Method, device and equipment for identifying traffic accident behavior and computer readable medium

Abstract

The invention discloses a method, device and equipment for identifying traffic accident behaviors and a computer readable medium, and relates to the technical field of computers. A specific embodiment of the method comprises the following steps: inputting a traffic accident image into a target detection model, and identifying traffic prohibition signs and the number of the traffic prohibition signs; intercepting a sub-graph of a traffic prohibition sign in the traffic accident image, identifying a traffic prohibition corresponding to the traffic prohibition sign, analyzing a traffic accident text corresponding to the traffic accident image according to the traffic prohibition, and determining a traffic accident behavior; and screening out a traffic accident retrieval behavior from the traffic accident behavior and the traffic accident similar behavior similar to the traffic accident behavior, querying the knowledge base according to the traffic accident retrieval behavior, and obtaining and pushing a traffic illegal behavior corresponding to the traffic accident query behavior, a traffic law article corresponding to the traffic accident query behavior and a corresponding investigation operation corresponding to the traffic accident query behavior. According to the embodiment, the time consumed for identifying traffic accident behaviors can be shortened, and the identification accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and in particular, to a method and device for identifying traffic accident behavior, equipment and computer readable medium. BACKGROUND

[0002] In the field of vehicle insurance, accurate determination of traffic accident liability is the core link to determine the compensation amount, control the cost of claims settlement and prevent moral risk. The traditional vehicle insurance claims settlement highly depends on manual work, which requires claims settlement staff to repeatedly compare accident scene pictures, traffic rules, historical cases and investigation guidelines, which is time-consuming, highly subjective and easy to cause disputes. Traffic police also face difficulties such as lack of standards for evidence interpretation, lack of support for legal reasoning, and limited police forces, resulting in different judgments for the same case, high rate of reconsideration, and difficulty in balancing efficiency and accuracy.

[0003] In the process of implementing the present application, the inventors found that at least the following problems exist in the prior art: When facing a large number of traffic accident behaviors, the efficiency of manual liability determination is bottlenecked. SUMMARY

[0004] Therefore, the embodiments of the present application provide a method and device for identifying traffic accident behavior, equipment and computer readable medium, which can shorten the time-consuming of identifying traffic accident behavior and improve the identification accuracy.

[0005] To achieve the above-mentioned purpose, according to one aspect of the embodiments of the present application, a method for identifying traffic accident behavior is provided, comprising:

[0006] inputting a traffic accident image into a target detection model, wherein the target detection model identifies traffic prohibition signs and the number of the traffic prohibition signs in the traffic accident image according to color features of the traffic prohibition signs and difference features between the traffic prohibition signs and non-traffic prohibition signs;

[0007] cutting a sub-image of the number of traffic prohibition signs in the traffic accident image, identifying traffic prohibitions corresponding to the traffic prohibition signs in the sub-image of the traffic prohibition signs, analyzing traffic accident texts corresponding to the traffic accident image according to the traffic prohibitions, and determining traffic accident behavior;

[0008] screening traffic accident search behavior from the traffic accident behavior and similar traffic accident similar behavior, querying a knowledge base according to the traffic accident search behavior, and obtaining and pushing traffic violation behavior corresponding to the traffic accident search behavior, corresponding traffic law and corresponding investigation operation.

[0009] Before the traffic accident image is input into the target detection model, the method further comprises:

[0010] According to the traffic accident history image including the traffic prohibition sign, the interference image generated from the traffic accident history image, the traffic accident history image including the non-traffic prohibition sign, and the interference image generated from the traffic accident history image including the non-traffic prohibition sign, a three-dimensional attention module in a YOLOv8 model is input, the three-dimensional attention module obtains color features of the traffic prohibition sign, and difference features of the traffic prohibition sign and the non-traffic prohibition sign;

[0011] The output data of the three-dimensional attention module is input into a prediction head in the YOLOv8 model, the prediction head includes a sign type target probability and a sign bounding box probability, and the target detection model is obtained by training a loss function of the YOLOv8 model, the loss function includes a sign type target probability loss function and a sign bounding box loss function.

[0012] The traffic prohibition sign in the traffic accident history image includes a partial sign bounding box;

[0013] The interference image is an image processed by data enhancement on the traffic accident history, and the data enhancement processing includes one or more of the following: dirt, inclination angle, and size adjustment;

[0014] The non-traffic prohibition sign includes one or more of the following: indication, warning, road sign, tourist area sign, notice sign, and auxiliary sign.

[0015] The subgraph of the corresponding traffic prohibition sign in the traffic accident image is intercepted, the traffic prohibition corresponding to the traffic prohibition sign is identified in the subgraph of the traffic prohibition sign, the traffic accident text corresponding to the traffic accident image is analyzed according to the traffic prohibition, and the traffic accident behavior is determined, including:

[0016] The subgraph of the corresponding traffic prohibition sign is intercepted in the traffic accident image according to the bounding box of the traffic prohibition sign, and the number of the subgraph is the same as the number of the traffic prohibition sign;

[0017] According to the number of traffic prohibition signs, the traffic prohibition corresponding to the traffic prohibition sign is identified in the subgraph of the traffic prohibition sign by using a graph-text multi-modal large model;

[0018] The traffic accident text corresponding to the traffic accident image is analyzed by using a large language model according to the traffic prohibition as a prompt word, and the traffic accident behavior of each party in the traffic accident image is determined.

[0019] The traffic accident retrieval behavior is screened out from the traffic accident behavior and the traffic accident similar behavior similar to the traffic accident behavior, including:

[0020] convert the traffic accident behavior into a traffic accident behavior vector, and obtain the traffic accident similar behavior based on the traffic accident behavior vector in the standard traffic violation behavior and the general traffic violation behavior in the knowledge base;

[0021] screen traffic accident retrieval behaviors from the traffic accident behavior and the traffic accident similar behavior by using screening prompt words in an input large language model.

[0022] The method further comprises:

[0023] analyzing original violation behaviors corresponding to traffic regulations and traffic law articles by a large language model, and identifying the standard traffic violation behavior corresponding to the original violation behaviors;

[0024] expanding the standard traffic violation behavior by using keywords in the standard traffic violation behavior, and extracting traffic violation behaviors corresponding to the traffic law articles from vehicle claim records to obtain the general traffic violation behavior;

[0025] constructing a knowledge base with the standard traffic violation behavior, the general traffic violation behavior, traffic law articles corresponding to the standard traffic violation behavior and the general traffic violation behavior, and investigation operations corresponding to the traffic law articles.

[0026] The method further comprises:

[0027] querying the knowledge base by using the traffic accident retrieval behavior vector to obtain traffic violation behaviors, traffic law articles, and investigation operations corresponding to the traffic accident query behavior in the knowledge base;

[0028] push the traffic violation behaviors, the traffic law articles, and the investigation operations corresponding to the traffic accident query behavior to a mobile terminal to identify the traffic violation behaviors, the traffic law articles, and the investigation operations by colors on the mobile terminal.

[0029] According to a second aspect of the embodiments of the present application, a device for identifying traffic accident behaviors is provided, comprising:

[0030] a recognition module configured to input a traffic accident image into a target detection model, and identify traffic prohibition signs and the number of the traffic prohibition signs in the traffic accident image according to color features of the traffic prohibition signs and difference features between the traffic prohibition signs and non-traffic prohibition signs.

[0031] The behavior module is configured to capture a sub-image of the traffic prohibition sign corresponding to the number in the traffic accident image, identify the traffic prohibition corresponding to the traffic prohibition sign in the sub-image of the traffic prohibition sign, analyze the traffic accident text corresponding to the traffic accident image according to the traffic prohibition, and determine the traffic accident behavior.

[0032] The retrieval module is configured to filter traffic accident retrieval behaviors from the traffic accident behaviors and similar behaviors of similar traffic accidents, query a knowledge base according to the traffic accident retrieval behaviors, and obtain and push traffic violation behaviors, corresponding traffic laws, and corresponding investigation operations corresponding to the traffic accident retrieval behaviors.

[0033] According to a third aspect of an embodiment of the present application, an electronic device for identifying traffic accident behaviors is provided, comprising:

[0034] one or more processors;

[0035] a storage device configured to store one or more programs,

[0036] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0037] According to a fourth aspect of an embodiment of the present application, a computer readable medium having a computer program stored thereon is provided, the program being executed by a processor to implement the method as described above.

[0038] An embodiment of the above application has the following advantages or beneficial effects: inputting a traffic accident image into a target detection model, identifying a traffic prohibition sign in the traffic accident image and a number of the traffic prohibition signs; capturing a sub-image of the traffic prohibition sign corresponding to the number in the traffic accident image, determining a traffic prohibition according to the number of the traffic prohibition signs, analyzing a traffic accident text corresponding to the traffic accident image according to the traffic prohibition, and determining a traffic accident behavior; and obtaining and pushing traffic violation behaviors, corresponding traffic laws, and corresponding investigation operations corresponding to the traffic accident retrieval behaviors by querying a knowledge base. The above operations do not require manual intervention, fully consider the role of traffic prohibition signs, can shorten the time for identifying traffic accident behaviors, and improve the identification accuracy.

[0039] Further effects of the above non-conventional optional mode will be described in conjunction with the specific embodiments below. BRIEF DESCRIPTION OF DRAWINGS

[0040] The accompanying drawings serve to better understand the present application and do not constitute an improper limitation thereof. Among them:

[0041] Figure 1is a main flowchart of a method for identifying traffic accident behaviors according to an embodiment of the present application;

[0042] Figure 2 is a flowchart of training a target detection model according to an embodiment of the present application;

[0043] Figure 3 is a flowchart of analyzing traffic accident texts corresponding to traffic accident images according to an embodiment of the present application;

[0044] Figure 4 is a flowchart of screening traffic accident retrieval behaviors according to an embodiment of the present application;

[0045] Figure 5 is a flowchart of constructing a knowledge base according to an embodiment of the present application;

[0046] Figure 6 is a flowchart of identifying traffic accident query behaviors according to an embodiment of the present application;

[0047] Figure 7 is a main structure diagram of a device for identifying traffic accident behaviors according to an embodiment of the present application;

[0048] Figure 8 is an exemplary system architecture diagram to which embodiments of the present application can be applied;

[0049] Figure 9 is a structure diagram of a computer system of a terminal device or a server suitable for implementing embodiments of the present application. DETAILED DESCRIPTION

[0050] Exemplary embodiments of the present application are described herein with reference to the accompanying drawings, which are meant to be exemplary and not limiting. Therefore, it should be recognized that many modifications and variations of the disclosed embodiments can be made by those of ordinary skill in the art without departing from the scope and spirit of the application. Also, for the sake of brevity and clarity, descriptions of well-known functions and constructions are omitted from the following description.

[0051] For the responsibility identification of traffic accident behaviors, subjective interpretation of accident image elements is relied on artificial experience, and there is a lack of systematic, orderly and retrievable mapping mechanism of "illegal behaviors-legal basis-investigation actions", which directly leads to a lack of legal basis for the responsibility identification conclusion and a lack of direction in the investigation process, and finally results in poor accuracy of responsibility identification, high compensation amount and frequent controversial cases. Secondly, there is a lack of unified knowledge base and effective reasoning, and it is difficult to generate objective and compliant investigation guidelines.

[0052] In comparison, there are some technical solutions in the current industry, but there are still significant limitations, and they have not really solved the fine, adversarial and strategy generation needs in the vehicle insurance liability scene.

[0053] In summary, in the face of a large number of traffic accident behaviors, the efficiency bottleneck of manual processing is highlighted.

[0054] In order to improve the processing efficiency of a large number of traffic accident behaviors, the following technical solutions in the embodiments of the present application can be used.

[0055] Referring to Figure 1 , Figure 1 is the main flowchart of the method for identifying traffic accident behaviors according to the embodiments of the present application, and the traffic violation behaviors and investigation operations are determined by identifying the traffic prohibition signs in the traffic accident images. As Figure 1 That is, 100 specifically includes the following steps:

[0056] S101, input the traffic accident image into a target detection model, and the target detection model identifies the traffic prohibition signs and the number of traffic prohibition signs in the traffic accident image according to the color features of the traffic prohibition signs and the difference features between the traffic prohibition signs and non-traffic prohibition signs.

[0057] In the practice process of the embodiments of the present application, only analyzing the traffic accident scene pictures cannot obtain accurate and complete traffic accident description. In the actual traffic accident scene investigation process, the investigation personnel will give a traffic accident description and take pictures of the traffic accident scene.

[0058] In the subsequent traffic accident vehicle insurance claim settlement process, the claim settlement personnel are limited by the manual recognition efficiency and the attention blind area, and often miss the traffic prohibition signs implied in the photos, which ultimately leads to the deviation of the responsibility identification and brings the claim loss that can be avoided. It can be seen that the traffic prohibition sign is an important basis for identifying traffic accident behaviors.

[0059] In the embodiments of the present application, the traffic signs include traffic prohibition signs, indication signs, warning signs, direction signs, tourist area signs, notice signs and auxiliary signs. As an example, the traffic prohibition signs include left turn prohibition, straight ahead prohibition, or U-turn prohibition.

[0060] Considering that the main color of the traffic prohibition sign is different from that of other traffic signs, the traffic signs can be divided into two categories: traffic prohibition signs and non-traffic prohibition signs. The non-traffic prohibition signs include one or more of the following: indication signs, warning signs, direction signs, tourist area signs, notice signs and auxiliary signs.

[0061] For the application scenario of traffic accidents, a target detection model can be used to detect traffic accident images, identify traffic violation signs and the number of traffic violation signs in the traffic accident images according to color features of the traffic violation signs and difference features between the traffic violation signs and non-traffic violation signs, and then determine whether the traffic accident behavior is illegal according to the traffic violation signs and the number of traffic violation signs.

[0062] In an embodiment of the present application, a target detection model is used to realize automatic sorting of traffic accident images. The purpose of automatic sorting is to quickly obtain traffic accident images with traffic violation signs to obtain traffic violations by using traffic violation signs in the traffic accident images.

[0063] By constructing a target detection model, traffic violation signs in traffic accident images are identified, and the identification accuracy of traffic violation behaviors that do not drive according to traffic violation signs is significantly improved.

[0064] Referring to Figure 2 , Figure 2 is a flowchart of training a target detection model according to an embodiment of the present application. Specifically, the following steps are included:

[0065] S201, inputting a three-dimensional attention module in a YOLOv8 model according to traffic accident historical images including traffic violation signs, interference images generated from the traffic accident historical images, traffic accident historical images including non-traffic violation signs, and interference images generated from the traffic accident historical images including non-traffic violation signs, the three-dimensional attention module obtaining color features of the traffic violation signs and difference features between the traffic violation signs and the non-traffic violation signs.

[0066] Existing target detection models have high error rates when detecting traffic violation signs, and the core reason is that traffic accident images often have complex interference factors, and different types of traffic violation signs have high visual feature similarity and only pattern detail differences. Class confusion is easy to occur. For example, a left turn prohibition is misjudged as a U-turn prohibition.

[0067] In an embodiment of the present application, the role of the target detection model is to identify traffic violation signs and the number of traffic violation signs. The target detection model is trained based on a YOLOv8 model. As an example, the target detection model can be trained based on other YOLO models.

[0068] The YOLOv8 model adopts a backbone network-neck network-prediction head three-order architecture: the backbone network uses a C2f module to strengthen the extraction of small target feature through branch convolution-feature splicing; the neck network is a FPN-PAN double-path structure, which realizes the fusion of high-level semantic and low-level detail features; the prediction head adopts an anchor-free design, and directly outputs target coordinates and class probability by using an EfficientRepulsionHead, thereby avoiding preset deviation of the anchor frame.

[0069] In the embodiment of the present application, the training data specifically includes traffic accident history images including traffic prohibition signs, interference images generated from the traffic accident history images, traffic accident history images including non-traffic prohibition signs, and interference images generated from the traffic accident history images including non-traffic prohibition signs. The target detection model is obtained by training the YOLOv8 model using the above training data.

[0070] The training data is exemplarily described below.

[0071] The traffic prohibition sign in the traffic accident history image includes a partial sign boundary box. As an example, the partial sign boundary box is a part of the complete sign boundary box, to simulate the scenario that the traffic prohibition sign is blocked.

[0072] The boundary box of the non-traffic prohibition sign in the traffic accident history image including the non-traffic prohibition sign can also be incomplete, to simulate the application scenario that multiple signs are blocked.

[0073] The traffic accident history images including traffic prohibition signs and the traffic accident history images including non-traffic prohibition signs can improve the accuracy of the target detection model in identifying the number of traffic prohibition signs. The categories of the non-traffic prohibition signs are labeled, such as indication signs, warning signs, direction signs, tourist area signs, notice signs, and auxiliary signs, to avoid interference of the non-traffic prohibition signs with the existence determination.

[0074] The interference image is an image processed by data enhancement of the traffic accident history data, and the data enhancement includes one or more of the following: dirt, inclination angle, and size adjustment.

[0075] The interference scene often occurs in the traffic accident image, such as traffic sign surface dirt, shooting angle inclination leading to traffic sign edge deformation, and background vehicle / pedestrian blocking the key area of the traffic sign. The interference image can be generated by enhancing the traffic accident history data.

[0076] The interference image is used to train the robust recognition ability of the target detection model for the core visual contour of the traffic prohibition sign. That is, whether the details of the traffic prohibition sign are complete or not, the contour feature can be used to determine whether it is a traffic prohibition sign.

[0077] In the embodiments of the present application, the interference images generated by the traffic accident history images, and the interference images generated by the traffic accident history images including non-traffic prohibition signs, that is, including traffic prohibition signs and non-traffic prohibition signs, are used to train the target detection model to distinguish the traffic prohibition signs from the non-traffic prohibition signs.

[0078] In the embodiments of the present application, the backbone network and the prediction head in the YOLOv8 model are adjusted in a targeted manner, the type classification weight is weakened, and the key task feature extraction is strengthened.

[0079] After the C2f module of YOLOv8, a three-dimensional attention module is added. The role of the three-dimensional attention module is to identify the differences of traffic signs from color, shape and category. The calculation formula of the three-dimensional attention module is shown in formula 1:

[0080]

[0081] Formula 1

[0082] In formula 1, represents the color and shape attention weight, focusing on the core visual features of the traffic prohibition sign. represents the category difference weight, focusing on multi-sign distinction, and solving non-prohibition interference. γ is a fusion coefficient, and σ(·) is a Sigmoid function. The role of the Sigmoid function is to normalize to the interval [0, 1].

[0083] For specific expansion, see formula 2:

[0084]

[0085] Formula 2

[0086] Wherein, is the color attention weight. For example, a higher weight is given to the red channel feature to highlight the red bottom feature. is the shape attention weight, and α is a balance coefficient (optimized to 0.6 through experiments). F is the feature map output by the backbone network. As an example, α is 0.6.

[0087] For specific expansion, see formula 3:

[0088]

[0089] Formula 3

[0090] Wherein, is the feature vector of the suspected traffic prohibition sign, is a standard feature vector of six types of non-traffic prohibition signs. The standard feature vectors of the six types of non-traffic prohibition signs are pre-stored in the model parameters, and the feature mean of each type of non-prohibition sign is calculated offline. is an L2 norm.

[0091] The three-dimensional attention module obtains the color features of the traffic prohibition sign and the difference features between the traffic prohibition sign and the non-traffic prohibition sign. That is, through the three-dimensional attention module, both the core visual features of the traffic prohibition sign and the comparison with the non-traffic prohibition sign can be captured in the feature extraction stage, and the interference of the non-prohibition sign can be reduced.

[0092] S202, the output data of the three-dimensional attention module is input into the prediction head of the YOLOv8 model, the prediction head includes sign type target probability and sign bounding box probability, and the target detection model is obtained by training using the loss function of the YOLOv8 model, the loss function includes a sign bounding box loss function and a sign type target probability loss function.

[0093] The prediction head of the YOLOv8 model includes sign type target probability and sign bounding box probability. The sign type target probability includes a multi-class classification branch and a binary classification branch.

[0094] The multi-class classification branch is used to output the probability that the traffic sign in the traffic accident image belongs to 8 types of targets. Among them, the 8 types of targets include background and 7 types of traffic signs. The multi-class classification branch can be implemented by using 3 layers of convolution layers and Softmax activation.

[0095] The binary classification branch can be generated by mapping the multi-class classification result. For example, class 1 is a traffic prohibition sign, and classes 2 to 7 correspond to 6 types of non-traffic prohibition signs. Class 1 is mapped to the traffic prohibition sign; classes 2 to 7 are mapped to the non-traffic prohibition sign. No additional convolution layer is needed to reduce the amount of calculation.

[0096] The bounding box regression probability is used to output the probability of the traffic prohibition sign bounding box in the traffic accident image. By using the anchor-free design of the YOLOv8 model, the center coordinates (x, y) of the traffic prohibition sign, the width and height (w, h) of the traffic prohibition sign, and the bounding box confidence of the traffic prohibition sign are output, which optimizes the positioning accuracy of the traffic prohibition sign and improves the number counting ability.

[0097] In addition, the total loss function of the YOLOv8 model combines the bounding box regression, multi-class classification, and existence judgment targets, as shown in formula 4:

[0098]

[0099] Formula 4

[0100] : bounding box loss function of sign, weight 1.0. Used to optimize traffic prohibition sign positioning

[0101] To improve the accuracy of the number of traffic prohibition signs.

[0102] The sign type target probability includes And .

[0103] : 8-class cross-entropy loss function, used to optimize the ability to distinguish multiple traffic signs. The 8-class cross-entropy loss is shown in formula 5:

[0104]

[0105] Formula 5

[0106] is the true label, is the predicted probability.

[0107] : binary cross-entropy loss function, to strengthen the existence determination. The binary cross-entropy loss is shown in formula 6.

[0108]

[0109] Formula 6

[0110] represents the true class label of the nth traffic sign candidate box, represents the confidence of the target detection model determining the nth traffic sign candidate box as a traffic prohibition sign.

[0111] In the embodiment of Figure 2 , the YOLOv8 model is trained using the training data to obtain the target detection model in the embodiment of the present application.

[0112] S102, a subgraph corresponding to the number of traffic prohibition signs is intercepted in the traffic accident image, and the traffic prohibition corresponding to the traffic prohibition sign is identified in the subgraph of the traffic prohibition sign. Analyze the traffic accident text corresponding to the traffic accident image according to the traffic prohibition to determine the traffic accident behavior.

[0113] In order to accurately identify the traffic prohibition corresponding to the traffic prohibition sign, avoid the interference of other objects in the traffic accident image, and intercept the subgraph corresponding to the number of traffic prohibition signs in the traffic accident image. The main object in the subgraph is the traffic prohibition sign, and the traffic prohibition corresponding to the traffic prohibition sign is identified in the subgraph of the traffic prohibition sign. According to the traffic prohibition, analyze the traffic accident text corresponding to the traffic accident image to determine the traffic accident behavior.

[0114] Referring to Figure 3 ,Figure 3 is a flowchart of analyzing traffic accident text corresponding to a traffic accident image according to an embodiment of the present application. Specifically, it includes the following steps:

[0115] S301, in the traffic accident image, the subgraphs corresponding to the traffic prohibition signs are intercepted according to the bounding boxes of the traffic prohibition signs, and the number of subgraphs is the same as the number of traffic prohibition signs.

[0116] There are corresponding traffic prohibition signs in the traffic accident image. Each traffic prohibition sign has a corresponding traffic prohibition. As an example, the traffic prohibition sign: no left turn, the corresponding traffic prohibition is: no left turn.

[0117] In order to improve the accuracy of identifying traffic prohibitions, analysis can be performed in subgraphs that only include traffic prohibition signs. In the case of identifying traffic prohibition signs in a traffic accident image, subgraphs corresponding to traffic prohibition signs can be intercepted according to the bounding boxes of traffic prohibition signs. How many traffic prohibition signs, then intercept the corresponding number of subgraphs. Each traffic prohibition sign has a corresponding subgraph.

[0118] S302, using a graph-text multimodal large model, in the traffic prohibition sign subgraph, according to the number of traffic prohibition signs, identifying the traffic prohibition corresponding to the traffic prohibition sign.

[0119] In an embodiment of the present application, the graph-text multimodal large model is used to identify the traffic prohibition corresponding to the traffic prohibition sign. The traffic prohibition sign subgraph is used to remind the graph-text multimodal large model of the existence of traffic prohibition identification. The graph-text multimodal large model uses the number of traffic prohibition signs to output a corresponding number of traffic prohibitions, thereby improving the accuracy of the graph-text multimodal large model in identifying traffic prohibitions.

[0120] As an example, the number of traffic prohibition signs is 2, and the graph-text multimodal large model is used in the subgraphs of the 2 traffic prohibition signs to identify the traffic prohibition corresponding to the traffic prohibition sign as: truck prohibited and speed limit 60.

[0121] S303, using the traffic prohibition as a prompt word, using a large language model to analyze the traffic accident text corresponding to the traffic accident image, and determining the traffic accident behavior of each party in the traffic accident image.

[0122] Traffic prohibitions can be used as a basis for judgment to analyze the traffic accident behavior corresponding to the traffic accident image. In order to improve the speed of analyzing the traffic accident image, the traffic prohibition is used as the prompt word of the large language model to analyze the traffic accident text corresponding to the traffic accident image. The analysis result of the large language model is the traffic accident behavior of each party in the traffic accident image.

[0123] The traffic accident text is a text of the occurrence of a traffic accident established by collecting information of a traffic accident scene. The traffic accident text and the traffic accident image can be obtained from a database of a server side. As an example, the database is a vehicle insurance claim system.

[0124] As an example, the traffic accident text is: On July 4, 2025, at about 5:40, a traffic accident occurred at an intersection. The parties involved: the target small passenger car was driving straight and collided with the third party truck. The moment of collision: the front of the target motor vehicle and the left side of the third motor vehicle were in contact.

[0125] The traffic ban sign corresponding to the traffic ban sign is identified by using a multi-modal large model: truck ban.

[0126] The large language model determines the traffic accident behavior corresponding to the traffic accident image: the third party truck may have a “violation of road traffic restrictions” behavior; the target small passenger car may have “excessive speed” and “not maintaining a safe distance” behaviors.

[0127] In Figure 3 Embodiments, the subgraph of the traffic ban sign is used to reduce the interference of other objects in the traffic accident image to accurately identify the traffic ban, and then automatically analyze the traffic accident text.

[0128] S103, screen out traffic accident retrieval behaviors from traffic accident behaviors and traffic accident similar behaviors similar to the traffic accident behaviors, query the knowledge base with the traffic accident retrieval behaviors, and identify and push the traffic violation behaviors corresponding to the traffic accident query behaviors and the corresponding investigation operations.

[0129] In embodiments of the present application, in order to improve the accuracy of identifying traffic violation behaviors, the number of traffic accident behaviors can be increased by using traffic accident similar behaviors similar to the traffic accident behaviors, and then the traffic accident retrieval behaviors are screened out in a larger range. The traffic accident retrieval behaviors are used as the basis for querying traffic violation behaviors.

[0130] Referring to Figure 4 , Figure 4 is a flowchart of screening out traffic accident retrieval behaviors according to embodiments of the present application. Specifically, the following steps are included:

[0131] S401, convert the traffic accident behavior into a traffic accident behavior vector, and obtain traffic accident similar behaviors based on the vector of the traffic accident behavior in the standard traffic violation behaviors and the general traffic violation behaviors in the knowledge base.

[0132] In embodiments of the present application, in order to accurately identify traffic violation behaviors, traffic accident similar behaviors of the traffic accident behavior are obtained in the knowledge base.

[0133] The traffic violation behaviors in the knowledge base are stored in a vector manner, and the traffic accident behaviors can be converted into traffic accident behavior vectors by using a vector conversion model. As an example, the vector conversion model includes BERT, RoBERTa or ChatGLM.

[0134] The standard traffic violation behaviors and the general traffic violation behaviors are stored in the knowledge base. The standard traffic violation behaviors are behaviors determined according to the traffic accident subject's violation behaviors and traffic law articles. The general traffic violation behaviors are behaviors expanded according to the similarity of the standard traffic violation behaviors.

[0135] Based on the similarity between the vector of the traffic accident behavior and the vector of the standard traffic violation behavior, and the similarity between the vector of the traffic accident behavior and the vector of the general traffic violation behavior, the traffic accident similar behavior is obtained.

[0136] S402, using the screening prompt word in the input large language model, screening the traffic accident retrieval behavior from the traffic accident behavior and the traffic accident similar behavior.

[0137] In the embodiments of the present application, a large language model is used to screen the traffic accident retrieval behavior from the traffic accident behavior and the traffic accident similar behavior. Specifically, the screening prompt word determined based on the traffic accident image is received to be screened by the large language model. As an example, the screening prompt word includes object names and / or scene names in the traffic accident image, such as: object names: car and pedestrian; scene name: intersection.

[0138] In the embodiments of the present application, a large language model is used to screen the traffic accident retrieval behavior from the traffic accident behavior and the traffic accident similar behavior. Specifically, the screening prompt word determined based on the traffic accident image is received to be screened by the large language model. As an example, the screening prompt word includes object names and / or scene names in the traffic accident image, such as: object names: car and pedestrian; scene name: intersection. Figure 4 In the embodiments of the present application, a large language model is used to screen the traffic accident retrieval behavior from the traffic accident behavior and the traffic accident similar behavior. Specifically, the screening prompt word determined based on the traffic accident image is received to be screened by the large language model. As an example, the screening prompt word includes object names and / or scene names in the traffic accident image, such as: object names: car and pedestrian; scene name: intersection.

[0139] Figure 5 Figure 5 is a flowchart of constructing a knowledge base according to an embodiment of the present application. Specifically, it includes the following steps:

[0140] S501, analyzing the original violation behavior corresponding to the traffic regulations and the traffic law articles by the large language model, and identifying the standard traffic violation behavior corresponding to the original violation behavior.

[0141] Based on the road traffic safety regulations, regulations and implementation measures, the traffic violation behaviors involved by the traffic accident subject and the traffic law articles corresponding to the traffic violation behaviors are obtained by the large language model. Combined with the experience of business experts, the standard traffic violation behaviors and the traffic law articles corresponding to the standard traffic violation behaviors are formed based on the above traffic violation behaviors and corresponding traffic law articles.

[0142] ​​S502, expand the standard traffic violation behavior by using the keywords in the standard traffic violation behavior, and extract the traffic violation behavior corresponding to the traffic law in the vehicle claim record to obtain the general traffic violation behavior.

[0143] In the embodiments of the application, the keywords can be used for expansion, and the vehicle claim record can also be used for expansion, and the general traffic violation behavior is obtained by the above two ways.

[0144] As an example, the keywords in the standard traffic violation behavior are input into the large language model to obtain the general traffic violation behavior. Among them, the keywords in the standard traffic violation behavior can be determined based on expert experience.

[0145] For example, the standard traffic violation behavior: when the motor vehicle changes lanes, it affects the normal driving of other vehicles. The keywords: motor vehicle lane change are input into the large language model, and the large language model outputs: motor vehicle lane change affects the driving of other vehicles; motor vehicle lane change without careful observation.

[0146] In addition, the vehicle claim record is obtained from the vehicle insurance claim system, and the traffic violation behavior corresponding to the traffic law is extracted from the vehicle claim record.

[0147] For example, the scene vehicle violation situation recorded by the surveyor in the claim case is used as the vehicle claim record. The vehicle claim record includes: the target vehicle drives at a high speed and causes a rear-end collision, and the three vehicle drivers cross the solid line to change lanes.

[0148] By using the large language model, the traffic violation behavior corresponding to the traffic law is extracted from the vehicle claim record by using the prompt word. The function of the prompt word is to inform the large model to extract the traffic violation behavior of each subject in the vehicle claim record. For example, the traffic violation behavior: driving at a high speed and crossing the solid line to change lanes. In order to ensure the effectiveness of the above traffic violation behavior, a business expert can be used for verification.

[0149] S503, using the standard traffic violation behavior, the general traffic violation behavior, the traffic law corresponding to the standard traffic violation behavior and the general traffic violation behavior, and the investigation operation corresponding to the traffic law, to construct a knowledge base.

[0150] The knowledge base includes the standard traffic violation behavior and the general traffic violation behavior, the traffic law corresponding to the standard traffic violation behavior and the general traffic violation behavior, and the investigation operation corresponding to the traffic law.

[0151] In addition, in order to improve the retrieval speed of the knowledge base, the data is stored in the knowledge base in the form of vectors.

[0152] The vectorization of the following data can be implemented by using a language model: standard traffic violations and general traffic violations, traffic law articles corresponding to the standard traffic violations and general traffic violations, and investigation operations corresponding to the traffic law articles. As an example, the language model is BERT, RoBERTa, or ChatGLM.

[0153] For the traffic law articles corresponding to the traffic violations, corresponding investigation operations are set. As an example, the investigation operations include: accident parties and accident operations. For example: traffic law article: rear-end collision. Accident party: rear vehicle; accident operation: check for drunk driving. Accident party: front vehicle; accident operation: check whether to change lanes.

[0154] In the embodiments of Figure 5 , on the basis of the standard traffic violations, the knowledge base is expanded to increase the accuracy of the traffic accident behaviors.

[0155] Referring to Figure 6 , Figure 6 is a flowchart of identifying traffic accident query behaviors according to an embodiment of the present application. Specifically, the following steps are included:

[0156] S601, using the vector of the traffic accident retrieval behavior as a query to the knowledge base, obtaining the traffic violations corresponding to the traffic accident query behavior, the traffic law articles corresponding to the traffic accident query behavior, and the investigation operations corresponding to the traffic accident query behavior in the knowledge base.

[0157] In order to accurately identify traffic accident behaviors, the knowledge base is used to implement the retrieval of traffic accident retrieval behaviors. The number of traffic accident behaviors is multiple, and each traffic accident behavior is queried. In the knowledge base, the traffic violations corresponding to the traffic accident query behavior, the traffic law articles corresponding to the traffic accident query behavior, and the investigation operations corresponding to the traffic accident query behavior are obtained. Using multiple traffic accident behaviors, multiple identification results of traffic accident behaviors are provided.

[0158] S602, pushing the traffic violations corresponding to the traffic accident query behavior, the traffic law articles corresponding to the traffic accident query behavior, and the investigation operations corresponding to the traffic accident query behavior to the mobile terminal, so as to identify the traffic violations, the traffic law articles, and the investigation operations by color on the mobile terminal.

[0159] In order to facilitate the traffic accident handling personnel to obtain information in time, the traffic violations corresponding to the traffic accident query behavior, the traffic law articles corresponding to the traffic accident query behavior, and the investigation operations corresponding to the traffic accident query behavior can be pushed to the mobile terminal. As an example, the application installed on the mobile terminal is pushed.

[0160] Moreover, the traffic violations, the traffic law articles, and the investigation operations can be identified by different colors in the mobile terminal. As an example, the traffic law articles are identified by red; the traffic violations are identified by yellow; and the investigation operations are identified by green.

[0161] Traffic violations, traffic laws and investigation operations are distinguished by colors to avoid misoperation on the mobile terminal.

[0162] In Figure 6 In the embodiment, the traffic accident retrieval behavior corresponding traffic laws and investigation operations are obtained by using the knowledge base to improve the acquisition speed.

[0163] In the above embodiment of the application, the traffic accident image is input into a target detection model to identify traffic prohibition signs and the number of traffic prohibition signs in the traffic accident image; a subgraph of the number of traffic prohibition signs in the traffic accident image is intercepted to determine a traffic prohibition; the traffic accident text corresponding to the traffic accident image is analyzed by the traffic prohibition to determine a traffic accident behavior; and the traffic accident retrieval behavior corresponding traffic violations, corresponding traffic laws and corresponding investigation operations are obtained and pushed by querying the knowledge base. The above operation does not require manual intervention, fully considers the role of traffic prohibition signs, can shorten the time consumption of identifying traffic accident behaviors, and improve the identification accuracy.

[0164] In the embodiment of the application, the intelligent and standardized level of vehicle insurance liability is significantly improved, the dependence on manual work and subjective bias is effectively reduced, and the objectivity and legal compliance of liability determination are enhanced. Thus, the liability risk is avoided, the customer experience is optimized, and the customer rights and interests are protected. At the same time, the output data can be used as an auxiliary technical reference for traffic accident liability identification, the evidence support and logical rigor of the liability determination process are enhanced, and the traffic accident processing is upgraded from experience judgment to data-driven and rule-oriented scientific decision-making mode.

[0165] Referring to Figure 7 , Figure 7 is a main structure schematic diagram of the device for identifying traffic accident behaviors according to the embodiment of the application. The device for identifying traffic accident behaviors can implement the method for identifying traffic accident behaviors, such as Figure 7 As shown in 700, the device for identifying traffic accident behaviors specifically includes:

[0166] The identification module 701 is configured to input a traffic accident image into a target detection model. The target detection model identifies traffic prohibition signs and the number of traffic prohibition signs in the traffic accident image according to color features of the traffic prohibition signs and difference features of the traffic prohibition signs and non-traffic prohibition signs.

[0167] The behavior module 702 is configured to intercept a subgraph of the number of traffic prohibition signs in the traffic accident image, identify traffic prohibition signs corresponding to the traffic prohibition signs in the subgraph of the traffic prohibition signs, analyze traffic accident text corresponding to the traffic accident image by the traffic prohibition, and determine a traffic accident behavior.

[0168] The retrieval module 703 is configured to filter out a traffic accident retrieval behavior from the traffic accident behavior and a traffic accident similar behavior similar to the traffic accident behavior, query a knowledge base with the traffic accident retrieval behavior, and obtain and push a corresponding traffic violation behavior, a corresponding traffic law, and a corresponding investigation operation corresponding to the traffic accident retrieval behavior.

[0169] In an embodiment of the present application, the identification module 701 is configured to input a three-dimensional attention module in a YOLOv8 model according to a traffic accident historical image including a traffic prohibition sign, a disturbance image generated from the traffic accident historical image, a traffic accident historical image including a non-traffic prohibition sign, and a disturbance image generated from the traffic accident historical image including the non-traffic prohibition sign, the three-dimensional attention module obtains color features of the traffic prohibition sign, and difference features between the traffic prohibition sign and the non-traffic prohibition sign.

[0170] The output data of the three-dimensional attention module is input into a prediction head in the YOLOv8 model, the prediction head includes a sign type target probability and a sign bounding box probability, and the target detection model is obtained by training a loss function of the YOLOv8 model, the loss function includes a sign type target probability loss function and a sign bounding box loss function.

[0171] In an embodiment of the present application, the traffic prohibition sign in the traffic accident historical image includes a partial sign bounding box.

[0172] The disturbance image is an image processed by data enhancement on the traffic accident historical image, and the data enhancement includes one or more of the following: dirt, inclination angle, and size adjustment.

[0173] The non-traffic prohibition sign includes one or more of the following: an indication sign, a warning sign, a road sign, a tourist area sign, a notice sign, and an auxiliary sign.

[0174] In an embodiment of the present application, the behavior module 702 is configured to intercept a subgraph of the corresponding traffic prohibition sign in the traffic accident image according to the bounding box of the traffic prohibition sign, and the number of the subgraph is the same as the number of the traffic prohibition sign.

[0175] A graph-text multi-modal large model is used to identify the corresponding traffic prohibition in the subgraph of the traffic prohibition sign according to the number of the traffic prohibition sign.

[0176] The traffic accident text corresponding to the traffic accident image is analyzed by using a large language model with the traffic prohibition as a prompt word, and the traffic accident behaviors of each party in the traffic accident image are determined.

[0177] In an embodiment of the present application, the retrieval module 703 is configured to convert the traffic accident behavior into a traffic accident behavior vector, and obtain the traffic accident similar behavior based on the traffic accident behavior vector in the standard traffic violation behavior and the general traffic violation behavior in the knowledge base;

[0178] The screening prompt word in the input large language model is used to screen the traffic accident retrieval behavior from the traffic accident behavior and the traffic accident similar behavior.

[0179] In an embodiment of the present application, the retrieval module 703 is configured to analyze the original violation behavior corresponding to the traffic regulation and the traffic regulation by a large language model, and identify the standard traffic violation behavior corresponding to the original violation behavior;

[0180] The key words in the standard traffic violation behavior are used to expand the standard traffic violation behavior, and the traffic violation behavior corresponding to the traffic regulation is extracted from the vehicle claim record to obtain the general traffic violation behavior;

[0181] The standard traffic violation behavior, the general traffic violation behavior, the traffic regulation corresponding to the standard traffic violation behavior and the general traffic violation behavior, and the investigation operation corresponding to the traffic regulation are used to construct a knowledge base.

[0182] In an embodiment of the present application, the retrieval module 703 is configured to query the knowledge base with the traffic accident retrieval behavior vector to obtain the traffic violation behavior corresponding to the traffic accident query behavior, the traffic regulation corresponding to the traffic accident query behavior and the investigation operation corresponding to the traffic accident query behavior;

[0183] The traffic violation behavior corresponding to the traffic accident query behavior, the traffic regulation corresponding to the traffic accident query behavior and the investigation operation corresponding to the traffic accident query behavior are pushed to the mobile terminal to identify the traffic violation behavior, the traffic regulation and the investigation operation by color on the mobile terminal.

[0184] Figure 8 An exemplary system architecture 800 is shown, which can apply the method for identifying traffic accident behavior or the device for identifying traffic accident behavior according to the embodiments of the present application.

[0185] As shown in Figure 8 The system architecture 800 can include terminal devices 801, 802, 803, a network 804 and a server 805. The network 804 is used as a medium to provide a communication link between the terminal devices 801, 802, 803 and the server 805. The network 804 can include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0186] The user can use the terminal devices 801, 802, 803 to interact with the server 805 through the network 804 to receive or send messages and the like. Various communication client applications can be installed on the terminal devices 801, 802, 803, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, and the like (only as examples).

[0187] The terminal devices 801, 802, 803 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0188] The server 805 can be a server providing various services, such as a background management server providing support for a user browsing a shopping website using the terminal device 801, 802, 803 (only as an example). The background management server can analyze and process received product information query requests and the like, and feed back the processing results (such as target push information, product information - only as examples) to the terminal device.

[0189] It should be noted that the method for identifying traffic accident behaviors provided by the embodiments of the present application is generally executed by the server 805, and correspondingly, the device for identifying traffic accident behaviors is generally arranged in the server 805.

[0190] It should be understood that Figure 8 The number of terminal devices, networks, and servers in

[0191] Reference is made below to Figure 9 which shows a structural schematic diagram of a computer system 900 of a terminal device suitable for use to implement the embodiments of the present application. Figure 9 The terminal device shown is only an example and should not bring any limitation to the functions and use range of the embodiments of the present application.

[0192] As shown in Figure 9 , the computer system 900 includes a central processing unit (CPU) 901 which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 902 or loaded from a storage portion 908 into a random access memory (RAM) 903. Various programs and data required for the operation of the system 900 are also stored in the RAM 903. The CPU 901, the ROM 902, and the RAM 903 are connected to each other through a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0193] The following components are connected to the I / O interface 905: an input part 906 including a keyboard, a mouse, etc.; an output part 907 including a display such as a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage part 908 including a hard disk, etc.; and a communication part 909 including a network interface card such as a LAN card, a modem, etc. The communication part 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as necessary. A removable medium 911 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is attached to the drive 910 as necessary, so that a computer program read out therefrom is installed in the storage part 908 as necessary.

[0194] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for executing the methods illustrated by the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the central processing unit (CPU) 901, the above-described functions defined in the system of the present disclosure are executed.

[0195] It should be noted that the computer-readable medium shown in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or component. In the present application, the computer-readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or component. The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0196] The flowcharts and block diagrams in the drawings illustrate the possible implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a program segment or a part of code containing one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur in different order than that shown in the drawings. For example, two blocks that are shown in succession can actually be executed substantially in parallel, and sometimes in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams or flowcharts, and the combination of blocks in the block diagrams or flowcharts, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0197] The modules described in the embodiments of the present application can be implemented in software or hardware. The described modules can also be arranged in a processor, for example, a processor can be described as including an identification module 701, a behavior module 702, and a retrieval module 703. In some cases, the names of these modules do not constitute a limitation on the modules themselves, for example, the identification module can also be described as "a module for inputting a traffic accident image into a target detection model, the target detection model identifies traffic violation signs and the number of traffic violation signs in the traffic accident image according to color features of the traffic violation signs and difference features between the traffic violation signs and non-traffic violation signs".

[0198] As another aspect, the present application also provides a computer readable medium, which can be included in the device described in the above embodiments, or can exist independently without being assembled into the device. The computer readable medium carries one or more programs, which, when executed by the device, cause the device to include:

[0199] inputting a traffic accident image into a target detection model, the target detection model identifies traffic violation signs and the number of traffic violation signs in the traffic accident image according to color features of the traffic violation signs and difference features between the traffic violation signs and non-traffic violation signs;

[0200] cutting a sub-image of the number of traffic violation signs in the traffic accident image, identifying a traffic violation corresponding to the traffic violation signs in the sub-image of the traffic violation signs, analyzing a traffic accident text corresponding to the traffic accident image according to the traffic violation, and determining a traffic accident behavior;

[0201] screening a traffic accident retrieval behavior from the traffic accident behavior and a similar behavior of a similar traffic accident, querying a knowledge base according to the traffic accident retrieval behavior, and obtaining and pushing a traffic violation behavior, a corresponding traffic law, and a corresponding investigation operation corresponding to the traffic accident query behavior.

[0202] According to the technical scheme of the embodiment of the present application, the traffic accident image is input into the target detection model, the traffic prohibition sign in the traffic accident image and the number of the traffic prohibition sign are identified, the subgraph of the corresponding traffic prohibition sign of the number is intercepted in the traffic accident image to determine the traffic prohibition, the traffic accident text corresponding to the traffic accident image is analyzed according to the traffic prohibition, the traffic accident behavior is determined, the corresponding traffic violation behavior, the corresponding traffic law article and the corresponding investigation operation corresponding to the traffic accident query behavior are obtained and pushed by querying the knowledge base. The above operation does not need manual intervention, fully considers the role of the traffic prohibition sign, can shorten the time consumption of identifying the traffic accident behavior, and improves the identification accuracy.

[0203] The above specific embodiments do not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can occur depending on design requirements and other factors. Any modification, equivalent replacement and improvement made within the spirit and principles of the present application should be included in the protection scope of the present application. It should be noted that the acquisition, storage and application of user personal information in the technical scheme of the present disclosure comply with relevant laws and regulations and do not violate public order and good customs.

Claims

1. A method of identifying traffic accident behavior, characterized by, The method comprises the following steps: inputting a traffic accident image into a target detection model, the target detection model identifying a traffic prohibition sign and the number of the traffic prohibition sign in the traffic accident image according to color features of the traffic prohibition sign and difference features between the traffic prohibition sign and a non-traffic prohibition sign; cutting a sub-image of the corresponding traffic prohibition sign in the traffic accident image, identifying a corresponding traffic prohibition in the sub-image of the traffic prohibition sign, analyzing a traffic accident text corresponding to the traffic accident image according to the traffic prohibition, and determining a traffic accident behavior; screening a traffic accident retrieval behavior from the traffic accident behavior and a similar behavior of a similar traffic accident, querying a knowledge base according to the traffic accident retrieval behavior, and obtaining and pushing a corresponding traffic violation behavior, a corresponding traffic law and a corresponding investigation operation corresponding to the traffic accident retrieval behavior.

2. The method of claim 1, wherein the traffic incident behavior is identified based on the at least one of the following: Before the traffic accident image is input into the target detection model, the method further comprises the following steps: ​ inputting a traffic accident historical image including a traffic prohibition sign, an interference image generated from the traffic accident historical image, a traffic accident historical image including a non-traffic prohibition sign and an interference image generated from the traffic accident historical image including the non-traffic prohibition sign into a three-dimensional attention module of a YOLOv8 model, the three-dimensional attention module obtaining color features of the traffic prohibition sign and difference features between the traffic prohibition sign and the non-traffic prohibition sign; output data of the three-dimensional attention module is input into a prediction head of the YOLOv8 model, the prediction head including a sign type target probability and a sign bounding box probability, the target detection model is obtained by training a loss function of the YOLOv8 model, the loss function including a sign type target probability loss function and a sign bounding box loss function.

3. The method of claim 2, wherein the traffic incident behavior is identified by, The traffic prohibition sign in the traffic accident historical image includes a partial sign bounding box; The interference image is an image processed by data enhancement on the traffic accident historical image, the data enhancement includes one or more of the following: dirt, inclination angle and size adjustment; The non-traffic prohibition sign includes one or more of the following: indication, warning, direction, tourist area, notice and auxiliary.

4. The method of claim 1, wherein the traffic incident behavior is identified by: The method for determining the traffic accident behavior comprises the following steps: cutting a sub-image of the corresponding traffic prohibition sign in the traffic accident image according to a bounding box of the traffic prohibition sign, the number of the sub-image being the same as the number of the traffic prohibition sign; using a graphic-text multimodal large model to identify the corresponding traffic prohibition of the traffic prohibition sign in the sub-image of the traffic prohibition sign according to the number of the traffic prohibition sign; The traffic accident image is input into a target detection model, and the target detection model identifies a traffic prohibition sign and a number of the traffic prohibition sign in the traffic accident image according to color characteristics of the traffic prohibition sign and difference characteristics of the traffic prohibition sign and a non-traffic prohibition sign.

5. The method of claim 1, wherein the traffic incident behavior is identified by: The traffic accident behavior is converted into a traffic accident behavior vector, and the traffic accident similar behavior is obtained based on the traffic accident behavior vector among standard traffic illegal behaviors and general traffic illegal behaviors in the knowledge base; The traffic accident behavior is converted into a traffic accident behavior vector, and the traffic accident similar behavior is obtained based on the traffic accident behavior vector among standard traffic illegal behaviors and general traffic illegal behaviors in the knowledge base; The traffic accident behavior is converted into a traffic accident behavior vector, and the traffic accident similar behavior is obtained based on the traffic accident behavior vector among standard traffic illegal behaviors and general traffic illegal behaviors in the knowledge base; 6. The method of claim 1 or 5, wherein the traffic incident behavior is identified by, The method further comprises: The original illegal behaviors corresponding to the traffic regulations and the traffic law articles are analyzed by the large language model, and the standard traffic illegal behaviors corresponding to the original illegal behaviors are identified; The standard traffic illegal behaviors are expanded by using keywords in the standard traffic illegal behaviors, and the traffic illegal behaviors corresponding to the traffic law articles are extracted from vehicle claim records, to obtain the general traffic illegal behaviors; The knowledge base is constructed by using the standard traffic illegal behaviors, the general traffic illegal behaviors, the traffic law articles corresponding to the standard traffic illegal behaviors and the general traffic illegal behaviors, and the investigation operations corresponding to the traffic law articles.

7. The method of claim 1, wherein the traffic incident behavior is identified by: The traffic accident query behavior corresponding traffic illegal behaviors, corresponding traffic law articles and corresponding investigation operations are obtained by querying the knowledge base with the traffic accident query behavior vector, and the traffic accident query behavior corresponding traffic illegal behaviors, corresponding traffic law articles and corresponding investigation operations are pushed to the mobile terminal. The traffic accident query behavior corresponding traffic illegal behaviors, corresponding traffic law articles and corresponding investigation operations are obtained by querying the knowledge base with the traffic accident query behavior vector, and the traffic accident query behavior corresponding traffic illegal behaviors, corresponding traffic law articles and corresponding investigation operations are pushed to the mobile terminal. The method comprises:

8. An apparatus for identifying traffic accident behavior, characterized by, The traffic accident image is input into a target detection model, and the target detection model identifies a traffic prohibition sign and a number of the traffic prohibition sign in the traffic accident image according to color characteristics of the traffic prohibition sign and difference characteristics of the traffic prohibition sign and a non-traffic prohibition sign; The traffic accident behavior is converted into a traffic accident behavior vector, and the traffic accident similar behavior is obtained based on the traffic accident behavior vector among standard traffic illegal behaviors and general traffic illegal behaviors in the knowledge base; The traffic accident behavior is converted into a traffic accident behavior vector, and the traffic accident similar behavior is obtained based on the traffic accident behavior vector among standard traffic illegal behaviors and general traffic illegal behaviors in the knowledge base; The method further comprises:

9. An electronic device for identifying traffic accident behavior, characterized in that, The original illegal behaviors corresponding to the traffic regulations and the traffic law articles are analyzed by the large language model, and the standard traffic illegal behaviors corresponding to the original illegal behaviors are identified; The standard traffic illegal behaviors are expanded by using keywords in the standard traffic illegal behaviors, and the traffic illegal behaviors corresponding to the traffic law articles are extracted from vehicle claim records, to obtain the general traffic illegal behaviors; The knowledge base is constructed by using the standard traffic illegal behaviors, the general traffic illegal behaviors, the traffic law articles corresponding to the standard traffic illegal behaviors and the general traffic illegal behaviors, and the investigation operations corresponding to the traffic law articles. The traffic accident query behavior corresponding traffic illegal behaviors, corresponding traffic law articles and corresponding investigation operations are obtained by querying the knowledge base with the traffic accident query behavior vector, and the traffic accident query behavior corresponding traffic illegal behaviors, corresponding traffic law articles and corresponding investigation operations are pushed to the mobile terminal. The traffic accident query behavior corresponding traffic illegal behaviors, corresponding traffic law articles and corresponding investigation operations are obtained by querying the knowledge base with the traffic accident query behavior vector, and the traffic accident query behavior corresponding traffic illegal behaviors, corresponding traffic law articles and corresponding investigation operations are pushed to the mobile terminal. The method comprises: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement a method as claimed in any of claims 1-7.

10. A computer readable medium having stored thereon a computer program, characterized in that, The program, which when executed by a processor, implements a method as claimed in any of claims 1-7.

Images