean distance between the endpoint of the centerline of the first lane in the current image frame and the endpoint of the centerline of the first lane in the next image frame, and the angle formed by the vector determined by the two endpoints and the current vehicle driving direction. If the Euclidean distance between the endpoint of the centerline of the first lane in the current image frame and the endpoint of the centerline of the first lane in the next image frame is less than a preset distance threshold, and the angle formed by the vector determined by the two endpoints and the current vehicle driving direction is less than a preset angle value, it can be determined that the centerline of the first lane in the current image frame and the centerline of the first lane in the next image frame are the same, otherwise it is considered that the two are different.

[0077] In some embodiments, a first distance between the current vehicle and the center line of the first lane in the current image frame can be determined based on the position information of the current vehicle in the current image frame and the position information of the center line of the first lane in the current image frame; a second distance between the current vehicle and the center line of the first lane in the next image frame can be determined based on the position information of the current vehicle in the current image frame and the position information of the center line of the first lane in the next image frame; a third distance between the current vehicle and the center line of the first lane in the next image frame can be determined based on the position information of the current vehicle in the next image frame and the position information of the center line of the first lane in the next image frame. If it is then determined that the first distance is less than half the lane width (half-width of the lane), the second distance is greater than half-width of the lane, and the third distance is less than half-width of the lane, it can be determined that the next image frame satisfies the vehicle crossing line condition.

[0078] In some embodiments, a judgment may be made as to whether each image frame satisfies the vehicle crossing-line condition, and the number of consecutive image frames that satisfy the image frame vehicle crossing-line condition may be counted. When it is determined that the number of consecutive image frames that satisfy the image frame vehicle crossing-line condition is less than a continuous crossing-line frame number threshold, it may be determined that the driving behavior of the current vehicle corresponding to each consecutive image frame is crossing the line. By marking the consecutive image frames, information on whether the current vehicle has crossed the line may be obtained.

[0079] According to the above technical means, by determining the number of consecutive image frames that meet the vehicle crossing the line condition, and based on the number of consecutive image frames and the threshold of the number of continuous frames of crossing the line, it is possible to directly determine whether there is any crossing of the line behavior corresponding to the current vehicle in the continuous image frames, thereby realizing the labeling of the continuous image frames.

[0080] In some embodiments of the present application, the driving environment data includes the number of lanes in the image frame and road status data, where the number of lanes can be the total number of lanes in multiple image frames or the total number of lanes in one image frame, and the road status data can include the slope and curvature of the road on which the vehicle is currently traveling, and the threshold value for the number of frames for continuous crossing the line can be dynamically adjusted based on the driving environment data.

[0081] In some embodiments, in the process of dynamically adjusting the line crossing duration frame number threshold, the number of lanes in the first reference image frame can be obtained; the line crossing duration frame number threshold is determined based on the road state data of the lane corresponding to the current vehicle in the first reference image frame, the number of lanes in the first reference image frame, and the current vehicle's driving speed within a preset time period before the corresponding moment in the first reference image frame.

[0082] In some embodiments, the first reference image frame is any one of a plurality of continuous image frames, and the line crossing continuous frame number threshold can be determined based on the slope, curvature, etc. of the lane corresponding to the current vehicle in the first reference image frame, the number of lanes in the first reference image frame, and the driving speed of the current vehicle within a preset time period before the corresponding moment of the first reference image frame (such as the average driving speed, the maximum driving speed, etc.). Since the road state data of the lane corresponding to the current vehicle in different first reference image frames, the number of lanes, and the driving speed of the current vehicle may all change, the determined line crossing frame number threshold may show dynamic changes.

[0083] In some embodiments, the current line crossing duration frame number threshold can also be updated in combination with the basic line crossing duration frame number threshold, the road complexity image coefficient, etc., so as to obtain the line crossing duration frame number threshold that changes according to the real-time driving environment changes, wherein the basic line crossing duration frame number threshold can be the initial line crossing duration frame number threshold.

[0084] For example, the line crossing duration frame number threshold N max It can be adjusted dynamically according to formula (4):

[0085]

[0086] Among them, N base is the basic maximum continuous frame number; v avg (t-Δt) is the average vehicle speed over a period of time; γ is the road complexity influence coefficient, and Cpl(t) is the complexity evaluation function, which is usually calculated based on the current road curvature, slope, number of intersections, and lane changes. This application can sum the road curvature and the number of lanes.

[0087] According to the above technical means, by combining the number of lanes in the first reference image frame, road status data and the current vehicle's driving speed within a preset time period before the corresponding moment of the first reference image frame, a line crossing duration frame number threshold that changes dynamically with the driving environment can be determined, so that the driving behavior decision data can be labeled based on the dynamically changing line crossing duration frame number threshold, and more accurate labeling results can be obtained.

[0088] In some embodiments of the present application, the driving behavior decision threshold includes a lane change duration threshold, and the annotation information includes information on whether the current vehicle is changing lanes. In the process of annotating the driving decision data based on the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width and the driving behavior decision threshold, a plurality of continuous crossing-line image frames in which the annotation information is that the current vehicle is crossing the line can be obtained; the distance of the current vehicle in the crossing-line image frame is determined according to the position information of the current vehicle in the crossing-line image frame and the position information of the center line of the corresponding first lane in the crossing-line image frame; if the distance is less than and the lane width satisfies the first condition, the crossing-line image frame is determined as a candidate image frame for lane change; the candidate lane change image frame is annotated according to the time information of the candidate lane change image frame and the lane change duration threshold to obtain information on whether the current vehicle is changing lanes.

[0089] In some embodiments, the first condition may be that the distance between the current vehicle and the center line of the corresponding first lane is less than half the lane width, or that the distance between the current vehicle and the center line of the corresponding first lane is less than or equal to half the lane width.

[0090] In some embodiments, the distance between the current vehicle and the center line of the first lane in each cross-line image frame can be determined based on the position information of the current vehicle in the cross-line image frame and the position information of the center line of the first lane corresponding to the current vehicle; then, when it is determined that the distance satisfies the first condition, the corresponding cross-line image frame can be determined as a lane change candidate image frame.

[0091] In some embodiments, the lane changing duration of the current vehicle can be determined based on the time information of the lane changing candidate image frame; based on the lane changing duration and the lane changing duration threshold, it can be determined whether the current vehicle has changed lanes. If the lane changing duration is less than the lane changing duration threshold, it can be determined that the current vehicle has changed lanes; otherwise, it can be determined that the current vehicle has not changed lanes.

[0092] In some embodiments, for the lane changing candidate image frames determined to be the current vehicle changing lanes, it can be marked as the driving behavior of the current vehicle changing lanes; for the lane changing candidate image frames determined to be the current vehicle not changing lanes, it can be further determined whether the driving behavior corresponding to the current vehicle is straight driving, following the vehicle, or abnormal driving behavior.

[0093] According to the above technical means, a lane changing candidate image frame is determined from a plurality of consecutive line crossing image frames, and based on the time information of the lane changing candidate image frame and the lane changing duration threshold, it can be directly determined whether the candidate lane changing image frame is an image frame for the current vehicle to change lanes, thereby achieving the labeling of the lane changing candidate image frame.

[0094] In some embodiments of the present application, the driving environment data includes the traffic flow in the image frame. In the process of dynamically adjusting the lane change time threshold based on the driving environment data, the traffic flow of vehicles in the second reference image frame and the distance information between the current vehicle and the first vehicle in the second reference image frame can be obtained; the lane change time threshold is determined based on the traffic flow and distance information in the second reference image frame.

[0095] In some embodiments, the second reference image frame is any one of the lane change candidate image frames; the first vehicle may be another vehicle adjacent to the current vehicle. The traffic volume in the second reference image frame may be determined based on the vehicle density in the second reference image frame, and the spacing information between the current vehicle and the first vehicle may be the distance between the current vehicle and the first vehicle. In the case where the first vehicle includes multiple vehicles, the average value of the distance between the current vehicle and each of the first vehicles may be used as the spacing information.

[0096] In some embodiments, the lane change time threshold may be updated in combination with a basic lane change time threshold, a traffic density influence coefficient, etc., wherein the basic lane change time threshold may be an initial lane change time threshold.

[0097] For example, the lane change time threshold T change It can be adjusted dynamically by formula (5):

[0098] T change = T base +k flow TraD(t)-k gap G avg (t) (5);

[0099] Among them, T base is the basic lane change duration; k flow is the traffic density influence coefficient; TraD(t) is the current traffic density, which is usually calculated based on the number of vehicles in the road section; k gap G is the spacing influence coefficient; avg (t) is the average distance between surrounding vehicles.

[0100] According to the above technical means, since the traffic flow of vehicles in the second reference image frame and the distance information between the current vehicle and the first vehicle in the second reference image frame can change with the change of the driving environment of the current vehicle, the lane change time threshold determined according to the traffic flow of vehicles in the second reference image frame and the distance information between the current vehicle and the first vehicle in the second reference image frame can adapt to the changes in the vehicle driving environment, so that the driving behavior decision data can be labeled based on the lane change time threshold, which can improve the labeling accuracy.

[0101] In some embodiments of the present application, in the process of determining the lane width of the lane corresponding to the current vehicle in each image frame based on the position information of the center line of the first lane and the position information of the current vehicle in each image frame, the initial lane width of the lane corresponding to the current vehicle in the third reference image frame can be determined based on the position information of the current vehicle in the third reference image frame and the position information of the center line of the first lane corresponding to the current vehicle in the third reference image frame; if the absolute value of the difference between the initial lane width and the lane width of the fourth reference image frame is greater than a preset threshold, the width information of the lane in a preset number of image frames before the third reference image frame is obtained; based on the initial lane width and the lane width information, the target lane width of the lane corresponding to the current vehicle in the third reference image frame is determined.

[0102] In some embodiments, the third reference image frame is any one of the plurality of image frames; the fourth reference image frame is the next image frame of the third reference image frame. The lane width information in the preset number of image frames before the third reference image frame may be an average value, a maximum value, etc. of the lane width in the plurality of image frames before the third reference image frame.

[0103] In some embodiments, if the absolute value of the difference between the initial lane width and the lane width of the fourth reference image frame is greater than a preset threshold, it may indicate that the lane width undergoes a significant mutation when the current vehicle enters the driving environment corresponding to the fourth reference image frame from the driving environment corresponding to the third reference image frame. The width of the lane corresponding to the current vehicle in the fourth reference image frame may be determined based on the width of the lane in the third reference image frame and the width of the lane in the image frame before the third reference image frame.

[0104] In some embodiments, the lane width of the lane in the fourth reference image frame, i.e., the target lane width, can be determined based on the lane width of the corresponding lane in the third reference image frame and the average lane width in multiple image frames before the third reference image frame, as well as the previous influence coefficient of the current frame (the third reference image frame). For example, the target lane width W corresponding to the fourth reference image frame is i It can be calculated by the following formula (6):

[0105] W i = |W base | + k wl W avg (t - Δt) (6);

[0106] Among them, |W base | is the basic lane width threshold. When the width calculation is abnormal, this value is reset first; k wl is the influence coefficient of the previous segment of the current frame, W avg (t-Δt) is the average width of the front section of the road within the time Δt.

[0107] According to the above technical means, by determining the initial width of the third reference image frame, when the absolute value of the difference between the initial width and the lane width of the next image frame (fourth reference image frame) of the third reference image is greater than a preset threshold, the lane width of the lane corresponding to the current vehicle in the fourth reference image frame is calculated according to the road section characteristics through the initial lane width and the lane width information in a preset number of image frames before the third reference image frame, so that a more accurate lane width value can be obtained, thereby ensuring the accuracy of subsequent driving behavior decision data labeling.

[0108] In some embodiments of the present application, after the driving behavior decision data is labeled based on the driving behavior decision threshold and the labeling information is obtained, a manually labeled driving behavior decision data set can also be obtained; the driving behavior decision samples are labeled based on the driving behavior decision threshold to obtain predicted labeling information; the labeling accuracy for the driving behavior decision samples is determined based on the actual labeling information and the predicted labeling information; if the labeling accuracy is less than a second preset threshold, the driving behavior decision threshold is updated to obtain a new driving behavior decision threshold, so as to label the driving behavior decision data based on the new driving behavior decision threshold.

[0109] In some embodiments, the driving behavior decision data set includes driving behavior decision samples and corresponding real annotation information. The driving behavior decision samples can be image frames collected during the driving process of the autonomous driving vehicle. The real annotation information can be used to identify the actual driving behavior of the vehicle corresponding to the driving behavior decision samples. The actual driving behavior may include changing left lanes, changing right lanes, crossing the line, going straight, etc.

[0110] In some embodiments, the driving behavior decision threshold obtained after dynamic adjustment according to the driving environment data can be used to determine the driving behavior of the vehicle in the driving behavior decision sample, and the predicted labeling information can be obtained based on the result of the determination. The predicted labeling information can be used to identify the predicted driving behavior of the vehicle corresponding to the driving behavior decision sample, and the predicted driving behavior can also include left lane change, right lane change, crossing the line, straight driving, etc.

[0111] In some embodiments, the driving behavior decision samples may include multiple driving behavior decision samples, and the actual driving behavior and predicted driving behavior corresponding to each driving behavior decision sample may be different. If the actual labeling information and the predicted labeling information corresponding to the same driving behavior decision sample are different, it means that the predicted labeling information obtained based on the driving behavior decision threshold is wrong; otherwise, it means that the predicted labeling information obtained based on the driving behavior decision threshold is correct. The labeling accuracy for the driving behavior decision samples can be determined by the number of correct predicted labeling information and the total number of driving behavior decision samples.

[0112] In some embodiments, the second preset threshold may be a labeling accuracy threshold. If the labeling accuracy is less than the second preset threshold, it means that the error rate of the predicted labeling information obtained based on the driving behavior decision threshold is high, and the driving behavior decision threshold may be set incorrectly. In this case, the driving behavior decision threshold can be updated to obtain a new driving behavior decision threshold, so that the driving behavior decision data can be labeled through the new driving behavior decision threshold.

[0113] In some embodiments, the driving behavior decision threshold can be continuously verified based on a manually labeled driving behavior decision data set until the labeling accuracy of the driving behavior decision samples obtained based on the driving behavior decision threshold is greater than or equal to a second preset threshold, and then the updating of the driving behavior decision threshold can be stopped.

[0114] According to the above technical means, the labeling accuracy of the driving behavior decision samples based on the driving behavior decision threshold is determined by using the real labeling information of the driving behavior decision samples in the manually labeled driving behavior decision data set. When the labeling accuracy is less than the second preset threshold, the driving behavior decision threshold is updated, which can improve the accuracy of the driving behavior decision threshold and ensure the accuracy of subsequent labeling of the driving behavior decision data.

[0115] In some embodiments of the present application, in the process of labeling driving behavior decision data, decision data of abnormal driving behavior can also be labeled. Abnormal driving behavior at least includes the current vehicle changing lanes without turning on the lights and moving away from the center line of the lane. Abnormal driving behavior can also include the current vehicle speeding, slowing down, etc.

[0116] In some embodiments, after the driving behavior decision data is labeled to obtain the labeling information, reference driving behavior decision data whose labeling information indicates abnormal driving behavior of the current vehicle can be obtained; based on the reference driving behavior decision data, and the labeling information corresponding to the reference driving behavior decision data, an abnormal driving behavior data set is established to train the driving decision model based on the abnormal driving behavior data set.

[0117] In some embodiments, each reference driving behavior decision data and the corresponding annotation information can be used as a training sample in the abnormal driving behavior data set, and then the driving decision model is trained based on multiple training samples. The driving behavior decision model can be a model for determining the current driving behavior of the vehicle.

[0118] In some embodiments, in addition to training the driving behavior decision model based on a driving behavior data set established based on reference driving behavior decision data of abnormal driving behavior and corresponding annotation information, the driving behavior decision data corresponding to other annotation information obtained through steps S101 to S102 can also be used as training samples of the driving behavior decision model, that is, the driving behavior decision model can be trained using driving behavior decision data containing normal driving behavior and abnormal driving behavior.

[0119] According to the above-mentioned technical means, by obtaining reference driving behavior decision data whose annotation information indicates abnormal driving behavior of the current vehicle, and establishing an abnormal driving behavior data set for training a driving behavior decision model based on the reference driving behavior decision data and the corresponding annotation information, the diversity of the driving behavior decision model training samples can be ensured, thereby improving the training accuracy of the driving behavior decision model.

[0120] In an embodiment of the present application, the driving behavior decision data of the current vehicle is obtained, and the driving behavior decision data includes the driving environment data of the current vehicle; the driving behavior decision data is annotated based on the driving behavior decision threshold to obtain the annotation information, and the annotation information is used to identify the driving behavior corresponding to the current vehicle. In this way, since the driving behavior decision threshold can be dynamically adjusted based on the driving environment data, the driving behavior decision threshold can adapt to the changes in the driving environment of the current vehicle. Therefore, the driving behavior decision data is annotated based on the driving behavior decision threshold, which can improve the accuracy of the annotation. Furthermore, the annotated driving decision data can be used as training data to ensure the training accuracy of the driving behavior decision model.

[0121] Next, the implementation process of the application embodiment in the actual application scenario is introduced.

[0122] Figure 2 A flowchart of a method for self-labeling an autonomous driving decision data set provided in an embodiment of the present application is provided. The method can be implemented through the following steps S201 to S204, and each step is described separately below.

[0123] S201. Obtain original autonomous driving data (equivalent to the “driving behavior decision data” in other embodiments).

[0124] In some embodiments, based on automatic driving or expert driver driving style, at least the driving trajectory, status information and corresponding map information of the driving vehicle can be obtained. If there is a target vehicle around the vehicle, the same information should be included.

[0125] In some embodiments, the status information should at least include the driving status data of the vehicle. If there is a target vehicle, it is considered a valid target vehicle based on a certain distance range between the target vehicle and the vehicle, and its status data is obtained; the map information should at least include the center line of the lane where the vehicle is located and the center lines of the lanes in the vicinity. The coordinate system of the lane center line should be the same as the coordinate system of the vehicle's driving trajectory. Each lane center line should have a unique identifier, and the lane center lines should have a successor relationship to make the road section complete.

[0126] S202: Preprocess the original autonomous driving data to obtain sample segments.

[0127] In some embodiments, preprocessing includes but is not limited to data parsing, screening, synchronization, slicing, etc. Through preprocessing, the original data set can be divided into structured segment data so that unified processing can be performed on different data segments.

[0128] In some embodiments, data parsing can parse the original data set into structured data according to the requirements of the annotation task fields. This application mainly targets basic driving behaviors (changing lanes left or right, lane keeping, etc.); data synchronization can perform timestamp, coordinate system, and data unit synchronization. Timestamp synchronization is based on the positioning of the vehicle and adopts the minimum interval principle; the coordinate system is converted to the vehicle (Vehicle Coordinate System, VCS) coordinate system; data initial screening can eliminate data with mismatched fields, incomplete fields, and outliers; data slicing can be performed by sliding window slicing at fixed time intervals to output sample segments of fixed time length.

[0129] In some embodiments, data parsing can be parsed into structured data using different data fields according to the needs of the annotation task. This application mainly targets basic driving behaviors (going straight, changing lanes left or right); data synchronization performs timestamp, coordinate system, and data unit synchronization. Among them, timestamp synchronization is based on the positioning time of the vehicle (GPS / IMU), and the minimum interval principle is used for time synchronization. Coordinate system synchronization is uniformly converted to the vehicle (VCS) coordinate system; data initial screening eliminates data with mismatched fields, incomplete fields, and abnormal values. Among them, field mismatch usually means that there is a data type error, and incomplete fields are manifested as missing data such as vehicle position and vehicle speed in a single frame record.

[0130] Exemplarily, data slicing may be performed at fixed time intervals of 1 second with each segment length being 10 seconds and a single frame being 0.1 seconds, and a fixed time length sample segment may be output.

[0131] In some embodiments, data preprocessing requires adjustable sample ratios based on sample post-verification needs, and road scene types can be divided into straight roads, curves, diversions, and confluences.

[0132] S203 : Adaptively label the sample segments according to a dynamic threshold (equivalent to the “driving behavior decision threshold” in other embodiments).

[0133] In some embodiments, Figure 3 As shown, adaptive annotation of sample fragments according to dynamic thresholds can be achieved by Figure 3 The following steps are based on Figure 3 A method for adaptively labeling sample segments based on a dynamic threshold is described.

[0134] S301: Obtain a fragment frame sequence and a lane centerline set.

[0135] In some embodiments, it can be assumed that F = {f1, f2, f3, ..., f n} is a sample segment frame sequence, where each sample segment frame contains the vehicle position P i =(x i ,y i ); L = {l1,l2,l3,…,l n} is the set of lane centerlines in the map. Each lane consists of a series of point sets, and each lane has a unique identifier ID l .

[0136] S302, analyzing the vehicle position of the own vehicle (equivalent to the "current vehicle" in other embodiments) frame by frame.

[0137] In some embodiments, the vehicle position of the ego vehicle in each sample segment frame may be determined based on the position coordinates of the ego vehicle in the vehicle body coordinate system).

[0138] S303: Determine whether the lane centerline corresponding to the vehicle can be matched.

[0139] In some embodiments, the minimum projection distance can be matched according to the minimum projection distance. i , calculate the center line of each lane l i The minimum projection distance d ij =min p∈lj ||P i -p||2, select the lane centerline corresponding to the minimum distance as the best matching lane; if Then f i Matching Lane l i .

[0140] If yes, execute steps S304 to S306; otherwise, execute steps S307 to S308.

[0141] S304. Record the lane centerline.

[0142] In some embodiments, identification information (ID) of the successfully matched lane centerline may be recorded.

[0143] S305 . Convert the coordinates of the lane centerline into a Frenet coordinate system.

[0144] In some embodiments, the successfully matched lane centerline l i , which can be converted into a representation in the Frenet coordinate system.

[0145] S306: Calculate lane width and left and right boundary lines.

[0146] In some embodiments, if the lane boundary lines in the map are already associated, the Euclidean distance W between the left and right boundary lines is directly calculated according to formula (3): i Otherwise, the lane width threshold is dynamically calculated based on the road structure according to formula (4). The half-width W of the lane where the vehicle is located half =W i / 2, the left and right boundary lines of the vehicle pass through the center line and deviate to the left and right -W half and W half get.

[0147] For example, in a road diverging section, due to the existence of a triangular diversion area, the lane width will appear smaller when calculated at the fork. For example, at the beginning of the branch road merging into the main road, the lane width may be close to 3.5m, but as the vehicle approaches full merging into the main road, the lane width may be reduced to less than 3.5m, or even only 2 meters or narrower. At this time, the lane width needs to be dynamically calculated based on the width of the previous section so that it can accurately match the lane centerline.

[0148] S307: Determine whether the distance between the vehicle and the lane centerline exceeds a distance threshold.

[0149] If yes, execute the following step S308; otherwise, return to step S302 and continue execution.

[0150] S308. Record and analyze the anomaly.

[0151] In some embodiments, if any frame fails to match (i.e., all d ij >θ d , where θ d is a preset threshold), the segment is marked as abnormal and subsequent processing is terminated.

[0152] In some embodiments, step S304 to step S306 and step S307 to step S308 may be performed in parallel. After step S306 or step S308, the following step S309 may be performed.

[0153] S309: Calculate the projection distance from the vehicle position to the lane centerline to obtain a projection distance sequence.

[0154] The distance sequence between the vehicle position and the lane centerline projection between frames D = {d1, d2, d3, ..., d n}, where d i =||P i -Proj(P i ,l j )||2, Proj is the projection operation.

[0155] S410: Analyze whether the vehicle has crossed the line.

[0156] In some embodiments, the lane crossing behavior recognition can be determined based on the successive relationship between lane lines and the projection distance sequence between the vehicle position and the lane center line. If the successive relationship between lane lines in the sample segment changes: ID i,i+1 = 0, and the following crossing judgment criteria and crossing time limit are met at the same time, it is considered as crossing the line.

[0157] Cross-line judgment criteria: If there exists i, i+1 in the segment such that d i,li <W i,half And d i+1,li >W i,half And d i+1,l(i+1) <W i,half , considering that lane width mutation will lead to failure of crossing judgment, we further introduce time series difference, that is, analyze (d i,li -d i-1,l(i-1) )-(d i+1,l(i+1) -d i,li )>θ Δd , where θ Δd is the threshold value.

[0158] Cross-line duration limit: Set the maximum number of frames of cross-line duration to N max (equivalent to the "cross-line continuous frame number threshold" in other embodiments) If the number of cross-line frames exceeds N max After the frame, the line crossing behavior is excluded, N max Based on the fact that it can be dynamically adjusted through formula (5).

[0159] For example, the lane width W of the current frame is i = 3.5m, the distance from the vehicle position to the center line of the lane d i,li =1.68m; the centerline of the next matching lane changes and is no longer the subsequent lane, and the lane width remains W i+1 =3.5m, and the distance from the vehicle position in this frame to the center line of the lane in the previous frame is d i+1,li = 1.77m, from the vehicle position in this frame to the lane centerline in this frame d i+1,l(i+1)=1.73m; at the same time, the crossing line N is completed only between the previous and next frames max = 2, and there is no distance from the next frame's self-vehicle position to the lane centerline of this frame greater than 1.75m, and there is no distance decrease θ Δd >0, it can be determined that the vehicle has crossed the line in the current frame.

[0160] If so, execute the following steps S311 to S313; otherwise, execute the following steps S314 to S319. Steps S311 to S313 and steps S314 to S319 can be executed in parallel.

[0161] S311. Confirm the line crossing behavior and record the line crossing time and timestamp.

[0162] S312, confirm whether the crossing line meets the standard.

[0163] If yes, execute the following step S313; otherwise, return to step S309 and continue execution.

[0164] In some embodiments, if it is determined that the vehicle has crossed the line, the lane change criteria are identified according to the following lane change criteria:

[0165] Continuous cross-line frame confirmation: Based on the cross-line frame, identify continuous k frames (k <N max ) satisfies d i <W half , and the lane center lines of these frames are continuous, they are considered as lane change candidate frames.

[0166] Lane changing time (frame) constraint: To ensure that the lane changing behavior occurs within a reasonable time after crossing the line, the maximum lane changing time T is dynamically adjusted based on formula (6) change (equivalent to the "lane change duration threshold" in other embodiments).

[0167] After the line crossing frame, T is calculated according to the traffic characteristics of the road section. change = 10. And there are multiple consecutive frames k = 3 after the crossing frame, the distance between the vehicle position and the center line is less than 1.75m, and the lane successor relationship changes (for example: the unique identifiers of the center lines of the lanes following the crossing frame are 0x001, 0x002, 0x003, ..., and the lane change frames following the three consecutive frames in the subsequent 10 frames are different from the crossing frame, such as the identifier of the lane following the first frame is 0x010, 0x011, 0x012, ...).

[0168] It can be understood that this step mainly determines whether there are blank areas or overlapping areas between lane boundaries. On the one hand, due to the unequal width of the left and right lanes in the actual lanes, and on the other hand, due to the existence of sensor errors and calculation errors, the calculation of lane boundaries has certain noise. At this time, if the filtering algorithm is blindly used for averaging, the lane change frame judgment will not match the actual situation. This application adopts a dynamic lane change frame calculation range to reduce the impact of this part.

[0169] S413: Confirm the lane-changing behavior and record the lane-changing time and timestamp.

[0170] In some embodiments, if it is determined that there is a lane change behavior, the distance between the vehicle position in the lane change frame and the lane boundary line of the crossing frame is determined. If the distance is smaller, it is determined to change lanes to this boundary (for example, the projection distance d from the left boundary line of the lane center line of the crossing frame lane is greater than the projection distance d from the left boundary line of the lane center line of the crossing frame lane). left_proj = 0.2m, distance d from the projection of the right boundary line right_proj =3.7m, then it is determined to change lanes to the left).

[0171] S314: Analyze vehicle status data.

[0172] In some embodiments, the vehicle status data includes speed difference and distance difference.

[0173] S315: Determine whether the distance from the vehicle to the center line of the lane is less than half the width of the lane.

[0174] If so, execute the following step S316; otherwise, return to step S308 and continue execution.

[0175] S316: Determine whether the vehicle is executing or following the vehicle based on whether there is a target vehicle ahead.

[0176] In some embodiments, if it is determined that there is no crossing of the line, the following criteria are used to determine whether to go straight or follow the vehicle:

[0177] Straight judgment: All frames satisfy d i <W half , and the lane centerline has a continuous ID i,i+1 , and the vehicle state changes significantly (such as speed change Δv>θ v , position change Δx>θ x ).

[0178] Following vehicle determination: Based on the straight-ahead determination, if there is a target vehicle ahead and the relative position and speed difference between the two vehicles meets the following vehicle standard (such as the distance difference D rel <θ D , speed difference Δv rel <θv rel ), it is deemed as following the vehicle.

[0179] In some embodiments, if the distances between the lane centerline of all frames in the segment are less than half the lane width (e.g., the distances in 100 frames are such as: 1.68m, 1.64m, 1.71m, ..., lane width 3.5m), and the lane centerlines of all frames have a successive relationship, and the change in the vehicle speed or position in the vehicle status information is not less than a certain range (e.g., the speed of each frame is greater than 10m / s), then it is determined that the vehicle is moving straight.

[0180] S204: Screen abnormal samples and perform post-validation on the dynamic threshold.

[0181] In some embodiments, abnormal samples generally identify situations that are not consistent with driving habits, and such samples are generally still beneficial for training. Figure 4 The abnormal samples 400 are common abnormal sample types in this application, including slow driving 401, driving away from the center line 402, crossing the line for a long time 403, changing lanes without turning on the lights 404, speeding 405, etc.

[0182] In some embodiments, according to the trajectory in the sample segment and the vehicle state information, if there is a d i Much larger than W half , the position difference between the two exceeds the lane width and there is no lane centerline within a radius of 2m around the vehicle, that is, the vehicle position is far away from the lane centerline and is identified as an abnormal sample.

[0183] In some embodiments, based on the vehicle trajectory and lane centerline information in the sample segment, if the distance between the vehicle position and the lane centerline in the entire segment sample is less than a certain range (e.g., 0.1 m) for more than half of the frames (e.g., 100 frames in the sample segment, more than 50 frames), that is, the vehicle's driving trajectory has long-term lane-crossing behavior, and it is identified as an abnormal sample.

[0184] In some embodiments, based on the vehicle status information in the segment sample, if the vehicle's driving speed in the entire segment sample (for example, the vehicle's speed is divided into x and y direction speeds according to a unified coordinate system) is less than a certain range (for example, 0.2 m / s) for more than half of the frames in the x and y directions, that is, the vehicle is slowing down, it is identified as an abnormal sample.

[0185] In some embodiments, based on the vehicle status information in the fragment sample, if it is confirmed as a lane changing sample in S300, no lighting signal appears in the frame from the crossing line frame to the frame with the maximum lane changing duration (generally, if the vehicle has lights on, the flag of the signal will jump, such as the bus_turn_left_light signal jumping from 0 to 1), that is, the lane changing behavior without lighting is identified as an abnormal sample.

[0186] It is understandable that the present application includes such decision-making behaviors into abnormal samples to ensure sample accuracy while further refining classification and improving the effectiveness of labeling.

[0187] In the embodiment of the present application, the introduction of the annotation library is intended to optimize and adjust the threshold in S203, such as Figure 5 As shown in the figure, it is a flow chart for introducing the standard library for post-verification. Figure 5 Take as an example to illustrate the method of dynamic threshold posterior positive verification.

[0188] S501: Set an initial threshold.

[0189] The initial threshold may be the maximum line crossing duration frame number, the maximum lane change duration, etc.

[0190] S502: Identify abnormal samples.

[0191] S503: Use the original sample library and the manually annotated sample library for parallel verification.

[0192] A manually annotated standard sample library containing various decision-making behaviors is constructed. Samples of various decisions (such as going straight, changing lanes left or right) are evenly distributed in a ratio based on general driving conditions (such as 1:1:3). The total number N is 2000, which is convenient for rapid iterative optimization.

[0193] S504: Determine whether the marking accuracy meets the standard.

[0194] If so, execute the following steps S505 to S507; otherwise, execute step S508.

[0195] S505. Analyze the proportion of abnormal samples.

[0196] S506. Confirm whether the proportion of abnormal samples is too high.

[0197] If yes, execute step S507; otherwise, execute step S508.

[0198] S507, optimizing the initial threshold using the gradient descent method.

[0199] The processing logic of S203 is applied to both the standard sample library and the original data set, and the feedback information of the standard sample library is used to fine-tune the parameters according to the gradient descent method. For example, the standard sample indicates that at the initial threshold (such as N max =10, T change =30) is 82% in the sampling inspection accuracy, which rises to 90% after the threshold adjustment, indicating that the adjustment direction is consistent with the gradient direction, otherwise the threshold needs to be adjusted in the opposite direction.

[0200] When adjusting parameters, consider both sample diversity and accuracy to avoid losing sample diversity due to excessive pursuit of accuracy, or vice versa. For example, dynamically adjust parameters based on the proportion of abnormal samples. If there are too many abnormal samples (such as more than 10%), moderately relax certain thresholds to retain more samples.

[0201] S508. Review regularly and update the manual standard sample library.

[0202] The standard sample library is reviewed regularly to ensure that the balance of sample distribution still conforms to the actual driving conditions over time. If necessary, the sample library is updated and the process of step S204 is repeated.

[0203] It can be understood that the self-labeling method for autonomous driving decision data provided in this application adopts adjustable dynamic thresholds and unified management of abnormal samples, which can adapt to complex road conditions and thus improve the accuracy and diversity of driving behavior decisions.

[0204] The driving decision data labeling method provided in the present application obtains the driving behavior decision data of the current vehicle, and the driving behavior decision data includes the driving environment data of the current vehicle; the driving behavior decision data is labeled based on the driving behavior decision threshold to obtain the labeling information, and the labeling information is used to identify the driving behavior corresponding to the current vehicle. In this way, since the driving behavior decision threshold can be dynamically adjusted based on the driving environment data, the driving behavior decision threshold can adapt to the changes in the driving environment of the current vehicle. Therefore, the driving behavior decision data is labeled based on the driving behavior decision threshold, which can improve the accuracy of the labeling. Furthermore, the labeled driving decision data can be used as training data to ensure the training accuracy of the driving behavior decision model.

[0205] The present application embodiment provides a driving decision data labeling device, Figure 6 A schematic diagram of the structure of a driving decision data labeling device provided in an embodiment of the present application is shown in FIG. Figure 6 As shown, the driving decision data labeling device 600 includes:

[0206] The first acquisition module 601 is used to acquire driving behavior decision data of the current vehicle; the driving behavior decision data includes the driving environment data of the current vehicle;

[0207] The first labeling module 602 is used to label the driving behavior decision data based on a driving behavior decision threshold to obtain labeling information; wherein the driving behavior decision threshold is dynamically adjusted based on the driving environment data, and the labeling information is used to identify the driving behavior corresponding to the current vehicle.

[0208] In some embodiments, the driving behavior decision data includes a plurality of image frames; the first annotation module 602 includes:

[0209] A first determination submodule, configured to determine a first lane centerline of a lane corresponding to the current vehicle in each image frame according to the position information of the current vehicle and the position information of each lane centerline in each image frame;

[0210] A second determination submodule, configured to determine a lane width of a lane corresponding to the current vehicle in each image frame according to the position information of the center line of the first lane and the position information of the current vehicle in each image frame;

[0211] The first labeling submodule is used to label the driving decision data based on the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width and the driving behavior decision threshold to obtain labeling information.

[0212] In some embodiments, the driving behavior decision threshold includes a threshold for the number of frames of crossing the line, and the annotation information includes information on whether the current vehicle crosses the line; the first annotation submodule includes:

[0213] a first determining unit, configured to determine whether the next image frame satisfies a vehicle crossing condition based on position information of the current vehicle in the current image frame, position information of the first lane centerline in the current image frame, position information of the first lane centerline in the next image frame, and the lane width if the first lane centerline in the current image frame and the first lane centerline in the next image frame are different;

[0214] A second determining unit is used to determine the number of consecutive image frames that meet the vehicle crossing line condition;

[0215] The first marking unit is used to mark the continuous image frames according to the frame number and the line crossing continuous frame number threshold to obtain information on whether the current vehicle crosses the line.

[0216] In some embodiments, the driving environment data includes the number of lanes and road state data in the image frame; the driving decision data labeling device 600 also includes:

[0217] A second acquisition module is used to acquire the number of lanes in a first reference image frame; the first reference image frame is any one of the multiple continuous image frames;

[0218] The first determination module is used to determine the line crossing duration frame number threshold based on the road state data of the lane corresponding to the current vehicle in the first reference image frame, the number of lanes in the first reference image frame, and the driving speed of the current vehicle within a preset time period before the corresponding moment of the first reference image frame.

[0219] In some embodiments, the driving behavior decision threshold includes a lane change duration threshold, and the annotation information includes information on whether the current vehicle changes lanes; the first annotation submodule also includes:

[0220] A first acquisition submodule is used to acquire a plurality of continuous crossing-line image frames in the plurality of image frames, in which information marked is that the current vehicle has crossed the line;

[0221] A third determination submodule is used to determine the distance between the current vehicle and the corresponding first lane center line in the crossing-line image frame according to the position information of the current vehicle in the crossing-line image frame and the position information of the corresponding first lane center line in the crossing-line image frame;

[0222] a fourth determination submodule, configured to determine the line-crossing image frame as a lane-changing candidate image frame if the distance and the lane width satisfy a first condition;

[0223] The second marking submodule is used to mark the candidate lane-changing image frame according to the time information of the lane-changing candidate image frame and the lane-changing duration threshold, so as to obtain information on whether the current vehicle changes lanes.

[0224] In some embodiments, the driving environment data includes the traffic volume in the image frame; the driving decision data labeling device 600 further includes:

[0225] A third acquisition module is used to acquire the traffic volume of vehicles in a second reference image frame and the distance information between the current vehicle and the first vehicle in the second reference image frame; the second reference image frame is any one of the lane change candidate image frames; the first vehicle may be another vehicle adjacent to the current vehicle;

[0226] The second determination module is used to determine the lane change duration threshold according to the vehicle flow and the spacing information in the second reference image frame.

[0227] In some embodiments, the second determining submodule includes:

[0228] a third determining unit, configured to determine an initial lane width of a lane corresponding to the current vehicle in the third reference image frame according to the position information of the current vehicle in the third reference image frame and the position information of a center line of a first lane corresponding to the current vehicle in the third reference image frame; the third reference image frame being any one of the multiple image frames;

[0229] a first acquisition unit, configured to acquire lane width information in a preset number of image frames before the third reference image frame if an absolute value of a difference between the initial lane width and the lane width of a fourth reference image frame is greater than a first preset threshold; the fourth reference image frame is an image frame next to the third reference image frame;

[0230] A fourth determining unit is used to determine a target lane width of a lane corresponding to the current vehicle in the fourth reference image frame based on the initial lane width and the lane width information.

[0231] In some embodiments, the driving decision data labeling device 600 further includes:

[0232] A fourth acquisition module is used to acquire a manually annotated driving behavior decision data set; the driving behavior decision data set includes driving behavior decision samples and corresponding real annotation information;

[0233] A second labeling module is used to label the driving behavior decision sample based on the driving behavior decision threshold to obtain predicted labeling information;

[0234] A third determination module, configured to determine a labeling accuracy rate for the driving behavior decision sample according to the real labeling information and the predicted labeling information;

[0235] The first updating module is used to update the driving behavior decision threshold if the labeling accuracy is less than a second preset threshold, obtain a new driving behavior decision threshold, and label the driving behavior decision data based on the new driving behavior decision threshold.

[0236] In some embodiments, the driving decision data labeling device 600 further includes:

[0237] A fifth acquisition module is used to acquire reference driving behavior decision data of which the annotation information indicates that the driving behavior of the current vehicle is abnormal; the abnormal driving behavior at least includes that the current vehicle changes lanes without turning on the lights and moves away from the center line of the lane;

[0238] The first creation module is used to establish an abnormal driving behavior data set based on the reference driving behavior decision data and the annotation information corresponding to the reference driving behavior decision data, so as to train a driving decision model based on the abnormal driving behavior data set.

[0239] It should be noted that the description of the driving decision data labeling device in the embodiment of the present application is similar to the description of the corresponding method embodiment above, and has similar beneficial effects as the method embodiment, so it is not repeated here. For technical details not disclosed in this embodiment, please refer to the description of the method embodiment of the present application for understanding.

[0240] It should be noted that in the embodiment of the present application, if the above-mentioned driving decision data labeling method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can be essentially or partly reflected in the form of a software product that contributes to the relevant solution. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods of each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk. In this way, the embodiment of the present application is not limited to any specific combination of hardware and software.

[0241] Accordingly, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the driving decision data labeling method provided in the above embodiment is implemented.

[0242] Figure 7 A schematic diagram of the structure of a driving decision data labeling device provided in an embodiment of the present application, such as Figure 7 As shown, the driving decision data annotation device 700 includes: a memory 701, a processor 702, a communication interface 703 and a communication bus 704. The memory 701 is used to store instructions for annotation of executable driving decision data; the processor 702 is used to execute the annotation instructions for executable driving decision data stored in the memory 701 to implement the driving decision data annotation method provided in the above embodiment.

[0243] The description of the above driving decision data labeling device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the driving decision data labeling device embodiment of this application, please refer to the description of the method embodiment of this application for understanding.

[0244] It should be noted that, in this article, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises at least one ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0245] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0246] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0247] In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0248] A person of ordinary skill in the art can understand that: all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, disks or optical disks.

[0249] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can essentially or in other words, the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a product to execute all or part of the methods of each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0250] The above are only implementation methods of the present application, but the protection scope of the present application is not limited thereto. Any technical object familiar with the technical field can be easily thought of within the technical scope disclosed in the present application. Changes or substitutions should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for labeling driving decision data, characterized in that: include: Obtain driving behavior decision data of the current vehicle; The driving behavior decision data includes the driving environment data of the current vehicle; The driving behavior decision data is labeled based on a driving behavior decision threshold to obtain labeling information; wherein the driving behavior decision threshold is dynamically adjusted based on the driving environment data, and the labeling information is used to identify the driving behavior corresponding to the current vehicle.

2. The method according to claim 1, characterized in that The driving behavior decision data includes a plurality of image frames; the step of labeling the driving decision data based on the driving behavior decision threshold to obtain labeling information includes: Determine, according to the position information of the current vehicle and the position information of the center lines of each lane in each image frame, the first lane center line of the lane corresponding to the current vehicle in each image frame; Determining a lane width of a lane corresponding to the current vehicle in each of the image frames according to the position information of the center line of the first lane and the position information of the current vehicle in each of the image frames; Based on the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width and the driving behavior decision threshold, the driving decision data is labeled to obtain labeling information.

3. The method according to claim 2, characterized in that The driving behavior decision threshold includes a threshold for the number of frames of continuous crossing the line, and the annotation information includes information on whether the current vehicle crosses the line; The step of labeling the driving decision data based on the position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width, and the driving behavior decision threshold to obtain the labeling information includes: If the first lane centerline in the current image frame is different from the first lane centerline in the next image frame, determining whether the next image frame satisfies the vehicle crossing condition according to the position information of the current vehicle in the current image frame, the position information of the first lane centerline in the current image frame, the position information of the first lane centerline in the next image frame, and the lane width; Determining the number of consecutive image frames that meet the vehicle crossing line condition; The continuous image frames are marked according to the frame number and the line crossing continuous frame number threshold to obtain information on whether the current vehicle has crossed the line.

4. The method according to claim 3, characterized in that The driving environment data includes the number of lanes and road status data in the image frame; the method further includes: Acquire the number of lanes in a first reference image frame; the first reference image frame is any one of the multiple continuous image frames; The line crossing duration frame number threshold is determined based on the road state data of the lane corresponding to the current vehicle in the first reference image frame, the number of lanes in the first reference image frame, and the driving speed of the current vehicle within a preset time period before the corresponding moment of the first reference image frame.

5. The method according to claim 2, characterized in that: The driving behavior decision threshold includes a lane change duration threshold, and the annotation information includes information on whether the current vehicle changes lanes; The position information of the current vehicle in each image frame, the position information of the center line of the first lane, the lane width and the driving behavior decision threshold, and the driving decision data are annotated to obtain the annotated information, including: Acquire a plurality of continuous crossing-line image frames in which the label information indicates that the current vehicle has crossed the line; Determine, according to the position information of the current vehicle in the crossing-line image frame and the position information of the corresponding first lane center line in the crossing-line image frame, the distance between the current vehicle and the corresponding first lane center line in the crossing-line image frame; If the distance and the lane width satisfy a first condition, determining the line crossing image frame as a lane changing candidate image frame; The candidate lane-changing image frames are marked according to the time information of the lane-changing candidate image frames and the lane-changing duration threshold, so as to obtain information on whether the current vehicle is changing lanes.

6. The method according to claim 5, characterized in that The driving environment data includes the traffic volume in the image frame; the method further includes: Acquire the traffic volume of vehicles in a second reference image frame and the distance information between the current vehicle and the first vehicle in the second reference image frame; the second reference image frame is any one of the lane change candidate image frames; the first vehicle is another vehicle adjacent to the current vehicle; The lane change duration threshold is determined according to the vehicle flow and the distance information in the second reference image frame.

7. The method according to any one of claims 2 to 6, characterized in that Determining a lane width of a lane corresponding to the current vehicle in each image frame according to the position information of the center line of the first lane and the position information of the current vehicle in each image frame includes: Determining an initial lane width of a lane corresponding to the current vehicle in the third reference image frame according to the position information of the current vehicle in the third reference image frame and the position information of a center line of a first lane corresponding to the current vehicle in the third reference image frame; the third reference image frame is any one of the multiple image frames; If the absolute value of the difference between the initial lane width and the lane width of the fourth reference image frame is greater than a first preset threshold, obtaining lane width information in a preset number of image frames before the third reference image frame; the fourth reference image frame is an image frame next to the third reference image frame; Based on the initial lane width and the lane width information, a target lane width of a lane corresponding to the current vehicle in the fourth reference image frame is determined.

8. The method according to any one of claims 1 to 6, characterized in that The method further comprises: Obtaining a manually annotated driving behavior decision data set; the driving behavior decision data set includes driving behavior decision samples and corresponding real annotated information; Annotating the driving behavior decision sample based on the driving behavior decision threshold to obtain predicted annotation information; Determining a labeling accuracy rate for the driving behavior decision sample according to the real labeling information and the predicted labeling information; If the labeling accuracy is less than a second preset threshold, the driving behavior decision threshold is updated to obtain a new driving behavior decision threshold, so as to label the driving behavior decision data based on the new driving behavior decision threshold.

9. The method according to any one of claims 1 to 6, characterized in that The method further comprises: Acquire reference driving behavior decision data of which the annotation information indicates that the driving behavior of the current vehicle is abnormal; the abnormal driving behavior at least includes that the current vehicle changes lanes without turning on the lights and moves away from the center line of the lane; Based on the reference driving behavior decision data and the annotation information corresponding to the reference driving behavior decision data, an abnormal driving behavior data set is established to train a driving decision model based on the abnormal driving behavior data set.

10. A driving decision data labeling device, characterized in that: include: A first acquisition module is used to acquire driving behavior decision data of the current vehicle; The driving behavior decision data includes the driving environment data of the current vehicle; The first labeling module is used to label the driving behavior decision data based on a driving behavior decision threshold to obtain labeling information; wherein the driving behavior decision threshold is dynamically adjusted based on the driving environment data, and the labeling information is used to identify the driving behavior corresponding to the current vehicle.

11. A driving decision data labeling device, characterized in that: include: processor and memory; wherein, The memory is used to store a computer program that can be run on the processor; The processor is configured to execute the method according to any one of claims 1 to 9 when running the computer program.

12. A computer-readable storage medium, characterized in that: The storage medium stores computer program code, and when the computer program code is executed by a computer, the method according to any one of claims 1 to 9 is executed.

Images