Traffic accident auxiliary responsibility judgment method based on large language model

Abstract

The invention discloses a traffic accident auxiliary responsibility judgment method based on a large language model. The method comprises the following steps: collecting data information in a vehicle driving process before an accident occurs; integrating and transmitting data information of various data sources, preprocessing the data information to obtain structured data, and generating driving record description; calling a large language model to perform accident condition analysis according to the structured data and the driving record description; and the large language model generates a structured responsibility judgment report according to the analysis of the accident condition of the vehicle, and transmits the structured responsibility judgment report to a traffic control department to assist traffic accident responsibility judgment. According to the invention, by using the reasoning ability of the large language model, the efficiency of accident responsibility determination is improved, the time loss of multiple parties of accidents is reduced, the police work intensity of the traffic control department is reduced, the insurance process is accelerated, and the accident responsibility determination accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent determination of vehicle traffic accidents, and specifically relates to a traffic accident auxiliary determination method based on a large language model. Background Art

[0002] With the rapid development of cities and the advancement of autonomous driving technology, in the future information environment, vehicle networking technology and autonomous driving technology will become mature, and vehicle networking technology will be basically realized. Vehicles can communicate with each other quickly and obtain mutual information (relative speed, acceleration, location information, etc.).

[0003] Against the backdrop of my country's rapid economic development, the number of motor vehicles continues to increase. According to statistics from the Ministry of Public Security, the number of motor vehicles in China will reach 435 million in 2023, with 24.8 million newly registered motor vehicles, an increase of 16,000 from 2022. With the increase in the number of vehicles, traffic accidents also occur frequently.

[0004] The procedure for determining the responsibility of a traffic accident includes multiple steps such as setting up safety warnings, calling the police, sending the victim to a doctor, taking photos for evidence, evacuating the scene, and the final determination of responsibility. However, there are multiple bottlenecks in the traditional determination of responsibility for traffic accidents. First, the on-site processing time for minor accidents is too long. For minor accidents, the processing time usually takes at least 20 minutes. These minor accidents are one of the main causes of urban traffic congestion and may even cause secondary accidents. Secondly, the responsibility determination cycle for non-minor accidents (including general accidents, major accidents, and particularly major accidents) is long, and the damage assessment process is cumbersome. The traditional manual responsibility determination cycle is slow, the damage assessment process is cumbersome, and it usually takes several working days. Finally, the accident determination consumes a lot of police force. A large number of minor accidents may lead to a shortage of police force, which will affect the allocation of police force and reduce the efficiency of handling.

[0005] Prior art, a method, device, equipment and readable storage medium for determining the responsibility of an automatic driving accident of an intelligent networked vehicle (CN116244664B), the invention relates to a method, device, equipment and readable storage medium for determining the responsibility of an automatic driving accident of an intelligent networked vehicle. The method for determining the responsibility includes the following steps: S1, obtaining accident data of an automatic driving accident; S2, determining whether the automatic driving setting conditions are met, if so, proceeding to step S3, if not, proceeding to step S6; S3, determining whether the ADS has a fault before the automatic driving accident occurs, if so, proceeding to step S5, if not, proceeding to step S4; S4, determining whether the setting conditions for human intervention in vehicle operation are met, if so, proceeding to step S6, if not, proceeding to step S5; S5, determining that the result is the responsibility of the car company, proceeding to step S7; S6, determining that the result is the responsibility of the driver; S7, ending.

[0006] The defect is that the implementation method of the patent is to determine the responsibility after the occurrence of an autonomous driving accident. It determines whether the cause of the accident is the responsibility of the driver or the responsibility of the intelligent networked car autonomous driving accident, and does not determine the responsibility of multiple parties in the traffic accident (traffic accident participants); at the same time, the implementation method of the patent does not make detailed judgments on traffic accidents, and does not automatically generate auxiliary judgment reports. Summary of the invention

[0007] The purpose of the present invention is to address the above-mentioned deficiencies in the prior art and to provide a method for assisting in determining responsibility for traffic accidents based on a large language model, so as to solve the problem that the existing traffic accident responsibility determination fails to identify the responsibilities of multiple parties in a traffic accident.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is:

[0009] A method for assisting in determining responsibility for traffic accidents based on a large language model comprises the following steps:

[0010] S1. Collecting data information of the vehicle during driving before the accident;

[0011] S2, integrating and transmitting data information from various data sources, preprocessing the data information, obtaining structured data, and generating a driving record description;

[0012] S3. Based on the structured data and driving record description, a large language model is used to analyze the accident situation;

[0013] S4. The large language model generates a structured accountability report based on the analysis of the accident condition of the vehicle, and transmits the structured accountability report to the traffic management department to assist in traffic accident accountability.

[0014] Furthermore, the data information in S1 includes:

[0015] Vehicle information, including basic information about vehicle operation, vehicle operation status information, and safety device status information;

[0016] Road and environmental data, which includes road conditions, traffic signals, and weather data;

[0017] External vehicle and environment information, including other vehicles’ driving behaviors and surrounding obstacles;

[0018] Furthermore, the data information is preprocessed in S2, including:

[0019] Remove noise from sensor data, fill in missing data, and correct abnormal data;

[0020] Convert each data into a unified coordinate system and timestamp;

[0021] Through feature extraction, external data such as weather and road conditions are converted into descriptions with contextual semantics;

[0022] The preprocessed structured data is transmitted to the large language model in real time, and a driving record description is generated.

[0023] Furthermore, the S3 comprises the following sub-steps:

[0024] S31. Before the accident occurs, the big language model is called through the API interface, and the structured data is input into the big language model to generate a driving record description;

[0025] S32. At the accident occurrence stage, the large language model determines whether an emergency has occurred with the current vehicle based on the driving record description, and further determines whether a traffic accident has occurred;

[0026] S33. In the post-accident phase, the large language model generates descriptions and key memories with the new driving record description frequency;

[0027] S34. The large language model generates a snapshot of the accident based on the key memories described by different driving records at three time stages: before the accident, during the occurrence stage, and after the accident.

[0028] Furthermore, the S31 specifically includes:

[0029] Start the vehicle behavior description function, input the structured data into the called large language model, and use the prompt word template of the large language model to obtain the model's response content;

[0030] Convert the model's response content into a driving record description in a time series format;

[0031] Based on the driving record description and real-time data in the current time series format, a phased summary of the time period is generated and stored as "key memory-phased driving record data summary" as a new context input for the next time series.

[0032] Further, the S32 includes the following sub-steps:

[0033] S321. Using the “key memory—phased driving data record summary” in S31 as a new context input of the large language model, the large language model identifies whether there is an emergency in the structured description; if no emergency occurs, the phased description of the time series is stored as “key memory—driving record description”, and determines whether the vehicle is stopped by vehicle speed, acceleration, and vehicle start state parameters;

[0034] If the vehicle stops, the large language model is called to summarize the trip information, generate the "key memory-stage driving data record summary" and end the traffic accident judgment process;

[0035] If the vehicle does not stop, the data related to the time series in the large model context input is deleted, and the phased description memory is used as the context input of the next time series;

[0036] If an emergency occurs, enter S322;

[0037] S322: If an emergency occurs to the vehicle, the large language model determines whether the vehicle is involved in a traffic accident and generates an analysis record;

[0038] If the vehicle has not been involved in a traffic accident, the emergency feature description and analysis record is stored as key memory, and a judgment is made as to whether the vehicle has stopped. If the vehicle has stopped, the large language model summarizes the trip information based on all stored key memories, and ends the traffic accident judgment process.

[0039] If the vehicle does not stop, in the process of generating the driving record description of the next time series, the data input in the previous time series is deleted, and the key memory-emergency description and analysis record and the key memory-stage driving record data summary are used as the new context input of the next time series;

[0040] If the vehicle is involved in a traffic accident, the process proceeds to S333;

[0041] S323, the large language model increases the frequency of the driving record description, and generates the driving record description as a new context input of the large language model;

[0042] S324, the large language model determines whether the traffic accident is over;

[0043] If the traffic accident is not over, a driving record description and a key memory-stage driving record data summary are generated, and the key memory-stage driving record data summary is stored;

[0044] If the traffic accident is over, the process proceeds to S33.

[0045] Furthermore, the S34 specifically includes:

[0046] The stored key memories, including the characteristic description and analysis records of the emergency, and the summary of the driving record data of the previous stage before the accident, are used as the new context input of the large language model to generate an accident snapshot to provide an overview of the entire accident process.

[0047] Furthermore, the S4 specifically includes:

[0048] The driving record data log generated during the accident phase is used as context input, and the data is visualized using a large language model to generate a curve icon that changes over time. The accident is analyzed by combining key memories before, during, and after the accident, driving record descriptions, and accident snapshots to generate a structured accountability report.

[0049] Furthermore, the curve icons that change with time include: a curve showing speed changing with time, a curve showing acceleration changing with time, and a curve showing the change of relative distance between the leading vehicle and the trailing vehicle.

[0050] The traffic accident auxiliary judgment method based on the large language model provided by the present invention has the following beneficial effects:

[0051] 1. The present invention improves the efficiency of accident responsibility determination, reduces the time loss of multiple parties involved in the accident (participants, insurance companies, and traffic police), reduces the workload of the traffic control department's police force, speeds up the insurance process, and improves the accuracy of accident responsibility determination.

[0052] 2. The present invention utilizes the reasoning ability of a large language model to generate a driving record description. After an emergency or traffic accident occurs to a vehicle, a driving record description and a structured accident responsibility auxiliary judgment example are generated in real time. A traffic accident auxiliary judgment method based on time series visual charts of speed, acceleration, distance, etc. are generated based on the collected real-time data and uninterrupted driving record descriptions. The generation of a detailed and accurate driving record (accident) description can greatly reduce the time spent by traffic control departments, insurance companies, etc. in liability identification, and provide accurate and detailed liability identification assistance to the accident identification party, thereby reducing the time loss of the accident party, traffic control department, and insurance company; immediate handling of traffic accidents can greatly reduce the road congestion caused by handling accidents and the potential risk of accidents caused by road congestion.

[0053] 3. The present invention uses a large language model to process the various information collected to determine whether an emergency, traffic accident, accident is over, or vehicle is stopped, and generates a time series-based accident snapshot, a structured accountability report, and a curve of speed, acceleration, and relative distance to other vehicles that changes over time to assist vehicle drivers, insurance companies, and traffic control departments in determining and judging responsibilities. Systematic process and method. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 The figure is a flow chart of the auxiliary judgment method of traffic accidents based on a large language model of the present invention.

[0055] Figure 2 This is an example of a prompt word template for generating a driving record according to the present invention.

[0056] Figure 3The generation process of the time-based driving record of the present invention is described.

[0057] Figure 4 It is a flow chart of traffic accident judgment of the present invention.

[0058] Figure 5 This is an accident snapshot case of the present invention.

[0059] Figure 6 1 is an example of a graph showing the change of the present invention over time.

[0060] Figure 7 This is an example of the accident liability determination result of the present invention. DETAILED DESCRIPTION

[0061] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0062] Definitions of Abbreviations and Key Terms in the Present Invention

[0063] LLMs: Large Language Models. It is a natural language processing (NLP) model based on deep learning. It understands, generates and predicts language by training a large amount of text data. LLMs can process and generate complex language tasks such as text generation, translation, summarization, question answering, sentiment analysis, etc. Through large-scale parameters and training data, it captures the deep laws and contextual relationships of language.

[0064] Time series: A time series is a set of data points arranged in time order. Usually, these data are collected at a specific time interval and have the attribute of time order. In a time series, each data point represents a value observed at a certain moment or in a certain period of time, and is usually closely related to the time variable.

[0065] API: Application Programming Interface. It is a set of definitions, protocols, and tools for building software and applications, which specifies how software components interact, including communication between libraries, operating systems, and microservices.

[0066] Prompt: In the field of large language models, "prompt" refers to the input text or instructions given by the user, which is used to guide the model to generate the corresponding output. When users interact with large language models, they ask a question, a sentence, a paragraph, or any type of text content, which are called prompts.

[0067] Key memory: refers to the information retained by the model in the current conversation in the text generation type language model. This information is used to help understand and generate user interactions, enabling it to provide more contextual, continuous and relevant responses.

[0068] ABS: Anti-lock Braking System is a car safety system used to prevent the wheels from locking during emergency braking, thereby maintaining the vehicle's controllability and stability.

[0069] Example 1

[0070] The present embodiment provides a method for assisting in determining responsibility for traffic accidents based on a large language model, which specifically comprises four processing stages and three time stages;

[0071] The three time stages include: pre-accident stage, accident stage, and post-accident stage;

[0072] The four processing stages include:

[0073] The first stage is the information collection stage, which corresponds to the stage before the accident.

[0074] The second stage is the data transmission and preprocessing stage; this stage corresponds to the stage before the accident occurs;

[0075] The third stage is the accident situation analysis stage, which corresponds to the accident occurrence stage;

[0076] The fourth stage is accident judgment and result output, which corresponds to the post-accident stage;

[0077] Based on this, reference Figure 1 and Figure 4 , the present invention specifically includes the following contents:

[0078] Step S1, information collection stage: collecting data information during the vehicle driving process before the accident, which specifically includes the following contents:

[0079] Vehicle information refers to various information about the vehicle currently being driven by the driver, including basic information about the vehicle's operation, vehicle operating status information, and safety device status information.

[0080] The basic information of the vehicle's current operating status includes time, position, speed, acceleration, vehicle steering angle, vehicle ID, driver information, etc.;

[0081] The vehicle's safety status information refers to the activation status of the vehicle's passive safety devices, such as airbags, ABS (anti-lock braking system), etc. This information can be obtained through the vehicle's own acceleration sensor, speed sensor and other sensors to determine the vehicle's current operating status and safety status, and help determine the severity of the accident.

[0082] Road and environmental data refers to road conditions, traffic signals, weather data, and current weather conditions including rain, snow, slippery roads, temperature, wind speed, etc. It can be obtained through video cameras, radar detectors, and infrared sensors around the vehicle to monitor the vehicle's surrounding environment information and detect vehicle environmental safety in real time.

[0083] External vehicle and environment information refers to the driving behavior of other vehicles besides the current vehicle. It obtains the speed, acceleration, distance, relative position, video data of the front, rear, left and right sides of other vehicles around it through sensors such as radar and cameras. In addition, it also includes surrounding obstacles such as whether there are pedestrians and obstacles around the vehicle, which are obtained using ultrasonic sensors, radars and other equipment.

[0084] The specific information collected is shown in Table 1;

[0085] Table 1 Information collection table

[0086]

[0087]

[0088] Step S2, data transmission and preprocessing stage; used to integrate and transmit data information from various data sources, and preprocess the data information to obtain structured data and generate a driving record description, which specifically includes the following contents:

[0089] The main purpose of this phase is to integrate and transmit data from various data sources, including real-time information from other vehicles, environmental data obtained from API interfaces, and data from current vehicle sensors, and to clean, standardize, and preprocess these raw data. The preprocessed data will be converted into structured text data that can be used by large language models to generate accurate and real-time driving record descriptions.

[0090] During the data integration and transmission stage, information from various data sources is collected in real time and uniformly transmitted to the processing platform. These data include vehicle sensors (such as GPS, IMU, vehicle speed sensors), surrounding traffic environment (such as other vehicles, traffic signals, road conditions), and external environment (such as weather information, road traffic, etc.). Vehicles exchange information with other vehicles and road infrastructure in real time through on-board communication systems (such as CAN bus, V2X communication) to ensure that all relevant data can be transmitted to the system in a timely and accurate manner. Efficient transmission protocols and low-latency communication technologies ensure that large language models can obtain complete driving data in real time.

[0091] In the data preprocessing stage, the raw data is cleaned, standardized, and formatted to be converted into structured text data that can be used by the large language model. First, the sensor data is de-noised, missing data is filled, and outliers are corrected. Then, the data is unified into a standard format, such as converting the vehicle's speed, acceleration, and position data into a unified coordinate system and timestamp. Next, feature extraction is used to convert external data such as weather and road conditions into descriptions with contextual semantics. Finally, the processed structured data will be transmitted to the language model in real time to support it in generating accurate driving record descriptions before, during, and after an accident.

[0092] Step S3, the accident situation analysis phase, is used to call the large language model to analyze the accident situation according to the structured data and driving record description. The language model called by the API here is a common domestic language model (such as Tongyi Qianwen, Wenxin Yiyan, DeepSeekV3, etc.); after the data transmission and preprocessing phase, the processed structured data is obtained. In order to ensure the privacy and security of the data and prevent the risk of data leakage, the proposed auxiliary judgment method calls the large language model through the API. In the entire auxiliary judgment process, there is no data transmission to the cloud, and all are integrated in the vehicle currently driven by the driver. Specifically, it includes three time periods before the accident, during the accident, and after the accident. According to the various information collected and processed in the previous stage, the large language model produces different reflections and response result outputs, which specifically include the following sub-steps:

[0093] Step S31: Before the accident occurs, the large language model service is called through the API interface for reasoning. The specific method is to use the preset Prompt (prompt word) template and combine the customized program to input the pre-processed structured information into the language model. Figure 2 The language model prompt template used covers the task requirements, data structure, and data items represented by multiple structured data variables. The implementation steps are as follows: first, start the vehicle behavior description function; second, input the preprocessed structured data; then, use the prompt word template of the large language model to obtain the model's response content.

[0094] refer to Figure 3 , converting the response results of the large language model into a driving record description in a time series format. The generation of driving record description text is based on time series data, and the text description is continuously generated according to the changes in the collected real-time data. Before the accident, the duration of each time series is T minutes, and every a 1 The driving text description is updated every second.

[0095] The description of driving records based on time series is an ongoing process, covering all time periods before, during, and after the accident. Next, the system enters the next step to generate a phased summary of the time period based on the real-time data and driving record description of the current time series (within T minutes) to avoid too much context input to the language model causing information overload and ensure that the description is structured and concise. The content of the phased summary will summarize the changes in the vehicle's operating status during the time period. This phased summary will serve as an overview description of the time period, helping the large language model to keep the key information in mind in subsequent analysis, and will be stored as "Key Memory-Phase Driving Record Data Summary" as the new context input for the next time series.

[0096] Step S32: At the accident occurrence stage, the large language model determines whether an emergency occurs to the current vehicle based on the driving record description, and further determines whether a traffic accident occurs, which specifically includes the following sub-steps:

[0097] Step S321: According to the "key memory - summary of driving data records at different stages" in step S31, the large language model is required to identify whether there are emergencies in the structured description, such as sudden change in speed, sudden drop in acceleration, airbag deployment, anti-lock braking system (ABS) locking, raining, etc. If no emergencies occur, the phased description of the time series is stored as "key memory - driving record description", and the vehicle speed, acceleration, vehicle start state and other parameters are used to determine whether the vehicle has stopped. If the vehicle has stopped, the language model is called and a new task is specified - based on all stored key memories, the large model is required to summarize the travel information, generate "key memory - summary of driving data records at different stages" and end the traffic accident judgment process. If the vehicle has not stopped, the data related to the time series in the context input of the large model is deleted, and the phased description memory is used as the context input of the next time series, and the system continues to cycle.

[0098] If an emergency occurs, enter S322;

[0099] Step S322: If an emergency occurs to the vehicle, the large language model determines whether the vehicle has a traffic accident and generates an analysis record;

[0100] Specifically, if the vehicle encounters an emergency, a new task is assigned to the large language model, requiring it to perform emergency feature analysis based on the key memory (stage-by-stage driving data record summary) and vehicle structured data of the previous step. In the analysis, acceleration threshold, collision force, installation device status, speed change, etc. are used as accident judgment indicators to determine whether the vehicle has a traffic accident and generate analysis records. If the analysis result shows that the vehicle has not had an accident, the emergency feature description and analysis record are stored as key memory (emergency description and analysis), and the vehicle speed, acceleration, opening state and other parameters are used to determine whether the vehicle has stopped. If the vehicle has stopped, the large language model is called and a new task is assigned - requiring the large language model to summarize the travel information based on all stored key memories, and the traffic accident judgment process is terminated. If the vehicle has not stopped, in the process of generating the driving record description of the next time series, the data input in the previous time series is deleted, and the key memory-emergency description and analysis record, key memory-stage driving record data summary are used as the new context input of the next time series, and the system continues to cycle.

[0101] If the vehicle has a traffic accident, proceed to step S333;

[0102] Step S333: The large language model increases the frequency of the driving record description, and generates the driving record description as a new context input of the large language model;

[0103] When a vehicle is involved in a traffic accident, the system will assign a new task—requiring the large language model to increase the frequency of driving record descriptions in order to capture more detailed traffic accident content and generate richer driving record descriptions as new context inputs for the large language model. Specifically, the frequency of driving record descriptions generated by the system is determined by a 1 Convert to a 2 , and a 2 >a 1 After increasing the frequency of driving record description, proceed to the next step;

[0104] Step S334: The large language model determines whether the traffic accident is over;

[0105] If the traffic accident is not over, a driving record description and a key memory-stage driving record data summary are generated, and the key memory-stage driving record data summary is stored;

[0106] After the system adds the driving record description, it updates the new task of the large language model - judging whether the accident is over, with whether the vehicle stops, whether the speed and acceleration change, and the vehicle start state as judgment indicators. If it is judged that the accident is not over, it continues to loop, generating the driving record description and key memory - stage driving record data summary, and only storing the key memory - stage driving record data summary; if it is judged that the accident is over, it enters the post-accident stage and goes to step S33.

[0107] Step S33: In the post-accident stage, the large language model generates descriptions and key memories with the new driving record description frequency;

[0108] Specifically, after an accident is detected, the large language model is updated with a new task, requiring the large language model to continue monitoring the vehicle status (such as whether the warning lights are activated after parking, whether traffic is restored), and to describe the frequency a according to the new driving record 3 Generate a new description. Since the vehicle is often stopped after an accident, the vehicle data features do not change much. In order to avoid inputting a large driving record description frequency into the language model and causing information overload, the new driving record description frequency a 3 Less than the initial driving record description frequency a 1 After each time series ends, the system will generate and store a summary of the driving record data for subsequent generation and analysis. After generating descriptions and key memories with the new driving record description frequency, the process proceeds to step S34;

[0109] Step S34, the large language model generates a snapshot of the accident based on key memories of different driving records described in three time periods: before the accident, during the accident, and after the accident.

[0110] Specifically, based on the key memories described in different driving records at three time stages before, during, and after the accident, an accident snapshot is generated, which provides an overview of the key time and vehicle behavior before, during, and after the accident, helping the large language model to conduct subsequent accident responsibility analysis and assist in determining responsibility. Specifically, the previously stored key memories, including the characteristic description and analysis records of the emergency, and the summary of the driving record data of the previous stage before the accident, are used as new context inputs for the large language model to generate an accident snapshot, providing an overview of the entire accident process, helping the subsequent language model to determine responsibility and understand the scene.

[0111] refer to Figure 5 ,The accident snapshot includes the timestamp, the key memory generated by the large language model in the previous step—the phased summary of driving description, vehicle operation information, environmental information, the activation status of safety devices, the event type judgment result generated by the large language model based on the data input, the accident status judgment result, and multiple information descriptions based on time changes.

[0112] Step S4: The large language model generates a structured accountability report based on the analysis of the accident condition of the vehicle, and transmits the structured accountability report to the traffic control department to assist in traffic accident accountability.

[0113] The driving record data log generated during the accident phase is used as context input, and the data is visualized using a large language model to generate a curve icon that changes over time. The accident is analyzed by combining key memories before, during, and after the accident, driving record descriptions, and accident snapshots to generate a structured accountability report.

[0114] refer to Figure 6 and Figure 7 In this process, the large language model first generates multiple time-varying curve charts based on the vehicle's driving record data log, such as the speed-time curve, the acceleration-time curve, and the relative distance curve between the front and rear vehicles. In addition, the time or time period of the emergency will be marked in the figure. Subsequently, the large language model task is updated - based on the time-varying chart, the cause of the accident is analyzed in combination with the key memory before, during and after the accident, the driving record description, and the accident snapshot. The content of the accident judgment includes the basic information of the accident, the content of the responsibility judgment, the analysis after the accident, the cause analysis, the traffic regulations violated, the judgment result, etc.

[0115] Although the specific implementation of the invention is described in detail in conjunction with the drawings, it should not be understood as limiting the scope of protection of this patent. Within the scope described in the claims, various modifications and variations that can be made by those skilled in the art without creative work still fall within the scope of protection of this patent.

Claims

1. A method for assisting in judging responsibility for traffic accidents based on a large language model, characterized in that: The following steps are involved: S1. Collecting data information of the vehicle during driving before the accident; S2, integrating and transmitting data information from various data sources, preprocessing the data information, obtaining structured data, and generating a driving record description; S3. Based on the structured data and driving record description, a large language model is used to analyze the accident situation; S4. The large language model generates a structured accountability report based on the analysis of the accident condition of the vehicle, and transmits the structured accountability report to the traffic management department to assist in traffic accident accountability.

2. The method for assisting in determining responsibility for traffic accidents based on a large language model according to claim 1 is characterized in that: The data information in S1 includes: Vehicle information, including basic information about vehicle operation, vehicle operation status information, and safety device status information; Road and environmental data, which includes road conditions, traffic signals, and weather data; External vehicle and environment information, which includes the driving behavior of other vehicles and surrounding obstacles.

3. The method for assisting in determining responsibility for traffic accidents based on a large language model according to claim 1, characterized in that: The data information is preprocessed in S2, including: Remove noise from sensor data, fill in missing data, and correct abnormal data; Convert each data into a unified coordinate system and timestamp; Through feature extraction, external data such as weather and road conditions are converted into descriptions with contextual semantics; The preprocessed structured data is transmitted to the large language model in real time, and a driving record description is generated.

4. The method for assisting in determining responsibility for traffic accidents based on a large language model according to claim 1, characterized in that: The S3 comprises the following sub-steps: S31. Before the accident occurs, the big language model is called through the API interface, and the structured data is input into the big language model to generate a driving record description; S32. At the accident occurrence stage, the large language model determines whether an emergency has occurred with the current vehicle based on the driving record description, and further determines whether a traffic accident has occurred; S33. In the post-accident phase, the large language model generates descriptions and key memories with the new driving record description frequency; S34. The large language model generates a snapshot of the accident based on the key memories described by different driving records at three time stages: before the accident, during the occurrence stage, and after the accident.

5. The method for assisting in determining responsibility for traffic accidents based on a large language model according to claim 4 is characterized in that: The S31 specifically includes: Start the vehicle behavior description function, input the structured data into the called large language model, and use the prompt word template of the large language model to obtain the model's response content; Convert the model's response content into a driving record description in a time series format; Based on the driving record description and real-time data in the current time series format, a phased summary of the time period is generated and stored as "Key Memory—Phase Driving Record Data Summary" as a new context input for the next time series.

6. The method for assisting in determining responsibility for traffic accidents based on a large language model according to claim 4 is characterized in that: The S32 comprises the following sub-steps: S321. Based on the "key memory—phased driving data record summary" in S31 as a new context input of the large language model, the large language model identifies whether there is an emergency in the structured description; if no emergency occurs, the phased description of the time series is stored as "key memory—driving record description", and whether the vehicle is stopped is determined by the vehicle speed, acceleration, and vehicle start state parameters; If the vehicle stops, the large language model is called to summarize the travel information, generate "key memory-stage driving data record summary" and end the traffic accident judgment process; If the vehicle does not stop, the data related to the time series in the large model context input is deleted, and the phased description memory is used as the context input of the next time series; If an emergency occurs, enter S322; S322: If an emergency occurs to the vehicle, the large language model determines whether the vehicle is involved in a traffic accident and generates an analysis record; If the vehicle has not been involved in a traffic accident, the emergency feature description and analysis record is stored as key memory, and a judgment is made as to whether the vehicle has stopped. If the vehicle has stopped, the large language model summarizes the trip information based on all stored key memories, and ends the traffic accident judgment process. If the vehicle does not stop, in the process of generating the driving record description of the next time series, the data input in the previous time series is deleted, and the key memory-emergency description and analysis record and the key memory-stage driving record data summary are used as the new context input of the next time series; If the vehicle is involved in a traffic accident, the process proceeds to S333; S323, the large language model increases the frequency of the driving record description, and generates the driving record description as a new context input of the large language model; S324, the large language model determines whether the traffic accident is over; If the traffic accident is not over, a driving record description and a key memory-stage driving record data summary are generated, and the key memory-stage driving record data summary is stored; If the traffic accident is over, the process proceeds to S33.

7. The method for assisting in determining responsibility for traffic accidents based on a large language model according to claim 4 is characterized in that: The S34 specifically includes: The stored key memories, including the characteristic description and analysis records of the emergency, and the summary of the driving record data of the previous stage before the accident, are used as the new context input of the large language model to generate an accident snapshot to provide an overview of the entire accident process.

8. The method for assisting in determining responsibility for traffic accidents based on a large language model according to claim 1, characterized in that: The S4 specifically includes: The driving record data log generated during the accident phase is used as context input, and the data is visualized using a large language model to generate a curve icon that changes over time. The accident is analyzed by combining key memories before, during, and after the accident, driving record descriptions, and accident snapshots to generate a structured accountability report.

9. The method for assisting in determining responsibility for traffic accidents based on a large language model according to claim 8 is characterized in that: The time-varying curve icons include: a speed-varying curve, an acceleration-varying curve, and a relative distance-varying curve between a leading vehicle and a trailing vehicle.

Images