Screening method and system suitable for special-shaped vehicles in automatic driving training
By combining shape image processing algorithms and neural network models, the shape and multi-scale features of special-shaped vehicles are extracted and identified, real-time screening of special-shaped vehicles in autonomous driving training is achieved, and the problems of small data volume and long training cycle are solved, improving the problem of Training efficiency and safety.
Patent Information
- Application Number
- CN202310550274.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-16
- Publication Date
- 2024-11-22
AI Technical Summary
The existing autonomous driving training methods have less data when dealing with special-shaped vehicles, resulting in poor training results. The existing data screening methods cannot meet the real-time requirements, resulting in too long algorithm training cycle.
Using a combination of shape image processing algorithm and neural network model, data is obtained through sensors, shape features and multi-scale features are extracted, and the identification and screening of special-shaped vehicles is carried out, and the union results are used to achieve real-time screening and data uploading, shortening the training cycle.
The quality of training samples in the special-shaped vehicle scenarios has been improved, the overall training cycle has been reduced, the detection ability of special-shaped vehicles has been enhanced, and the risk of traffic accidents has been reduced.
Smart Images

Figure CN119007156A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of autonomous driving technology, and in particular to a method and system for screening special-shaped vehicles in autonomous driving training. Background Art
[0002] Before a vehicle can achieve autonomous driving on the road, it is generally necessary to undergo complex training under various road conditions. Current autonomous driving training is often conducted in the following ways: first, data is obtained based on the camera, sensors, and lidar on the training vehicle, and then the corresponding algorithm is trained and optimized based on the data.
[0003] However, during the autonomous driving training process, due to the small number of road conditions in extreme scenarios, the amount of data for detection training for special-shaped vehicles such as modified vehicles and accident vehicles is often small; especially in the existing autonomous driving training, multiple training vehicles are often used for actual testing training on different roads, which leads to the fact that the proportion of data corresponding to special-shaped vehicles is also extremely low when considering bulk data. Therefore, during the autonomous driving training process, the training processing effect for special-shaped vehicles is often poor, and in actual autonomous driving, there is often a risk of being unable to effectively avoid risks when encountering special-shaped vehicles, which in turn causes serious traffic accidents.
[0004] Therefore, in autonomous driving training, it is necessary to screen the data corresponding to special-shaped vehicles for targeted processing. However, the existing data screening methods are mostly manual screening or offline program screening. Among them, manual screening is to screen the data after annotating it by category by manual labelers; offline program screening is automatic screening based on manually set rules. However, this type of data screening method often has the defect of a long screening cycle, and the data screening for special-shaped vehicles in autonomous driving training needs to have a certain degree of real-time performance to optimize the algorithm in real time, so that the latest optimization algorithm can be adopted when encountering the corresponding extreme scenario next time, thereby shortening the training cycle of the corresponding algorithm as a whole. Summary of the invention
[0005] The purpose of this application is to provide a method and system for screening special-shaped vehicles in autonomous driving training, so as to improve the technical problem that the existing manual screening or offline program screening cannot meet the real-time requirements of data screening corresponding to special-shaped vehicles in autonomous driving training, and shorten the corresponding training cycle.
[0006] To achieve the above objectives, this application proposes the following technical solutions:
[0007] In the first aspect, the technical solution provides a method for screening special-shaped vehicles in autonomous driving training, including:
[0008] Acquire the raw data uploaded by each sensor of the training vehicle; wherein the raw data is stored in the on-board chip of the training vehicle;
[0009] Based on the shape image processing algorithm, shape feature extraction and shape feature recognition are performed on each detection target in the original data to obtain a first judgment result, and at the same time, based on the neural network model, an adaptive detection frame is added to each detection target in the original data, and multi-scale feature extraction and multi-scale feature recognition are performed to obtain a second judgment result; wherein the first judgment result and the second judgment result both include: whether the detection target is a special-shaped vehicle, and the type of the special-shaped vehicle;
[0010] Taking a union of the first judgment result and the second judgment result to obtain a final judgment result corresponding to the detection target in each of the original data;
[0011] Based on the final judgment result, the original data corresponding to the special-shaped vehicle is screened from the original data as the target data, and the target data and the corresponding special-shaped vehicle type are uploaded to the cloud.
[0012] Furthermore, the shape feature extraction and shape feature recognition of each detection target in the original data based on the shape image processing algorithm to obtain a first judgment result includes:
[0013] Extract shape features of each detection target in the original data based on the image segmentation algorithm;
[0014] When the detection target is judged to be a special-shaped vehicle based on the shape feature, whether the detection target is a rolled-over vehicle is judged based on the rotation angle of the shape feature, or whether the detection target is a fallen vehicle is judged based on the three-dimensional posture of the shape feature.
[0015] Furthermore, the adding of an adaptive detection frame to each detection target in the original data based on the neural network model and performing multi-scale feature extraction and multi-scale feature recognition to obtain a second judgment result includes:
[0016] Acquire a multi-scale feature corresponding to each of the original data, and when it is determined that there are a number of targets to be detected in the multi-scale feature, add a focus area corresponding to the target detection target and a focus exclusion area corresponding to the invalid detection target to the multi-scale feature;
[0017] The focused area in the multi-scale feature is identified to obtain a second determination result.
[0018] Furthermore, the adding of an adaptive detection frame to each detection target in the original data based on the neural network model and performing multi-scale feature extraction and multi-scale feature recognition to obtain a second judgment result includes:
[0019] When it is determined based on the binary classification model that there is a special-shaped vehicle, the category of the special-shaped vehicle is detected based on the multi-classification model.
[0020] Further, obtaining the neural network model includes:
[0021] Obtain several training samples and construct training sets and validation sets respectively;
[0022] Training an initial neural network model based on the training set, and verifying the trained initial neural network model based on the verification set;
[0023] The training process based on the training set and the verification process based on the verification set are repeatedly performed until the neural network model is obtained.
[0024] In the second aspect, the technical solution provides a screening system for special-shaped vehicles in autonomous driving training, including:
[0025] A first acquisition module is used to acquire the raw data uploaded by each sensor of the training vehicle; wherein the raw data is stored in the on-board chip of the training vehicle;
[0026] A judgment module, configured to extract and recognize shape features of each detection target in the original data based on a shape image processing algorithm to obtain a first judgment result, and to extract and recognize multi-scale features of each detection target in the original data based on a neural network model after adding an adaptive detection frame to obtain a second judgment result; wherein the first judgment result and the second judgment result both include: whether the detection target is a special-shaped vehicle, and the type of the special-shaped vehicle;
[0027] A result fusion module, used for taking a union of the first judgment result and the second judgment result to obtain a final judgment result corresponding to the detection target in each original data;
[0028] The screening module is used to screen the raw data corresponding to the special-shaped vehicle from the raw data based on the final judgment result as the target data, and upload the target data and the corresponding special-shaped vehicle type to the cloud.
[0029] Furthermore, the judging module includes:
[0030] A first extraction unit, used for extracting shape features of the detection target in each original data based on an image segmentation algorithm;
[0031] The first judgment unit is used to judge whether the detection target is a special-shaped vehicle based on the shape feature, whether the detection target is a rollover vehicle based on the rotation angle of the shape feature, or whether the detection target is a fallen vehicle based on the three-dimensional posture of the shape feature.
[0032] Furthermore, the judging module includes:
[0033] A first adding unit is used to obtain a multi-scale feature corresponding to each of the original data, and when it is determined that there are a number of targets to be detected in the multi-scale feature, add a key focus area corresponding to the target detection target and a key exclusion area corresponding to the invalid detection target to the multi-scale feature;
[0034] The second judgment unit is used to identify the key focus area in the multi-scale feature to obtain a second judgment result.
[0035] In a third aspect, the technical solution provides an electronic device, comprising at least one processor, wherein the processor is coupled to a memory, wherein a computer program is stored in the memory, and wherein the computer program is configured to execute the method described when executed by the processor.
[0036] In a fourth aspect, the present technical solution provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is used to execute the described method.
[0037] It can be seen from the above technical solutions that the technical solution of the present application provides a screening method for special-shaped vehicles in autonomous driving training, so as to improve the existing autonomous driving training methods and general screening methods that cannot meet the training requirements for extreme scenarios under special-shaped vehicles.
[0038] In this technical solution, the shape image processing algorithm and the neural network algorithm are introduced into the screening process at the same time. Since the shape image processing algorithm is based on pre-set rules, and the neural network model corresponding to the neural network algorithm is also pre-optimized and trained in a targeted manner, the acquired raw data is processed based on them to meet the real-time requirements. At the same time, considering the possible missed detection when a single shape image processing algorithm and a neural network model are used to process raw data, the final screening target is determined based on the union of the first judgment result and the second judgment result corresponding to them; thereby screening out the raw data corresponding to the special-shaped vehicle to the greatest extent. The training samples in the special-shaped vehicle scene in the autonomous driving training are increased, the quality of a single training is improved, and the overall training cycle is reduced.
[0039] It should be appreciated that all combinations of the foregoing concepts, as well as additional concepts described in greater detail below, may be considered to be part of the inventive subject matter of the present disclosure, provided such concepts are not mutually inconsistent.
[0040] The foregoing and other aspects, embodiments and features of the present application can be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the present application, such as the features and / or beneficial effects of the exemplary embodiments, will be apparent from the following description or learned through practice of the specific embodiments according to the teachings of the present application. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in various figures may be represented by the same reference numeral. For clarity, not every component is labeled in each figure. Embodiments of various aspects of the present application will now be described by way of example and with reference to the accompanying drawings, in which:
[0042] Figure 1 This is a flow chart of the method for screening special-shaped vehicles in automatic driving according to this embodiment;
[0043] Figure 2 A flowchart for obtaining a first judgment result;
[0044] Figure 3 is a flow chart for obtaining the second judgment result;
[0045] Figure 4 is another flow chart for obtaining the second judgment result;
[0046] Figure 5 The flowchart for obtaining the neural network model;
[0047] Figure 6 This is a block diagram of the system for screening special-shaped vehicles in automatic driving according to this embodiment;
[0048] Figure 7 1 is a structural diagram of the electronic device described in this embodiment. DETAILED DESCRIPTION
[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution of the embodiments of the present application will be clearly and completely described below in conjunction with the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application. Unless otherwise defined, the technical terms or scientific terms used herein should be the common meanings understood by people with general skills in the field to which this application belongs.
[0050] The words "first", "second" and similar words used in the patent application specification and claims of this application do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, unless the context clearly indicates otherwise, the singular form of "a", "an" or "the" and other similar words do not indicate a quantitative limitation, but rather indicate the presence of at least one. Words such as "include" or "comprise" mean that the elements or objects appearing before "include" or "comprise" include the features, wholes, steps, operations, elements and / or components listed after "include" or "comprise", and do not exclude the existence or addition of one or more other features, wholes, steps, operations, elements, components and / or their collections. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0051] In order to enable autonomous driving vehicles to cope with various road conditions, they need to be trained for autonomous driving in various scenarios. However, due to the small number of extreme scenarios under special-shaped vehicles, the autonomous driving training data in this scenario is extremely low in both quantity and proportion compared to the training data in other scenarios, which makes it impossible to effectively achieve the training purpose. When technicians intend to conduct targeted screening of training data corresponding to extreme scenarios, they find that the existing screening methods generally have the defect of long screening cycles. And it is unable to meet the real-time requirements of training data screening in extreme scenarios. Therefore, this embodiment aims to provide a screening method suitable for special-shaped vehicles in autonomous driving training to improve the above technical defects.
[0052] In conjunction with the accompanying drawings, a method for screening special-shaped vehicles in autonomous driving training shown in this embodiment is specifically introduced below.
[0053] like Figure 1 As shown, the method comprises the following steps:
[0054] Step S102, obtaining the original data uploaded by each sensor of the training vehicle; wherein the original data is stored in the on-board chip of the training vehicle.
[0055] In specific implementation, the raw data includes image data obtained by the camera mounted on the self-training vehicle, and point cloud data obtained by the millimeter wave radar or laser radar. Generally, the number of training vehicles exceeds 200, and the amount of raw data generated in one day is a massive amount of data in the PB (Petabyte) level.
[0056] Step S104: perform shape feature extraction and shape feature recognition on each detection target in the original data based on the shape image processing algorithm to obtain a first judgment result, and at the same time, perform multi-scale feature extraction and multi-scale feature recognition after adding an adaptive detection frame to each detection target in the original data based on the neural network model to obtain a second judgment result.
[0057] In this embodiment, the first judgment result and the second judgment result both include: whether the detection target is a special-shaped vehicle, and the type of the special-shaped vehicle.
[0058] Based on this, as a specific implementation method, Figure 2 As shown, obtaining the first judgment result includes:
[0059] Step S1042: extract shape features of each detection target in the original data based on an image segmentation algorithm.
[0060] In a specific implementation, certain shapes, such as trapezoid or diamond, are not common in ordinary vehicles and can be used to distinguish special-shaped vehicles from ordinary vehicles.
[0061] Step S1044: when the detection target is determined to be a special-shaped vehicle based on the shape feature, whether the detection target is a rolled-over vehicle is determined based on the rotation angle of the shape feature, or whether the detection target is a fallen vehicle is determined based on the three-dimensional posture of the shape feature.
[0062] As a preferred embodiment, when determining the specific type of special-shaped vehicle, the corresponding size or color can be introduced to combine with the shape to identify the specific special-shaped vehicle, such as combining the shape feature and the corresponding color to determine whether it is a vehicle on fire.
[0063] At the same time, as a specific implementation method, Figure 3 As shown, obtaining the second judgment result includes:
[0064] Step S1042': when it is determined based on the binary classification model that there is a special-shaped vehicle, the category of the special-shaped vehicle is detected based on the multi-classification model.
[0065] In a specific implementation, the binary classification model may be a focal loss binary classification model, and the multi-classification model may be a softmax (normalized exponential function) multi-classification model.
[0066] As a preferred implementation, in order to improve the accuracy of the second judgment result, as Figure 4 As shown, it also includes:
[0067] Step S1042″: obtaining a multi-scale feature corresponding to each of the original data, and when it is determined that there are a number of targets to be detected in the multi-scale feature, adding a key focus area corresponding to the target detection target and a key exclusion area corresponding to the invalid detection target to the multi-scale feature.
[0068] Step S1044', identifying the focus area in the multi-scale feature to obtain a second judgment result.
[0069] At this time, the addition of the key focus area and the key exclusion area avoids the recognition of invalid detection targets such as background noise, and achieves more accurate recognition of detection targets under multi-scale features.
[0070] In order to allow more raw data corresponding to special-shaped vehicles to be screened and used, the first judgment result and the second judgment result are both set to have a higher confidence level. In this embodiment, their confidence levels are both set to be no less than 95%. At the same time, the following steps are continued:
[0071] Step S106: taking a union of the first judgment result and the second judgment result to obtain a final judgment result corresponding to the detection target in each original data.
[0072] In this embodiment, step S104 and step S106 are both performed based on the vehicle-mounted chip.
[0073] Step S108: based on the final judgment result, original data corresponding to the special-shaped vehicle is screened from the original data as target data, and the target data and the corresponding special-shaped vehicle type are uploaded to the cloud.
[0074] As an optional implementation, Figure 5 As shown, the neural network model is specifically obtained through the following steps:
[0075] Step S202: Obtain a number of training samples, and construct a training set and a validation set respectively.
[0076] Step S204: training the initial neural network model based on the training set, and verifying the trained initial neural network model based on the verification set.
[0077] Step S206, repeatedly performing the training process based on the training set and the verification process based on the verification set until the neural network model is obtained.
[0078] In summary, it can be seen that the screening method described in this embodiment simultaneously introduces a shape image processing algorithm and a neural network model to meet the timeliness of screening special-shaped vehicles. At the same time, the combination of the two also enables more possible target data corresponding to special-shaped vehicles to be obtained in the end, thereby reducing the automatic driving training cycle for special-shaped vehicles.
[0079] The above program can be run in the processor, or it can be stored in the memory (or computer-readable storage medium), which includes permanent and non-permanent, removable and non-removable media. Information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer-readable storage media does not include temporary computer-readable media, such as modulated data signals and carrier waves.
[0080] These computer programs can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps of the functions specified in one or more blocks can be implemented by different modules corresponding to different steps.
[0081] Based on this, this embodiment also provides a screening system for special-shaped vehicles in autonomous driving training. Figure 6 As shown, the system comprises:
[0082] The first acquisition module is used to acquire the original data uploaded by each sensor of the training vehicle; wherein the original data is stored in the on-board chip of the training vehicle.
[0083] A judgment module is used to perform shape feature extraction and shape feature recognition on each detection target in the original data based on a shape image processing algorithm to obtain a first judgment result, and at the same time, based on a neural network model, add an adaptive detection frame to each detection target in the original data and perform multi-scale feature extraction and multi-scale feature recognition to obtain a second judgment result; wherein the first judgment result and the second judgment result both include: whether the detection target is a special-shaped vehicle, and the type of the special-shaped vehicle.
[0084] The result fusion module is used to take the union of the first judgment result and the second judgment result to obtain a final judgment result corresponding to the detection target in each original data.
[0085] The screening module is used to screen the raw data corresponding to the special-shaped vehicle from the raw data based on the final judgment result as the target data, and upload the target data and the corresponding special-shaped vehicle type to the cloud.
[0086] The system is used to implement the steps of the method for screening special-shaped vehicles in autonomous driving training disclosed in the above-mentioned embodiment, which have been described and will not be repeated here.
[0087] For example, the judgment module includes:
[0088] The first extraction unit is used to extract shape features of the detection target in each original data based on an image segmentation algorithm.
[0089] The first judgment unit is used to judge whether the detection target is a special-shaped vehicle based on the shape feature, whether the detection target is a rollover vehicle based on the rotation angle of the shape feature, or whether the detection target is a fallen vehicle based on the three-dimensional posture of the shape feature.
[0090] For example, the judgment module includes:
[0091] The first adding unit is used to obtain a multi-scale feature corresponding to each of the original data, and when it is determined that there are a number of targets to be detected in the multi-scale feature, add a key focus area corresponding to the target detection target and a key exclusion area corresponding to the invalid detection target to the multi-scale feature.
[0092] The second judgment unit is used to identify the key focus area in the multi-scale feature to obtain a second judgment result.
[0093] For another example, the judgment module includes:
[0094] The third judgment unit is used to detect the category of the special-shaped vehicle based on the multi-classification model when judging that there is a special-shaped vehicle based on the binary classification model.
[0095] For example, including:
[0096] The second acquisition module is used to acquire a number of training samples and construct a training set and a validation set respectively.
[0097] The iterative training module is used to train the initial neural network model based on the training set, and to verify the trained initial neural network model based on the verification set.
[0098] The training module is used to repeatedly perform the training process based on the training set and the verification process based on the verification set until the neural network model is obtained.
[0099] Since the system is built based on the method, it also has the advantage of high real-time performance when screening special-shaped vehicles in autonomous driving training based on the system.
[0100] This embodiment also provides an electronic device. Figure 7 As shown, it includes at least one processor, the processor is coupled to a memory, a computer program is stored in the memory, and the computer program is configured to execute the above method when executed by the processor.
[0101] This embodiment also provides a computer-readable storage medium on which a computer program is stored. The computer program is used to execute the above method.
[0102] Although the present application has been disclosed as above with preferred embodiments, it is not intended to limit the present application. A person with ordinary knowledge in the technical field to which the present application belongs can make various changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be determined by the definition of the claims.
Claims
1. A method for screening special-shaped vehicles in autonomous driving training. It is characterized in that include: Acquire the raw data uploaded by each sensor of the training vehicle; wherein the raw data is stored in the on-board chip of the training vehicle; Based on the shape image processing algorithm, shape feature extraction and shape feature recognition are performed on each detection target in the original data to obtain a first judgment result, and at the same time, based on the neural network model, an adaptive detection frame is added to each detection target in the original data, and multi-scale feature extraction and multi-scale feature recognition are performed to obtain a second judgment result; wherein the first judgment result and the second judgment result both include: whether the detection target is a special-shaped vehicle, and the type of the special-shaped vehicle; Taking a union of the first judgment result and the second judgment result to obtain a final judgment result corresponding to the detection target in each of the original data; Based on the final judgment result, the original data corresponding to the special-shaped vehicle is screened from the original data as the target data, and the target data and the corresponding special-shaped vehicle type are uploaded to the cloud.
2. The method for screening special-shaped vehicles in autonomous driving training according to claim 1, It is characterized in that The shape feature extraction and shape feature recognition of each detection target in the original data based on the shape image processing algorithm to obtain a first judgment result includes: Extract shape features of each detection target in the original data based on the image segmentation algorithm; When the detection target is judged to be a special-shaped vehicle based on the shape feature, whether the detection target is a rolled-over vehicle is judged based on the rotation angle of the shape feature, or whether the detection target is a fallen vehicle is judged based on the three-dimensional posture of the shape feature.
3. The method for screening special-shaped vehicles in autonomous driving training according to claim 1, It is characterized in that The method of adding an adaptive detection frame to each detection target in the original data based on the neural network model and then performing multi-scale feature extraction and multi-scale feature recognition to obtain a second judgment result includes: Acquire a multi-scale feature corresponding to each of the original data, and when it is determined that there are a number of targets to be detected in the multi-scale feature, add a focus area corresponding to the target detection target and a focus exclusion area corresponding to the invalid detection target to the multi-scale feature; The focused area in the multi-scale feature is identified to obtain a second determination result.
4. The method for screening special-shaped vehicles in autonomous driving training according to claim 1, It is characterized in that The method of adding an adaptive detection frame to each detection target in the original data based on the neural network model and then performing multi-scale feature extraction and multi-scale feature recognition to obtain a second judgment result includes: When it is determined based on the binary classification model that there is a special-shaped vehicle, the category of the special-shaped vehicle is detected based on the multi-classification model.
5. The method for screening special-shaped vehicles in autonomous driving training according to claim 1, It is characterized in that Acquiring the neural network model includes: Obtain several training samples and construct training sets and validation sets respectively; Training an initial neural network model based on the training set, and verifying the trained initial neural network model based on the verification set; The training process based on the training set and the verification process based on the verification set are repeatedly performed until the neural network model is obtained.
6. A screening system for special-shaped vehicles in autonomous driving training, It is characterized in that include: A first acquisition module is used to acquire the raw data uploaded by each sensor of the training vehicle; wherein the raw data is stored in the on-board chip of the training vehicle; A judgment module, configured to extract and recognize shape features of each detection target in the original data based on a shape image processing algorithm to obtain a first judgment result, and to extract and recognize multi-scale features of each detection target in the original data based on a neural network model after adding an adaptive detection frame to obtain a second judgment result; wherein the first judgment result and the second judgment result both include: whether the detection target is a special-shaped vehicle, and the type of the special-shaped vehicle; A result fusion module, used for taking a union of the first judgment result and the second judgment result to obtain a final judgment result corresponding to the detection target in each original data; The screening module is used to screen the raw data corresponding to the special-shaped vehicle from the raw data based on the final judgment result as the target data, and upload the target data and the corresponding special-shaped vehicle type to the cloud.
7. The screening system for special-shaped vehicles in autonomous driving training according to claim 6, It is characterized in that The judging module comprises: A first extraction unit, used for extracting shape features of the detection target in each original data based on an image segmentation algorithm; The first judgment unit is used to judge whether the detection target is a special-shaped vehicle based on the shape feature, whether the detection target is a rollover vehicle based on the rotation angle of the shape feature, or whether the detection target is a fallen vehicle based on the three-dimensional posture of the shape feature.
8. The screening system for special-shaped vehicles in autonomous driving training according to claim 6, It is characterized in that The judging module comprises: A first adding unit is used to obtain a multi-scale feature corresponding to each of the original data, and when it is determined that there are a number of targets to be detected in the multi-scale feature, add a key focus area corresponding to the target detection target and a key exclusion area corresponding to the invalid detection target to the multi-scale feature; The second judgment unit is used to identify the key focus area in the multi-scale feature to obtain a second judgment result.
9. An electronic device, It is characterized in that The method comprises at least one processor, wherein the processor is coupled to a memory, wherein a computer program is stored in the memory, and wherein the computer program is configured to execute the method according to any one of claims 1 to 5 when executed by the processor.
10. A computer-readable storage medium, It is characterized in that A computer program is stored thereon, and the computer program is used to execute the method according to any one of claims 1 to 5.
