Data screening method and device, vehicle, server and storage medium

Abstract

The invention discloses a data screening method and device, a vehicle, a server and a storage medium. The data screening method comprises the steps that scene data, collected by a sensor, of a driving road where a vehicle is located are acquired; performing first screening processing on the scene data according to a preset screening strategy; performing second screening processing on the scene data after the first screening processing according to a scene prediction model, and determining target scene data which carries a scene category label; sending the target scene data to a server; and storing the updated scene prediction model sent by the server. According to the method, the data scene screening performance can be stably improved, the more extreme scene data are screened, the better the performance is, meanwhile, the time needed by the data screening process can be shortened, and therefore the screening cost is reduced while the data screening efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of vehicle technology, and in particular to a data screening method, device, vehicle, server and storage medium. Background Art

[0002] Autonomous driving relies on data collected from sensors and lidars, as well as algorithm training, analysis, and processing of various types of data such as vehicle mechanics and the environment. Existing vehicles are equipped with sensors such as cameras, millimeter-wave radars, and lidars for road testing in various places, on various road conditions, and under various climatic conditions. On average, a vehicle in motion generates 1TB of data per hour, which can generate massive amounts of data up to PB for a multi-vehicle autonomous driving fleet. However, among the huge amount of data, only a very small amount of data from extreme scenarios (snowy days, special roads, special vehicles, etc.) is conducive to improving the performance of autonomous driving. Therefore, screening out extreme scenario data that is truly conducive to the iteration of autonomous driving algorithms can effectively improve scene recognition capabilities while avoiding negative effects such as low scene recognition efficiency caused by data redundancy.

[0003] In the related technology, fine-grained screening is performed by manual labelers or automatic screening based on preset unified rules. This method has obvious problems: first, the cycle required for data collection, transmission, labeling, and screening is long, and the data timeliness is poor, especially in the case of massive data; second, the transmission and storage costs of a large amount of unlabeled collected data, as well as the labor costs are high. Summary of the invention

[0004] In view of this, the present application provides a data screening method, device, storage medium and vehicle, which continuously improve the data scene screening performance by combining rule screening and neural network learning, effectively reduce the amount of data transmitted and stored, and reduce the cost of driving scene recognition while improving scene screening efficiency.

[0005] According to a first aspect of the present application, a data screening method is provided, comprising:

[0006] Obtaining scene data of the road on which the vehicle is traveling collected by sensors;

[0007] Performing a first screening process on the scene data according to a preset screening strategy;

[0008] Performing a second screening process on the scene data after the first screening process according to the scene prediction model to determine target scene data, where the target scene data carries a scene category label;

[0009] Sending the target scene data to the server so that the server updates the scene prediction model according to the target scene data;

[0010] The updated scene prediction model sent by the storage server.

[0011] Optionally, performing a second screening process on the scene data after the first screening process according to the scene prediction model to determine the target scene data includes:

[0012] Inputting the scene data after the first screening process into the first scene prediction model to obtain the first scene category and the confidence of the first scene category;

[0013] Inputting the scene data after the first screening process into a second scene prediction model to obtain a second scene category and a confidence level of the second scene category, wherein the parameter amount of the second scene prediction model is less than the parameter amount of the first scene prediction model;

[0014] Using a distillation loss function to process the confidence of the first scene category and the confidence of the second scene category that is the same as the first scene category, to obtain a first probability and a second probability of the scene data;

[0015] Determine the scene data whose difference between the first probability and the second probability is greater than a preset difference as the target scene data;

[0016] A scene category label of the target scene data is generated according to the first scene category or the second scene category.

[0017] Optionally, the distillation loss function adopts the following formula:

[0018]

[0019] Where q represents probability, z represents confidence, T represents distillation temperature coefficient, and i represents a scene category in j.

[0020] Optionally, performing a first screening process on the scene data according to a preset screening strategy includes:

[0021] Delete scene data whose average brightness value exceeds the preset brightness range.

[0022] Optionally, performing a first screening process on the scene data according to a preset screening strategy includes: deleting scene data with a definition less than a preset definition.

[0023] Optionally, performing a first screening process on the scene data according to a preset screening strategy includes:

[0024] If the similarity between any two scene data is greater than or equal to the preset similarity, one of the two scene data is deleted.

[0025] Optionally, the data screening method further includes:

[0026] The scene data is predicted and processed according to the updated scene prediction model to determine the scene information of the road on which the vehicle is traveling.

[0027] According to a second aspect of the present application, a data screening method is provided, comprising:

[0028] Receive target scene data sent by the vehicle;

[0029] Preprocess the target scene data;

[0030] The scene prediction model is trained according to the preprocessed target scene data to update the scene prediction model;

[0031] Send the updated scene prediction model to the vehicle.

[0032] According to a third aspect of the present application, a data screening device is provided, comprising:

[0033] An acquisition module, used to acquire scene data of the road on which the vehicle is traveling collected by a sensor;

[0034] A screening module, used to perform a first screening process on the scene data according to a preset screening strategy; and

[0035] Performing a second screening process on the scene data after the first screening process according to the scene prediction model to determine target scene data, where the target scene data carries a scene category label;

[0036] A communication module, used to send the target scene data to the server so that the server updates the scene prediction model according to the target scene data;

[0037] The storage module is used to store the updated scene prediction model sent by the server.

[0038] Optionally, a screening module is specifically used to input the scene data after the first screening process into the first scene prediction model to obtain the first scene category and the confidence of the first scene category;

[0039] Inputting the scene data after the first screening process into a second scene prediction model to obtain a second scene category and a confidence level of the second scene category, wherein the parameter amount of the second scene prediction model is less than the parameter amount of the first scene prediction model;

[0040] Using a distillation loss function to process the confidence of the first scene category and the confidence of the second scene category that is the same as the first scene category, to obtain a first probability and a second probability of the scene data;

[0041] Determine the scene data whose difference between the first probability and the second probability is greater than a preset difference as the target scene data;

[0042] A scene category label of the target scene data is generated according to the first scene category or the second scene category.

[0043] Optionally, the distillation loss function adopts the following formula:

[0044]

[0045] Where q represents probability, z represents confidence, T represents distillation temperature coefficient, and i represents a scene category in j.

[0046] Optionally, the screening module is specifically used to delete scene data whose average brightness value exceeds a preset brightness range.

[0047] Optionally, the screening module is specifically used to delete scene data with a clarity less than a preset clarity.

[0048] Optionally, the screening module is specifically configured to delete one of the two scene data if the similarity between the two scene data is greater than or equal to a preset similarity.

[0049] Optionally, the data screening device also includes: a prediction module, which is used to perform prediction processing on the scene data according to the updated scene prediction model to determine the scene information of the road on which the vehicle is traveling.

[0050] According to a fourth aspect of the present application, a data screening device is provided, comprising:

[0051] A communication module, used for receiving target scene data sent by the vehicle;

[0052] A preprocessing module, used for preprocessing target scene data;

[0053] A model training module, used to train the scene prediction model according to the preprocessed target scene data to update the scene prediction model;

[0054] The communication module is also used to send the updated scene prediction model to the vehicle.

[0055] According to the fifth aspect of the present application, a vehicle is provided, including a sensor and a processor, wherein the processor and a memory are coupled, and the memory stores a program or instruction running on the processor, and when the program or instruction is executed by the processor, the steps of the data screening method of the first aspect mentioned above are implemented.

[0056] According to the sixth aspect of the present application, a server is provided, including a processor, wherein the processor is coupled to a memory, and the memory stores a program or instruction running on the processor, and when the program or instruction is executed by the processor, the steps of the data screening method of the second aspect are implemented.

[0057] According to the seventh aspect of the present application, a computer-readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above-mentioned data screening method are implemented.

[0058] According to an eighth aspect of the present application, there is provided a data screening system, comprising the vehicle of the fifth aspect and the server of the sixth aspect, wherein the vehicle is communicatively connected to the server.

[0059] By means of the above technical solution, after obtaining the scene data, the vehicle first uses a pre-set preset screening strategy to eliminate low-quality data, and then uses the trained scene prediction model for secondary screening to select target scene data with higher scene recognition value, so as to use the target scene data with scene category labels as training samples for the scene prediction model. On the one hand, while realizing automatic screening, the data scene screening performance is continuously improved by combining rule screening and neural network learning, effectively reducing the amount of data transmitted and stored, shortening the long-cycle data screening process to near real-time, improving the scene data screening efficiency, and accurately generating the scene category labels of the target scene data through the scene prediction model, eliminating manual labeling and reducing the cost of manual screening. On the other hand, the data screening process is completed by the vehicle-side processor, effectively avoiding server overload due to excessive data volume. On the other hand, continuously updating the scene prediction model through high-quality target scene data helps to improve the prediction accuracy of the scene prediction model and provide reliable and high-precision data support for autonomous driving.

[0060] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0062] Figure 1 One of the flow charts of the data screening method provided in the embodiment of the present application is shown;

[0063] Figure 2 The second flowchart of the data screening method provided in the embodiment of the present application is shown;

[0064] Figure 3 The third flowchart of the data screening method provided in the embodiment of the present application is shown;

[0065] Figure 4 One of the structural schematic diagrams of the data screening device provided in the embodiment of the present application is shown;

[0066] Figure 5 A second structural schematic diagram of the data screening device provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0067] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0068] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0069] Those skilled in the art will appreciate that, unless expressly stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, or there may be intermediate elements. In addition, the "connection" or "connection" used herein may include wireless connection or wireless fusion. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0070] Now, exemplary embodiments according to the present application will be described in more detail with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in a variety of different forms and should not be interpreted as being limited to the embodiments set forth herein. It should be understood that these embodiments are provided to make the disclosure of the present application thorough and complete, and to fully convey the concepts of these exemplary embodiments to those of ordinary skill in the art.

[0071] In this embodiment, a data screening method is provided, which is applicable to vehicles, such as Figure 1 As shown, the method includes:

[0072] Step 110, acquiring scene data of the road on which the vehicle is traveling collected by a sensor;

[0073] The sensor may be a camera, a millimeter wave radar, a laser radar, etc. The sensor may not be located in front of, on the sides and / or behind the vehicle, as long as it can collect scene data of a sufficient size of the driving road. The scene data also includes image data collected by a camera and / or point cloud data collected by a radar.

[0074] Step 120, performing a first screening process on the scene data according to a preset screening strategy;

[0075] It is understandable that the preset filtering strategy is used to remove dirty data and can be set according to actual needs.

[0076] In a specific application scenario, step 120, that is, performing a first screening process on the scene data according to a preset screening strategy, includes the following methods:

[0077] Method 1: Delete scene data whose average brightness value exceeds the preset brightness range.

[0078] In this embodiment, the brightness value of each pixel in the image data or the brightness value of each point in the point cloud data is detected, and the average brightness value of the scene data is calculated based on this. The average brightness value is compared with the preset brightness range. If the average brightness value exceeds the preset brightness range, it means that the brightness of the scene data is too high or too low, which is not conducive to the model's recognition and prediction of the scene. At this time, the scene data whose average brightness value exceeds the preset brightness range is deleted. Thus, the scene data with poor brightness is eliminated, which not only reduces the amount of data, but also helps to improve the prediction accuracy of the scene information.

[0079] Method 2: Delete scene data whose definition is less than the preset definition.

[0080] In this embodiment, the clarity of the image scene data is detected by a clarity detection algorithm and compared with a preset clarity. If the comparison result is that the clarity is less than the preset clarity, it means that the clarity of the scene data is not enough to distinguish the scene information where the vehicle is currently located, and the scene data with a clarity less than the preset clarity is deleted, and only the scene data with a clarity greater than or equal to the preset clarity is retained. In this way, the scene data with insufficient clarity can be eliminated, ensuring the quality of the available scene data.

[0081] Method three: if the similarity between any two scene data is greater than or equal to a preset similarity, one of the two scene data is deleted.

[0082] In this embodiment, when comparing any two scene data in a large amount of scene data, if the similarity is greater than or equal to the preset similarity, it means that the scene information or other data predicted based on the two scene data may be the same. In order to avoid repeated prediction work, one of the two scene data is deleted, that is, only one of the two similar scene data is retained. In this way, the amount of data is effectively reduced while ensuring the quality of the scene data, more storage space is freed up, and the workload of subsequent screening, scene prediction, and model training can be reduced, greatly reducing the cost of data screening.

[0083] For example, a hash algorithm is used to calculate similarity. The hash algorithm can convert each image in the image data into a string of 0-1 codes to perform similarity calculation.

[0084] It should be noted that the preset brightness range, preset clarity and preset similarity can be reasonably set according to the screening accuracy, and the embodiments of the present application do not make specific limitations.

[0085] Step 130, performing a second screening process on the scene data after the first screening process according to the scene prediction model to determine target scene data, where the target scene data carries a scene category label;

[0086] In this embodiment, after obtaining the scene data, the low-quality data is first eliminated using a pre-set preset screening strategy, and then each scene data is classified through the scene prediction model, so that the value of each scene data can be distinguished, and a choice can be made to retain or delete it, so as to select the target scene data with higher scene recognition value, so as to facilitate the use of the target scene data with scene category labels as training samples for the scene prediction model. Thus, while achieving automatic screening, the data scene screening performance is continuously improved by combining rule screening and neural network learning, effectively reducing the amount of data transmitted and stored, shortening the long-cycle data screening process to near real-time, improving the scene data screening efficiency, and accurately generating the scene category label of the target scene data through the scene prediction model, eliminating manual labeling and reducing the cost of manual screening. Moreover, the data screening process is completed by the vehicle-side processor, which effectively avoids server overload due to excessive data volume.

[0087] It can be understood that the order of using a preset filtering strategy to perform the first filtering process and performing the second filtering process according to the scene prediction model is not specifically limited in the embodiment of the present application. In the scheme of this embodiment, the scene data can also be first subjected to a second filtering process according to the scene prediction model, and then the scene data after the second filtering process can be subjected to a first filtering process using a preset filtering strategy.

[0088] Specifically, the scene prediction model can adopt an open source neural network model, such as resnet-50.

[0089] Furthermore, if Figure 2 As shown, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, step 130, that is, performing a second screening process on the scene data after the first screening process according to the scene prediction model to determine the target scene data, specifically includes the following steps:

[0090] Step 131, inputting the scene data after the first screening process into a first scene prediction model to obtain a first scene category and a confidence level of the first scene category;

[0091] Step 132, inputting the scene data after the first screening process into a second scene prediction model to obtain a second scene category and a confidence level of the second scene category;

[0092] Among them, one of the first scene prediction model and the second scene prediction model is a pre-stored scene prediction model, and the other can be obtained by enlarging or reducing the scene prediction model. The first scene prediction model and the second scene prediction model are both used to predict the scene information of the demand scene. For example, the demand scene can be snowy, special roads, special vehicles, etc. The confidence is the unnormalized probability (logits) of the scene category. The parameter amount of the second scene prediction model is less than the parameter amount of the first scene prediction model. For example, the parameter amount of the first scene prediction model is 1.25 times, 1.5 times or 2 times that of the second scene prediction model.

[0093] It should be noted that the first scene category or the second scene category may include multiple possible scenes, and the first scene category obtained by the first scene prediction model and the second scene category obtained by the second scene prediction model may be the same or different.

[0094] In this embodiment, a scene prediction model is pre-trained, and the scene prediction model is enlarged or reduced to obtain a first scene prediction model and a second scene prediction model with the same function but different parameter quantities. The first scene prediction model and the second scene prediction model are used to predict the possible scene categories of the scene data and the confidence corresponding to the scene categories. In order to use the confidence of the first scene category obtained by the first scene prediction model and the confidence of the second scene category obtained by the second scene prediction model to analyze the scene data that can produce a large difference in prediction results based on two models with different parameter quantities, it helps to ensure that the target scene data has a higher training value, thereby improving the data scene screening performance.

[0095] Step 133, using a distillation loss function to process the confidence of the first scene category and the confidence of the second scene category that is the same as the first scene category, to obtain a first probability and a second probability of the scene data;

[0096] Specifically, the distillation loss function uses the following formula:

[0097]

[0098] In the formula, q represents probability, z represents confidence, T represents distillation temperature coefficient, i represents a scene category in j, and z i represents the confidence of the i-th scene category, z j Represents the confidence of all scene categories.

[0099] It is worth mentioning that the distillation temperature coefficient T is used to adjust the smoothness of the output results of the scene prediction model, which is related to the size of the parameters of the scene prediction model. When T is low, less attention is paid to negative labels (extreme scenes) during model training, especially those negative labels that are significantly lower than the average value, while when T is high, the values ​​related to negative labels will increase relatively. Therefore, the larger the number of parameters, the larger the value of the distillation temperature coefficient T can be. When the distillation temperature coefficient T = 1, the distillation loss function is the standard loss function (Softmax) formula. The higher T is, the smoother the output probability distribution of the loss function is, the greater the entropy of its distribution, the information carried by the negative labels will be relatively amplified, and the model training will pay more attention to negative labels.

[0100] For example, after inputting the scene data into the first scene prediction model and the second scene prediction model respectively, the first scene categories are snow scene (3) and gale scene (1), and the second scene categories are snow scene (2) and gale scene (2). Taking T = 0.5 as an example, the first probability of the snow scene calculated by the distillation loss function is exp(3 / 0.5) / exp(3 / 0.5+2 / 0.5) = 0.665, and the second probability is exp(2 / 0.5) / exp(3 / 0.5+2 / 0.5) = 0.335. The first probability of the gale scene is exp(1 / 0.5) / exp(1 / 0.5+2 / 0.5) = 0.253, and the second probability is exp(2 / 0.5) / exp(2 / 0.5+2 / 0.5) = 0.747.

[0101] Step 134, determining the scene data whose difference between the first probability and the second probability is greater than a preset difference as the target scene data;

[0102] Step 135 : Generate a scene category label for the target scene data according to the first scene category or the second scene category.

[0103] In this embodiment, the confidence of the same scene category obtained by different models is substituted into the distillation loss function, and the first probability of the scene data for the first scene prediction model and the second probability of the scene data for the second scene prediction model under each scene category are calculated. The difference between the first probability and the second probability is calculated, and the prediction difference of the same scene data for different scene prediction models can be determined by the difference, so as to determine which scene data are valuable target scene data. When the difference corresponding to a certain scene data is greater than the preset difference, it means that the result difference obtained by the scene data based on the two scene prediction models is large, and the scene data can be determined as the target scene data. When the difference corresponding to a certain scene data is less than or equal to the preset difference, it means that the result difference obtained by the scene data based on the two scene prediction models is small, and it can be excluded as worthless scene data. After determining the target scene data, the target scene data is marked according to the corresponding scene category to generate the scene category label of the target scene data. On the one hand, using the distillation loss function to correct the confidence can make the corrected confidence describe the scene category more accurately and improve the reliability of the output confidence. On the other hand, it can accurately generate scene category labels for target scene data, eliminating the complicated procedure of manual labeling, which not only reduces the workload of manual screening, but also reduces the time required for scene data labeling, greatly improving data screening efficiency.

[0104] It is understandable that for a scene data, multiple scene categories may be predicted by the scene prediction model, and the first probability and the second probability of each scene category need to be obtained when calculating the distillation loss function. For the above situation, if the difference between the first probability and the second probability in any scene category is greater than the preset difference, the scene data can be used as the target scene data, or if the number of scene categories where the difference between the first probability and the second probability is greater than the preset difference is greater than the preset number, the scene data is used as the target scene data.

[0105] Step 140, sending the target scene data to the server;

[0106] Step 150: storing the updated scene prediction model sent by the server.

[0107] In this embodiment, the vehicle sends the target scene data obtained after screening to the server. After receiving the target scene data, the server trains the scene prediction model according to the target scene data to achieve the purpose of updating the scene prediction model. The server then deploys the updated scene prediction model to the vehicle. After the vehicle receives the new scene prediction model, the scene prediction model before the update is deleted, and the new scene prediction model is stored to facilitate the use of the new scene prediction model for subsequent data screening, scene prediction and other operations. In this way, the scene prediction model is continuously updated through high-quality target scene data, and the more target scene data is screened, the better the performance, which helps to improve the prediction accuracy of the scene prediction model and provide reliable and high-precision data support for autonomous driving.

[0108] Furthermore, as a refinement and extension of the specific implementation methods of the above-mentioned embodiment, in order to fully illustrate the specific implementation process of this embodiment, the data screening method also includes: predicting and processing the scene data according to the updated scene prediction model to determine the scene information of the road on which the vehicle is traveling.

[0109] In this embodiment, when the user needs to predict the scene the vehicle is currently in, the scene data collected by the constant sensor is input into the updated scene prediction model to obtain multiple scene categories and their corresponding confidence levels. According to the confidence level, the scene category that best matches the actual environment is selected as the scene information of the road the vehicle is traveling on. This realizes the automated scene prediction function and provides reliable and high-precision data support for autonomous driving.

[0110] In this embodiment, a data screening method is provided, which is applicable to a server, such as Figure 3 As shown, the method includes:

[0111] Step 210, receiving target scene data sent by the vehicle;

[0112] Step 220, preprocessing the target scene data;

[0113] Step 230, training the scene prediction model according to the preprocessed target scene data;

[0114] Step 240, sending the updated scene prediction model to the vehicle.

[0115] In this embodiment, the vehicle sends the target scene data obtained after screening to the server. After receiving the target scene data, the server first pre-processes the target scene data to convert the target scene data into a form more suitable for model training, and then uses the processed target scene data as a training sample to train the scene prediction model to achieve the purpose of updating the scene prediction model. The server then deploys the updated scene prediction model to the vehicle. After the vehicle receives the new scene prediction model, the scene prediction model before the update is deleted, and the new scene prediction model is stored to facilitate the use of the new scene prediction model for subsequent data screening, scene prediction and other operations. In this way, the scene prediction model is continuously updated through high-quality target scene data, and the more target scene data is screened, the better the performance, which helps to improve the prediction accuracy of the scene prediction model and provide reliable and high-precision data support for autonomous driving.

[0116] Furthermore, if Figure 4 As shown, as a specific implementation of the above-mentioned data screening method, an embodiment of the present application provides a data screening device 400, which includes: an acquisition module 401, a screening module 402, a communication module 403 and a storage module 404.

[0117] The acquisition module 401 is used to acquire scene data of the road on which the vehicle is traveling collected by the sensor;

[0118] A screening module 402 is used to perform a first screening process on the scene data according to a preset screening strategy; and to perform a second screening process on the scene data after the first screening process according to a scene prediction model to determine target scene data, where the target scene data carries a scene category label;

[0119] The communication module 403 is used to send the target scene data to the server so that the server updates the scene prediction model according to the target scene data;

[0120] The storage module 404 is used to store the updated scene prediction model sent by the server.

[0121] In this embodiment, after obtaining the scene data, the vehicle first uses a pre-set preset screening strategy to eliminate low-quality data, and then uses the trained scene prediction model for secondary screening to select target scene data with higher scene recognition value, so as to use the target scene data with scene category labels as training samples for the scene prediction model. On the one hand, while realizing automatic screening, the data scene screening performance is continuously improved by combining rule screening and neural network learning, effectively reducing the amount of data transmitted and stored, shortening the long-cycle data screening process to near real-time, improving the scene data screening efficiency, and accurately generating the scene category labels of the target scene data through the scene prediction model, eliminating manual labeling and reducing the cost of manual screening. On the other hand, the data screening process is completed by the vehicle-side processor, which effectively avoids overloading the server due to excessive data volume. On the other hand, continuously updating the scene prediction model through high-quality target scene data helps to improve the prediction accuracy of the scene prediction model and provide reliable and high-precision data support for autonomous driving.

[0122] Furthermore, the screening module 402 is specifically used to input the scene data after the first screening process into the first scene prediction model to obtain the first scene category and the confidence of the first scene category; input the scene data after the first screening process into the second scene prediction model to obtain the second scene category and the confidence of the second scene category, wherein the parameter amount of the second scene prediction model is smaller than the parameter amount of the first scene prediction model; use the distillation loss function to process the confidence of the first scene category and the confidence of the second scene category that is the same as the first scene category to obtain the first probability and the second probability of the scene data; determine the scene data whose difference between the first probability and the second probability is greater than the preset difference as the target scene data; generate the scene category label of the target scene data according to the first scene category or the second scene category.

[0123] Furthermore, the distillation loss function adopts the following formula:

[0124]

[0125] Where q represents probability, z represents confidence, T represents distillation temperature coefficient, and i represents a scene category in j.

[0126] Furthermore, the screening module 402 is specifically configured to delete scene data whose average brightness value exceeds a preset brightness range.

[0127] Furthermore, the screening module 402 is specifically configured to delete scene data with a definition lower than a preset definition.

[0128] Furthermore, the screening module 402 is specifically configured to delete one of the two scene data if the similarity between the two scene data is greater than or equal to a preset similarity.

[0129] Furthermore, the data screening device 400 also includes: a prediction module (not shown in the figure), which is used to perform prediction processing on the scene data according to the updated scene prediction model to determine the scene information of the road on which the vehicle is traveling.

[0130] Furthermore, if Figure 5 As shown, as a specific implementation of the above-mentioned data screening method, an embodiment of the present application provides a data screening device 500, and the data screening device 500 includes: a communication module 501, a preprocessing module 502 and a model training module 503.

[0131] Wherein, the communication module 501 is used to receive target scene data sent by the vehicle;

[0132] A preprocessing module 502, used for preprocessing target scene data;

[0133] A model training module 503 is used to train the scene prediction model according to the preprocessed target scene data to update the scene prediction model;

[0134] The communication module 501 is also used to send the updated scene prediction model to the vehicle.

[0135] In this embodiment, the vehicle sends the target scene data obtained after screening to the server. After receiving the target scene data, the server first pre-processes the target scene data to convert the target scene data into a form more suitable for model training, and then uses the processed target scene data as a training sample to train the scene prediction model to achieve the purpose of updating the scene prediction model. The server then deploys the updated scene prediction model to the vehicle. After the vehicle receives the new scene prediction model, the scene prediction model before the update is deleted, and the new scene prediction model is stored to facilitate the use of the new scene prediction model for subsequent data screening, scene prediction and other operations. In this way, the scene prediction model is continuously updated through high-quality target scene data, and the more target scene data is screened, the better the performance, which helps to improve the prediction accuracy of the scene prediction model and provide reliable and high-precision data support for autonomous driving.

[0136] For the specific definition of the data screening device, please refer to the definition of the data screening method above, which will not be repeated here. Each module in the above-mentioned data screening device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0137] Based on the above Figures 1 to 3 The method shown in the embodiment of the present application accordingly provides a readable storage medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned Figures 1 to 3 Data filtering method shown.

[0138] Based on this understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the data screening method described in each implementation scenario of the present application.

[0139] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages and saves the hardware and software resources of the computer device, and supports the operation of information processing programs and other software and / or programs. The network communication module is used to realize communication between the components inside the storage medium, and communication with other hardware and software in the physical device.

[0140] Based on the above Figure 1 and Figure 2 The method shown, and Figure 4 In order to achieve the above-mentioned purpose, the embodiment of the present application further provides a vehicle, which includes a sensor and a processor, the processor and a memory are coupled, and the memory stores a program or instruction running on the processor, and the program or instruction is executed by the processor to achieve the above-mentioned Figure 1 and Figure 2 Data filtering method shown.

[0141] Based on the above Figure 3 The method shown, and Figure 5 In order to achieve the above-mentioned purpose, the embodiment of the present application further provides a server, which includes a processor, the processor and a memory are coupled, and the memory stores a program or instruction running on the processor, and the program or instruction is executed by the processor to achieve the above-mentioned Figure 3 Data filtering method shown.

[0142] Optionally, the vehicle may further include a user interface, a network interface, a radio frequency (RF) circuit, an audio circuit, a WI-FI module, etc. The user interface may include a display, an input unit such as a keyboard, etc., and the optional user interface may also include a USB interface, a card reader interface, etc. The network interface may optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), etc.

[0143] Those skilled in the art will appreciate that the vehicle structure provided in this embodiment does not constitute a limitation on the computer device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0144] Through the description of the above implementation methods, the technicians in this field can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform, or by hardware to obtain the scene data of the road where the vehicle is located collected by the sensor; the scene data is first filtered according to the preset filtering strategy; the scene data after the first filtering is second filtered according to the scene prediction model to determine the target scene data, and the target scene data carries the scene category label; the target scene data is sent to the server so that the server updates the scene prediction model according to the target scene data; the updated scene prediction model sent by the server is stored. On the one hand, the embodiment of the present application, while realizing automatic screening, combines rule screening and neural network learning to continuously improve the data scene screening performance, effectively reduce the amount of data transmitted and stored, shorten the long-cycle data screening process to near real-time, improve the scene data screening efficiency, and can accurately generate the scene category label of the target scene data through the scene prediction model, eliminating manual labeling and reducing the cost of manual screening. On the other hand, the data screening process is completed by the vehicle-side processor, which effectively avoids the server from being overloaded due to excessive data volume. On the other hand, continuously updating the scene prediction model through high-quality target scene data will help improve the prediction accuracy of the scene prediction model and provide reliable and high-precision data support for autonomous driving.

[0145] Those skilled in the art will appreciate that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily necessary for implementing the present application. Those skilled in the art will appreciate that the modules in the devices in the implementation scenario can be distributed in the devices of the implementation scenario according to the description of the implementation scenario, or can be changed accordingly and located in one or more devices different from the present implementation scenario. The modules of the above-mentioned implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0146] The above serial numbers of this application are only for description and do not represent the advantages and disadvantages of the implementation scenarios. The above disclosure is only a few specific implementation scenarios of this application, but this application is not limited to them, and any changes that can be thought of by technicians in this field should fall within the scope of protection of this application.

Claims

1. A data screening method, characterized in that: The method comprises: Obtaining scene data of the road on which the vehicle is traveling collected by sensors; Performing a first screening process on the scene data according to a preset screening strategy; Performing a second screening process on the scene data after the first screening process according to the scene prediction model to determine target scene data, wherein the target scene data carries a scene category label; Sending the target scene data to a server so that the server updates the scene prediction model according to the target scene data; The updated scene prediction model sent by the server is stored.

2. The data screening method according to claim 1, characterized in that: The performing a second screening process on the scene data after the first screening process according to the scene prediction model to determine the target scene data includes: Inputting the scene data after the first screening process into a first scene prediction model to obtain a first scene category and a confidence level of the first scene category; Inputting the scene data after the first screening process into a second scene prediction model to obtain a second scene category and a confidence level of the second scene category, wherein the parameter amount of the second scene prediction model is less than the parameter amount of the first scene prediction model; Using a distillation loss function to process the confidence of the first scene category and the confidence of the second scene category that is the same as the first scene category, to obtain a first probability and a second probability of the scene data; Determine the scene data where the difference between the first probability and the second probability is greater than a preset difference as the target scene data; The scene category label of the target scene data is generated according to the first scene category or the second scene category.

3. The data screening method according to claim 2, characterized in that: The distillation loss function adopts the following formula: Where q represents probability, z represents confidence, T represents distillation temperature coefficient, and i represents a scene category in j.

4. The data screening method according to claim 1, characterized in that: The first screening process of the scene data according to the preset screening strategy includes: deleting the scene data whose average brightness value exceeds a preset brightness range; and / or, Deleting the scene data whose definition is less than a preset definition; and / or, If the similarity between any two scene data is greater than or equal to a preset similarity, one of the two scene data is deleted.

5. The data screening method according to any one of claims 1 to 4, characterized in that: The method further comprises: The scene data is predicted and processed according to the updated scene prediction model to determine the scene information of the road on which the vehicle is traveling.

6. A data screening method, characterized in that: The method comprises: Receive target scene data sent by the vehicle; Preprocessing the target scene data; Training the scene prediction model according to the preprocessed target scene data to update the scene prediction model; The updated scene prediction model is sent to the vehicle.

7. A data screening device, characterized in that: The device comprises: An acquisition module, used to acquire scene data of the road on which the vehicle is traveling collected by a sensor; a screening module, configured to perform a first screening process on the scene data according to a preset screening strategy; and Performing a second screening process on the scene data after the first screening process according to the scene prediction model to determine target scene data, wherein the target scene data carries a scene category label; A communication module, used for sending the target scene data to a server, so that the server updates the scene prediction model according to the target scene data; A storage module is used to store the updated scene prediction model sent by the server.

8. A data screening device, characterized in that: The device comprises: A communication module, used for receiving target scene data sent by a vehicle; A preprocessing module, used for preprocessing the target scene data; A model training module, used to train the scene prediction model according to the preprocessed target scene data to update the scene prediction model; The communication module is also used to send the updated scene prediction model to the vehicle.

9. A vehicle, characterized in that: It includes a sensor and a processor, the processor is coupled to a memory, the memory stores a program or instruction running on the processor, and the program or instruction, when executed by the processor, implements the steps of the data screening method as described in any one of claims 1 to 5.

10. A server, characterized in that: It includes a processor, the processor is coupled to a memory, the memory stores a program or instruction running on the processor, and the program or instruction, when executed by the processor, implements the steps of the data screening method as claimed in claim 6.

11. A computer-readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by a processor, the steps of the data screening method according to any one of claims 1 to 6 are implemented.

Images