Accident report generation method and vehicle information processing system

Abstract

The invention provides an accident report generation method and a vehicle information processing system, and the method comprises the steps: determining a target time period of an accident occurrence moment, and obtaining vehicle end data reported by an accident vehicle and road end data reported by road side equipment in the target time period; generating at least one relation vector according to the vehicle end data, the road end data and a relation reasoning model obtained by pre-training; inputting each relation vector into a pre-trained task prediction model for task reasoning, and generating a task result; and inputting the task result into a pre-trained report generation model, and generating and pushing an accident report by the report generation model. According to the method, the complex dependency relationship between vehicles, between the vehicle and the environment and between different sensing modes can be automatically excavated, a causal chain on a time sequence is established, and intelligent attribution is performed on accident causes, so that objectivity, consistency and authenticity in a responsibility affirmation process are improved.

Description

Technical Field

[0001] This application relates to the field of vehicle data processing technology, and more specifically, to an accident report generation method and a vehicle information processing system. Background Technology

[0002] With the rapid development of intelligent transportation systems and autonomous driving technologies, vehicle-road cooperation and vehicle-road-cloud integration have become important directions for improving road traffic safety and operational efficiency. Against this backdrop, rapid response and liability determination mechanisms for vehicle collision accidents are facing higher demands. Especially in the event of minor collisions, how to ensure the rapid removal of vehicles involved in the accident while maintaining traffic safety, reducing traffic congestion, and providing objective and complete evidence for subsequent insurance claims and liability determination has become a critical issue that urgently needs to be addressed in the field of intelligent transportation.

[0003] Currently, determining liability in vehicle collision accidents primarily relies on on-site photographic evidence taken by the parties involved and on-site investigation by traffic police. However, this method has significant drawbacks, including strong subjectivity, poor accuracy, difficulty in guaranteeing authenticity, and low efficiency. Summary of the Invention

[0004] The purpose of this application is to address the shortcomings of the prior art by providing an accident report generation method and a vehicle information processing system, thereby solving the problems of strong subjectivity, poor accuracy, difficulty in guaranteeing authenticity, and low efficiency in the prior art.

[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows: In a first aspect, one embodiment of this application provides an accident report generation method, applied to a server in a vehicle information processing system, the vehicle information processing system comprising: the server, multiple vehicles, and multiple roadside devices, the method comprising: Determine the target time period in which the accident occurred, and acquire vehicle-side data reported by the accident vehicle and roadside data reported by the roadside equipment during the target time period. The vehicle-side data includes driving log data, driving image data, and point cloud data, while the roadside data includes video data and traffic light data. Based on the vehicle-side data, the road-side data, and the pre-trained relational reasoning model, at least one relation vector is generated, and each relation vector is used to indicate the dependency relationship between the participating entities when an accident occurs. Each of the aforementioned relation vectors is input into a pre-trained task prediction model for task inference to generate task results. The task prediction model is based on a multi-task prediction head, and the task results include: collision results, behavior results, and attribution results. The task results are input into a pre-trained report generation model, which then generates and pushes an accident report.

[0006] In one possible implementation, generating at least one relation vector based on the vehicle-side data, the roadside data, and a pre-trained relation inference model includes: The vehicle-side data and road-side data are fused to obtain a fused feature sequence, which includes multiple fused feature vectors arranged in chronological order. The fused feature sequence is input into a pre-trained relational reasoning model to generate at least one relational vector.

[0007] In one possible implementation, the data fusion of the vehicle-side data and the roadside data to obtain a fused feature sequence includes: The driving log data is processed using a pre-defined text processing model to extract text features and generate structured behavior tags. The driving image data is processed by a preset first image encoder to generate vehicle-side image features; The video data is processed by a preset second image encoder to generate roadside image features; The point cloud data is encoded using a preset point cloud encoder to generate a point cloud feature vector. The structured behavior labels, vehicle-side image features, roadside image features, point cloud feature vectors, and traffic light data are fused in chronological order to obtain the fused feature sequence.

[0008] In one possible implementation, the relational reasoning model includes a plurality of relational reasoning modules connected in sequence, each of which is based on a multi-head self-attention mechanism and a feedforward neural network. The step of inputting the fused feature sequence into a pre-trained relation inference model to generate at least one relation vector includes: Each of the aforementioned relation reasoning modules sequentially performs self-attention mechanism relation reasoning on the fused feature sequence to generate at least one relation vector.

[0009] In one possible implementation, the task prediction model includes: a collision prediction head, a behavior classification head, an attribution classification head, and a fusion module; the step of inputting each of the relation vectors into the pre-trained task prediction model for task inference to generate task results includes: Each of the aforementioned relation vectors is input into the collision prediction head to predict the collision probability and generate a collision result, which is used to indicate the probability of a collision occurring. Each of the aforementioned relation vectors is input into the behavior classification head to perform behavior classification prediction and generate behavior results, which are used to indicate the behavior classification of the accident vehicle. Each of the aforementioned relation vectors is input into the attribution classification head to perform accident attribution prediction and generate attribution results, which are used to indicate the cause of the accident. The collision results, the behavior results, and the attribution results are input into the fusion module to generate task results.

[0010] In one possible implementation, the report generation model includes: a scenario construction module, a responsibility weight assessment module, and a text generation module; The step of inputting the task results into a pre-trained report generation model, and having the report generation model generate and push an accident report, includes: The task results are input into the scenario construction module to construct a scenario and generate an accident semantic graph. The task results and the accident semantic graph are input into the responsibility weight assessment module to generate responsibility weight assessment results; The accident semantic graph and the responsibility weight assessment results are input into the text generation module to generate and push the accident report.

[0011] In one possible implementation, before determining the target time period in which the accident occurred, the method further includes: Construct a sensing network and data channels.

[0012] In one possible implementation, the process of collecting the vehicle-side data includes: Point cloud data of each vehicle is collected using the vehicle's lidar. Driving image data of each vehicle is acquired through the vehicle's camera unit; The driving log data of each vehicle is collected through the vehicle's electronic control unit; The point cloud data, driving image data, and driving log data of each vehicle are transmitted to the server via a data channel.

[0013] In one possible implementation, the process of collecting the roadside data includes: Video data and traffic light data of each vehicle during its journey are collected by multiple roadside devices, each of which includes millimeter-wave radar, lidar, and high-definition cameras.

[0014] Secondly, another embodiment of this application provides a vehicle information processing system, the system comprising: a server, multiple vehicles, and multiple roadside devices; the server is used to perform the steps of the method described in any of the first aspects.

[0015] Thirdly, another embodiment of this application provides an accident report generation device applied to a server in a vehicle information processing system, the vehicle information processing system including: the server, multiple vehicles, and multiple roadside devices, the device including: The determination module is used to determine the target time period in which the accident occurred, and to acquire vehicle-side data reported by the accident vehicle and roadside data reported by the roadside equipment during the target time period. The vehicle-side data includes driving log data, driving image data and point cloud data, and the roadside data includes video data and traffic light data. The relation reasoning module is used to generate at least one relation vector based on the vehicle-side data, the road-side data and the pre-trained relation reasoning model. Each relation vector is used to indicate the dependency relationship between the participating entities when an accident occurs. The task reasoning module is used to input the aforementioned relation vectors into a pre-trained task prediction model for task reasoning and to generate task results. The task prediction model is based on a multi-task prediction head, and the task results include: collision results, behavior results, and attribution results. The report generation module is used to input the task results into a pre-trained report generation model, which then generates and pushes an accident report.

[0016] Optionally, the relation reasoning module is specifically used for: The vehicle-side data and road-side data are fused to obtain a fused feature sequence, which includes multiple fused feature vectors arranged in chronological order. The fused feature sequence is input into a pre-trained relational reasoning model to generate at least one relational vector.

[0017] Optionally, the relation reasoning module is specifically used for: The driving log data is processed using a pre-defined text processing model to extract text features and generate structured behavior tags. The driving image data is processed by a preset first image encoder to generate vehicle-side image features; The video data is processed by a preset second image encoder to generate roadside image features; The point cloud data is encoded using a preset point cloud encoder to generate a point cloud feature vector. The structured behavior labels, vehicle-side image features, roadside image features, point cloud feature vectors, and traffic light data are fused in chronological order to obtain the fused feature sequence.

[0018] Optionally, the relational reasoning model includes multiple relational reasoning modules connected in sequence, each relational reasoning module being implemented based on a multi-head self-attention mechanism and a feedforward neural network; the relational reasoning module is specifically used for: Each of the aforementioned relation reasoning modules sequentially performs self-attention mechanism relation reasoning on the fused feature sequence to generate at least one relation vector.

[0019] Optionally, the task prediction model includes: a collision prediction head, a behavior classification head, an attribution classification head, and a fusion module; the step of inputting each of the relation vectors into the pre-trained task prediction model for task inference and generating task results includes: Each of the aforementioned relation vectors is input into the collision prediction head to predict the collision probability and generate a collision result, which is used to indicate the probability of a collision occurring. Each of the aforementioned relation vectors is input into the behavior classification head to perform behavior classification prediction and generate behavior results, which are used to indicate the behavior classification of the accident vehicle. Each of the aforementioned relation vectors is input into the attribution classification head to perform accident attribution prediction and generate attribution results, which are used to indicate the cause of the accident. The collision results, the behavior results, and the attribution results are input into the fusion module to generate task results.

[0020] Optionally, the report generation model includes: a scenario construction module, a responsibility weight assessment module, and a text generation module; the report generation module is specifically used for: The task results are input into the scenario construction module to construct a scenario and generate an accident semantic graph. The task results and the accident semantic graph are input into the responsibility weight assessment module to generate responsibility weight assessment results; The accident semantic graph and the responsibility weight assessment results are input into the text generation module to generate and push the accident report.

[0021] Optionally, before determining the target time period in which the accident occurred, the method further includes: Construct a sensing network and data channels.

[0022] Optionally, the process of collecting vehicle-side data includes: Point cloud data of each vehicle is collected using the vehicle's lidar. Driving image data of each vehicle is acquired through the vehicle's camera unit; The driving log data of each vehicle is collected through the vehicle's electronic control unit; The point cloud data, driving image data, and driving log data of each vehicle are transmitted to the server via a data channel.

[0023] Optionally, the process of collecting the roadside data includes: Video data and traffic light data of each vehicle during its journey are collected by multiple roadside devices, each of which includes millimeter-wave radar, lidar, and high-definition cameras.

[0024] Fourthly, another embodiment of this application provides a server, including: a processor, a storage medium, and a bus, wherein the storage medium stores machine-readable instructions executable by the processor, and when the server is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the steps of any of the methods described in the first aspect above.

[0025] Fifthly, another embodiment of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of any of the methods described in the first aspect above.

[0026] The beneficial effects of this application are as follows: By determining the target time period in which the accident occurred and acquiring vehicle-side data reported by the accident vehicle and roadside data reported by roadside equipment during the target time period, at least one relation vector is generated based on the vehicle-side data, roadside data, and a pre-trained relational reasoning model. These relation vectors are then input into a pre-trained task prediction model for task reasoning to generate task results. These results are then input into a pre-trained report generation model, which generates and pushes an accident report. This approach automatically uncovers complex dependencies between vehicles, between vehicles and the environment, and between different perception modalities, establishing causal chains in a time series and intelligently attributing the causes of accidents. This improves the objectivity, consistency, and authenticity of the liability determination process, forming an immutable and traceable digital evidence chain. Furthermore, it enables automatic capture and uploading of key data within milliseconds, avoiding the loss of crucial information due to on-site movement or human error, thus improving the speed and completeness of obtaining evidence for accident liability determination. In addition, it lowers the understanding threshold for non-professionals and significantly improves response efficiency and traffic management capabilities. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A system architecture diagram of the vehicle information processing system used in the accident report generation method provided in this application embodiment; Figure 2 A schematic flowchart of an accident report generation method provided in an embodiment of this application; Figure 3 A schematic flowchart illustrating the generation of at least one relation vector in the accident report generation method provided in this application embodiment; Figure 4 A schematic flowchart illustrating the process of obtaining a fused feature sequence in the accident report generation method provided in this embodiment of the application; Figure 5 A schematic diagram of a task prediction model in the accident report generation method provided in this application embodiment; Figure 6 A schematic flowchart illustrating the process of generating task results in the accident report generation method provided in this application embodiment; Figure 7 A schematic diagram of the report generation model in the accident report generation method provided in this application embodiment; Figure 8 This is a schematic diagram of a process in which an accident report generation method provided in an embodiment of this application generates and pushes an accident report by a report generation model. Figure 9 A schematic diagram of an accident report generation device provided in an embodiment of this application; Figure 10 This is a schematic diagram of a server structure provided in an embodiment of this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0030] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0031] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0032] Currently, the determination of liability in vehicle collision accidents mainly relies on the parties involved taking photos as evidence at the scene and the traffic police conducting on-site investigations.

[0033] However, manual evidence collection is highly subjective, prone to problems such as poor photo angles and omission of key information, and cannot obtain high-precision vehicle dynamic data (such as acceleration, steering angle, braking timing, etc.) at the moment of the accident, resulting in an incomplete chain of evidence. Secondly, data such as surveillance videos and vehicle logs are isolated and lack a unified time benchmark and spatial alignment mechanism, making it difficult to achieve cross-device and cross-modal data association and fusion, affecting the authenticity and accuracy of accident reconstruction. Thirdly, existing solutions generally lack the ability to conduct automated responsibility analysis based on multi-source data and fail to combine traffic regulations and behavioral semantics for causal reasoning, resulting in long responsibility determination cycles and a large component of subjective judgment, making it difficult to meet the needs of efficient, fair, and traceable management.

[0034] Therefore, the existing technology that relies on the parties involved to take photos and collect evidence on-site and on-site investigation by traffic police has obvious drawbacks such as strong subjectivity, poor accuracy, difficulty in guaranteeing authenticity, and low efficiency.

[0035] Based on the aforementioned problems, this application proposes an accident report generation method. By determining the target time period in which the accident occurred and acquiring vehicle-side data reported by the accident vehicle and roadside data reported by roadside equipment within that target time period, at least one relation vector is generated based on the vehicle-side data, roadside data, and a pre-trained relational reasoning model. These relation vectors are then input into a pre-trained task prediction model for task reasoning to generate task results. These results are then input into a pre-trained report generation model, which generates and pushes the accident report. This method improves the objectivity, consistency, and authenticity of the liability determination process. Furthermore, it enables automatic capture and uploading of key data within milliseconds, avoiding the loss of crucial information due to on-site movement or human error, and improving the speed and completeness of evidence acquisition for accident liability determination.

[0036] First, the relevant system architecture involved in the accident report generation method provided in the embodiments of this application will be described in detail. Figure 1 A system architecture diagram of the vehicle information processing system used in the accident report generation method provided in this application embodiment is shown below. Figure 1 As shown, the vehicle information processing system includes: a server, multiple vehicles, and multiple roadside devices.

[0037] The server can be a cloud server. The server executes the steps of the incident report generation method provided in this application embodiment to generate and push incident reports.

[0038] Accordingly, the vehicle information processing system may also include a client. After the server generates an accident report, it pushes it to the client so that the accident report can be displayed on the client. For example, the client device can be a display device with data transmission capabilities that is close to the user, such as a mobile terminal, television, computer, or PDA.

[0039] The roadside equipment is equipped with various high-precision sensors, such as millimeter-wave radar and high-definition cameras, enabling comprehensive and dynamic real-time monitoring of the accident area. In the instant of a minor accident such as a collision, millimeter-wave radar can accurately capture key information such as the vehicle's speed, acceleration, and relative position, while high-definition cameras record images of the accident scene from multiple angles, covering details such as the collision points, damage, and the surrounding environment, overcoming the drawbacks of manual photography, such as angle deviations and missing information.

[0040] The following describes in detail the accident report generation method provided in this application with reference to several embodiments.

[0041] Figure 2A flowchart illustrating an accident report generation method provided in this application embodiment is shown below. Figure 2 As shown, the method includes: S201. Determine the target time period in which the accident occurred, and obtain vehicle-side data reported by the accident vehicle and roadside data reported by the roadside equipment during the target time period.

[0042] Optionally, when an accident occurs, the time of the accident can be determined, and the target time period in which the time of the accident occurs can be determined.

[0043] The target time period can be a period of time before and after the accident. For example, the target time period can be the 30 seconds before the accident and the 10 seconds after the accident.

[0044] Optionally, vehicle-side data reported by the accident vehicle and roadside data reported by the roadside equipment can be acquired during the target time period. The accident vehicle may include at least one vehicle.

[0045] The vehicle-side data includes driving log data, driving image data, and point cloud data, while the road-side data includes video data and traffic light data.

[0046] Specifically, driving log data refers to structured operational status information recorded by the vehicle's electronic control unit (ECU) or other onboard controllers, including: vehicle operating parameters, timestamp information, driving behavior events, and outputs from the autonomous driving system. Vehicle operating parameters include: vehicle speed, acceleration, steering angle, throttle opening, braking status, gear position, and ABS / ESP operating status. Driving behavior events include: emergency braking, rapid acceleration, lane change operations, lane departure warnings, and other trigger records. Autonomous driving system outputs include: perception and decision logs (such as target recognition results and path planning instructions).

[0047] Specifically, driving image data refers to static images or short video clips of the accident site captured by the vehicle's forward, side, or surround-view cameras. Point cloud data refers to a set of three-dimensional spatial coordinates generated by vehicle-mounted LiDAR or millimeter-wave radar, describing the geometry and relative positions of surrounding objects, and is collected by the vehicle's inertial measurement unit (IMU) in conjunction with LiDAR or radar systems.

[0048] Specifically, video data refers to a continuous video stream of the accident site recorded by roadside high-definition cameras. Traffic light data refers to the real-time output of signal phase and timing information from the road traffic signal control system.

[0049] S202. Generate at least one relation vector based on vehicle-side data, road-side data, and a pre-trained relation reasoning model.

[0050] Optionally, the vehicle-side data and roadside data can be processed and then inferred using a pre-trained relational reasoning model to obtain at least one relational vector.

[0051] Among them, the relational reasoning model can be implemented based on a multi-layer self-attention mechanism. Through this mechanism, the model performs in-depth analysis of vehicle-side and road-side data, capturing complex relationships and causal chains between different time points and data modalities. For example, the relational reasoning model can automatically learn the strong correlation between the event of "the vehicle in front braking suddenly" and "the vehicle behind colliding".

[0052] The relation vectors indicate the dependencies between participating entities at the time of the accident. Specifically, the participating entities include: vehicles, environmental elements, and perception modalities. The dependencies between participating entities include: dependencies between vehicles, temporal dependencies of vehicles themselves across time steps, dependencies between vehicles and environmental elements, and dependencies between different perception modalities.

[0053] For example, the dependencies between vehicles could be: after the vehicle in front brakes suddenly, the vehicle behind fails to brake in time. The temporal dependencies of a vehicle itself across time steps could be: a vehicle begins to decelerate at t-5 seconds, increases braking pressure at t-2 seconds, and collides at t=0. The dependencies between a vehicle and environmental factors could be: a red light illuminates, the vehicle should stop but does not. The dependencies between different perception modalities could be: a camera detects brake lights illuminating, radar data shows the speed of the vehicle in front decreasing, and the log shows that the vehicle's ECU received a braking command.

[0054] For example, the relational reasoning model can be a Transformer-based sequence modeling architecture or a Graph Neural Network (GNN).

[0055] In one example, vehicle-side data and roadside data can be preprocessed and then input into a pre-trained relational reasoning model to generate at least one relational vector.

[0056] In another example, vehicle-side data and road-side data can be fused and then input into a pre-trained relational reasoning model to obtain at least one relational vector.

[0057] S203. Input each relation vector into the pre-trained task prediction model to perform task inference and generate task results.

[0058] Optionally, each relation vector can be input into a pre-obtained task prediction model for task inference to generate task results.

[0059] The task prediction model is used to perform task reasoning on each relation vector from multiple dimensions simultaneously, obtaining task results in multiple dimensions. The task prediction model is implemented based on a multi-task prediction head, and the task results include: collision results, behavior results, and attribution results.

[0060] Specifically, collision results are used to indicate the likelihood of an accident occurring, behavioral results are used to automatically identify and classify the specific driving behavior of each vehicle in each time period during the accident, and attribution results are used to indicate the root cause of the accident.

[0061] For example, the task prediction model can be implemented based on an architecture of "shared backbone + multi-task prediction head".

[0062] S204. Input the task results into the pre-trained report generation model, which will then generate and push an accident report.

[0063] Optionally, the task results can be input into a pre-trained report generation model, which will then convert the results into an incident report that can be understood and used by the user.

[0064] The report generation model can be a large language model.

[0065] For example, the report generation model receives the task results, performs logical reasoning, integrates the analysis results, and constructs a complete causal chain of the accident. For instance, it may comprehensively determine that "vehicle A was speeding before the collision (from behavior classification)" and that "the main cause of the accident was improper operation by the driver (from attribution analysis)".

[0066] In this embodiment, by determining the target time period in which the accident occurred and acquiring vehicle-side data reported by the accident vehicle and roadside data reported by roadside equipment during the target time period, at least one relation vector is generated based on the vehicle-side data, roadside data, and a pre-trained relational reasoning model. These relation vectors are then input into a pre-trained task prediction model for task reasoning to generate task results. These results are then input into a pre-trained report generation model, which generates and pushes an accident report. This process automatically uncovers complex dependencies between vehicles, between vehicles and the environment, and between different perception modalities, establishing causal chains in a time series and intelligently attributing the causes of accidents. This improves the objectivity, consistency, and authenticity of the liability determination process, forming an immutable and traceable digital evidence chain. Furthermore, it can automatically capture and upload key data within milliseconds, avoiding the loss of crucial information due to on-site movement or human error, thus improving the speed and completeness of obtaining evidence for accident liability determination. In addition, it lowers the understanding threshold for non-professionals and significantly improves response efficiency and traffic management capabilities.

[0067] In one possible implementation, Figure 3 A flowchart illustrating the generation of at least one relation vector in the accident report generation method provided in this application embodiment is shown below. Figure 3 As shown, in step S202 above, at least one relation vector is generated based on vehicle-side data, road-side data, and a pre-trained relation inference model, including: S301. Perform data fusion on vehicle-side data and road-side data to obtain a fused feature sequence.

[0068] Optionally, vehicle-side data and roadside data can be fused to obtain a fused feature sequence.

[0069] For example, a fusion model obtained through pre-training can be used to fuse vehicle-side data and roadside data to obtain a fused feature sequence.

[0070] The fused feature sequence includes multiple fused feature vectors arranged in chronological order. The fused feature sequence is a time-ordered, high-dimensional set of feature vectors containing vehicle-road cooperative holographic information; one fused feature vector corresponds to one time step.

[0071] S302. Input the fused feature sequence into the pre-trained relation reasoning model to generate at least one relation vector.

[0072] Optionally, the fused feature sequence is input into a pre-trained relation reasoning model to perform relation reasoning, resulting in at least one relation vector.

[0073] In one possible implementation, Figure 4 This is a flowchart illustrating the process of obtaining the fused feature sequence in the accident report generation method provided in this application embodiment, with reference to... Figure 4 As shown, in S301 above, vehicle-side data and road-side data are fused to obtain a fused feature sequence, including: S401. Extract text features from driving log data using a preset text processing model to generate structured behavior labels.

[0074] Optionally, the driving log data is processed by a pre-defined text processing model to extract text features and generate structured behavior labels. The text processing model first analyzes the driving log data and extracts key information (such as "the system issues a braking command" and "the driver takes over") and converts it into structured behavior labels.

[0075] Structured behavioral labels refer to discrete labels that abstract, categorize, and standardize the driving behaviors implicit in the original logs. Structured behavioral labels represent significant driving events that occur within a specific time window.

[0076] For example, the text processing model can be a large language model or a Transformer architecture model such as BERT or RoBERTa.

[0077] S402. The driving image data is processed by a preset first image encoder to generate vehicle-side image features.

[0078] Optionally, the driving image data is input into a preset first image encoder for image encoding processing to generate vehicle-side image features. The first image encoder is a ResNet image encoder.

[0079] For example, each frame of the driving image data is input into ResNet, its high-level semantic features are extracted and processed, and finally a one-dimensional feature vector representing the global content of the image is output, that is, the vehicle-side image features.

[0080] S403. The video data is processed by a preset second image encoder to generate roadside image features.

[0081] Optionally, the video data is input into a preset second image encoder for image processing and encoding to generate roadside image features. The second image encoder is a ResNet image encoder.

[0082] For example, each frame of the video data is input into ResNet, its high-level semantic features are extracted and processed, and finally a one-dimensional feature vector representing the global content of the image is output, namely, the roadside image features.

[0083] S404. The point cloud data is encoded using a preset point cloud encoder to generate point cloud feature vectors.

[0084] Optionally, the point cloud data is input into a preset point cloud encoder for point cloud encoding processing to extract information such as the position, shape, and velocity of objects in three-dimensional space, and output a point cloud feature vector. The point cloud encoder can be SpatialLM-Llama-1B.

[0085] S405. The structured behavior labels, vehicle-end image features, road-end image features, point cloud feature vectors, and traffic light data are fused in chronological order to obtain a fused feature sequence.

[0086] Optionally, structured behavior labels, vehicle-side image features, road-side image features, point cloud feature vectors, and traffic light data are aligned by timestamps, and after feature splicing and sequence organization, they are fused in chronological order to obtain a fused feature sequence, thereby preserving the causal chain and improving the accuracy of task prediction.

[0087] The fused feature sequence is essentially a tensor with shape (number of time steps, total feature dimension).

[0088] In one possible implementation, the relation inference model includes multiple relation inference modules connected in sequence, each module being implemented based on a multi-head self-attention mechanism and a feedforward neural network. In step S302 above, the fused feature sequence is input into the pre-trained relation inference model to generate at least one relation vector, including: Each relation reasoning module sequentially performs self-attention mechanism relation reasoning on the fused feature sequence to generate at least one relation vector.

[0089] Optionally, the fused feature sequence is sequentially subjected to self-attention mechanism relation reasoning through each relation reasoning module to generate at least one relation vector.

[0090] Specifically, the relational reasoning module uses a self-attention mechanism to ensure that each time step in the sequence pays attention to information from all other time steps and other modalities, automatically learning dependencies across time, entities, and modalities. It also enhances the expressive power by performing a non-linear transformation on the attention output through a feedforward neural network, and ensures the stability of deep network training through residual connections and normalization layers. This achieves a layer-by-layer abstraction from shallow local relations to deep global causality. The relation vector is a high-dimensional semantic vector.

[0091] In one possible implementation, Figure 5 This is a schematic diagram of the task prediction model in the accident report generation method provided in this application embodiment. Figure 6This is a flowchart illustrating the process of generating task results in the accident report generation method provided in this application embodiment, with reference to... Figure 5 as well as Figure 6 As shown, the task prediction model includes: a collision prediction head, a behavior classification head, an attribution classification head, and a fusion module. In step S203 above, the relation vectors are input into the pre-trained task prediction model for task inference, generating task results, including: S601. Input each relation vector into the collision prediction head to predict the collision probability and generate the collision result.

[0092] Optionally, the collision prediction head is used to predict the probability of an accident occurring. The collision prediction head can be implemented based on a multilayer perceptron (MLP) and a sigmoid output layer, and the loss function can be a binary cross-entropy.

[0093] The collision result is used to indicate the probability of a collision. Specifically, the collision result can be a continuous value (0~1) or a binary label.

[0094] S602. Input each relation vector into the behavior classification head to perform behavior classification prediction and generate behavior results.

[0095] Optionally, the behavior classification head is used to automatically identify and classify the specific driving behavior of each vehicle in each time period during an accident. The behavior classification head can be implemented based on a multilayer perceptron (MLP) and Softmax. The loss function can be Cross-Entropy.

[0096] Among them, the behavior result is used to indicate the behavior classification of the accident vehicle. The behavior result can be a discrete category (such as "normal driving", "emergency braking", "lane change conflict" etc.).

[0097] S603. Input each relation vector into the attribution classification head to perform accident attribution prediction and generate attribution results.

[0098] Optionally, the attribution classification head is used to analyze the root causes of the accident at a deeper level. The attribution classification head can be implemented based on a multilayer perceptron (MLP) and Softmax.

[0099] The attribution result is used to indicate the cause of the accident. The attribution result can be a category of accident cause (such as "traffic light violation", "following too closely", "obstructed view", etc.).

[0100] S604. Input the collision results, behavior results, and attribution results into the fusion module to generate the task results.

[0101] Optionally, the fusion module is used to inject and fuse collision results, behavioral results, and attribution results into rules to obtain the task result. Specifically, the fusion module is used to perform logical verification in conjunction with a traffic regulations knowledge base.

[0102] By using a collision prediction head, a behavior classification head, and an attribution classification head to predict each relation vector separately, the behavior classification head can utilize the hazard patterns learned by the collision prediction head, while the attribution analysis can refer to the results of the behavior classification. Furthermore, the derived "behavior," "risk," and "cause" are logically consistent and unlikely to contradict each other. In other words, the collision prediction head, behavior classification head, and attribution classification head co-evolve to jointly improve accuracy. A fusion module then integrates the collision results, behavior results, and attribution results to obtain the task result, consolidating the results of three independent tasks into a unified result and improving the interpretability of the obtained task result.

[0103] In one possible implementation, Figure 7 This is a schematic diagram of a report generation model in the accident report generation method provided in this application embodiment. Figure 8 This is a schematic diagram illustrating a process in the accident report generation method provided in this application, where an accident report is generated and pushed by a report generation model. (Refer to...) Figure 7 as well as Figure 8 As shown, the report generation model includes: a scenario construction module, a responsibility weight assessment module, and a text generation module; in step S204 above, the task results are input into the pre-trained report generation model, which then generates and pushes the accident report, including: S801. Input the task results into the scenario construction module to construct the scenario and generate an accident semantic graph.

[0104] Optionally, the scenario construction module combines a pre-defined timeline, spatial relationships, and traffic rule knowledge base to construct a structured accident semantic graph. The accident semantic graph includes multiple nodes and multiple edges. Specifically, the nodes can be participating entities such as vehicles, traffic lights, and road environment elements, while the edges can represent behavioral dependencies, causal relationships, and temporal sequences between entities. The accident semantic graph is a high-level abstract representation of the entire accident process.

[0105] Accident semantic graphs can provide a logical framework for subsequent responsibility delineation and text generation.

[0106] S802. Input the task results and accident semantic graph into the responsibility weight assessment module to generate the responsibility weight assessment results.

[0107] Optionally, the responsibility weight assessment module includes a built-in rule engine for quantitative analysis of multiple preset factors, outputting the responsibility weight of each party involved. These preset factors include: the degree of violation (e.g., speeding, running a red light, improper lane changing); causal contribution (the intensity of a behavior's impact on the accident); compliance with traffic regulations (compare with relevant clauses); and historical data reference (calling a database of similar cases for comparative judgment).

[0108] For example, the liability weights could be: Vehicle A - secondary liability (30%); Vehicle B - primary liability (70%).

[0109] The responsibility weight assessment module enables objective and interpretable responsibility determination, thereby avoiding subjective assumptions.

[0110] S803. Input the accident semantic graph and the responsibility weight assessment results into the text generation module to generate and push the accident report.

[0111] Optionally, the text generation module is implemented using a large language model (LLM) or a template-based NLG (natural language generation) system. The text generation module is used to organize and refine the accident semantic graph and responsibility weight assessment results according to a preset standard report format, perform compliance verification, generate and push the accident report.

[0112] For example, the text generation module can be a Qwen3-max model.

[0113] In one possible implementation, before determining the target time period in S201 where the accident occurred, the following step is also included: Construct a sensing network and data channels.

[0114] Optionally, the sensing network includes both vehicle-side and roadside-side terminals.

[0115] In the vehicle-side, taking autonomous vehicles as an example, autonomous vehicles are equipped with a series of onboard sensors, such as an inertial measurement unit (IMU) to measure motion parameters like acceleration and angular velocity in real time, accurately recording changes in the vehicle's attitude. Wheel speed sensors monitor wheel rotation speed in real time to obtain the vehicle's actual speed. Furthermore, onboard cameras are positioned at the front, rear, and sides of the vehicle to ensure coverage of all directions around it. When an accident occurs, these cameras automatically capture photos and videos at high frame rates, recording crucial information such as the vehicle's position, the point of impact, and the surrounding environment at the moment of the accident. Simultaneously, the vehicle's electronic control unit (ECU) records various operational data in real time, such as engine speed, accelerator pedal opening, and brake pressure, storing this data in onboard non-volatile memory for later upload to the cloud.

[0116] At the roadside end, multiple Roadside Units (RSUs) are strategically distributed along the road. Each RSU integrates various sensors, including millimeter-wave radar, lidar, and high-definition cameras. Millimeter-wave radar provides high-precision distance and velocity measurement capabilities, enabling real-time monitoring of the position, speed, and trajectory of all vehicles within the accident area. LiDAR, by emitting laser beams and receiving reflected signals, constructs a 3D point cloud model of the accident scene, providing strong support for accurate reconstruction of the accident scenario. High-definition cameras monitor the accident scene from different angles, capturing the entire process of the accident. The data collected by these roadside sensors is transmitted in real-time to a cloud server via fiber optic or wireless communication networks.

[0117] Optionally, a data channel is constructed, in which the in-vehicle equipment uploads stored real-time vehicle driving data and captured photos, videos, and other information to a cloud server via cellular networks (such as 4G or 5G) or Dedicated Short Range Communication (DSRC) technology. To ensure the reliability and stability of data transmission, data encryption and verification mechanisms are employed to prevent data from being tampered with or lost during transmission.

[0118] In this process, data collected by roadside units is transmitted to nearby edge computing nodes via fiber optic networks or wireless mesh networks. The edge computing nodes perform preliminary preprocessing on the data, such as data filtering and feature extraction, to reduce the data volume before uploading the processed data to a cloud server. The cloud server, acting as a data aggregation center, receives all data from both vehicle and roadside units and performs unified storage and management.

[0119] In one possible implementation, the process of collecting vehicle-side data includes: Point cloud data of each vehicle is collected by the vehicle's LiDAR; driving image data of each vehicle is collected by the vehicle's camera unit; driving log data of each vehicle is collected by the vehicle's electronic control unit; and the point cloud data, driving image data and driving log data of each vehicle are transmitted to the server through the data channel.

[0120] Optionally, the vehicle's lidar uses laser pulses to measure distance, emitting hundreds of thousands to millions of laser beams per second, and after receiving the reflected signals, calculates the time difference to generate point coordinates (x, y, z) in three-dimensional space. The output result is point cloud data.

[0121] Optionally, driving image data for each vehicle can be collected through the vehicle's camera unit. The camera unit includes: a forward-facing monocular / dual-lens camera, a surround-view camera, a side blind spot monitoring camera, and a driver monitoring camera. The driving image data includes: road scenes (e.g., lane lines, traffic signs, traffic lights), dynamic targets (e.g., vehicles, pedestrians, non-motorized vehicles), and driver status (e.g., fatigue, distraction).

[0122] Optionally, driving log data for each vehicle can be collected via the vehicle's electronic control unit (ECU). The ECU includes: Engine Control Module (ECM), Brake Control Module (BCM), Body Control Module (BCM), ADAS Domain Controller, and Gateway ECU. The sources and content of the driving log data include: vehicle speed, engine speed, and throttle opening from the powertrain ECU; brake pedal status and ABS trigger records from the brake ECU; steering wheel angle and EPS operating mode from the steering ECU; autonomous driving mode switching and decision commands from the ADAS controller; and CAN bus communication logs and fault codes (DTCs) from the gateway ECU. This data is typically aggregated to a central gateway via CAN / LIN / FlexRay bus and packaged uniformly by a T-Box or data recorder.

[0123] Optionally, point cloud data, driving image data, and driving log data of each vehicle can be transmitted to the server via a data channel. The data channel includes: physical layer: 4G / 5G / NB-IoT / C-V2X wireless communication module; protocol layer: TCP / IP, MQTT, HTTP / HTTPS, CoAP, etc.; security mechanisms: TLS encryption, authentication, and data anonymization (e.g., facial blurring); and device carrier: T-Box (Telematics Box) or vehicle gateway.

[0124] In one possible implementation, the process of collecting roadside data includes: Video data and traffic light data of each vehicle during its journey are collected by multiple roadside devices, each of which includes millimeter-wave radar, lidar, and high-definition cameras.

[0125] Optionally, roadside equipment can be installed at key road sections such as intersections, highway sections, tunnel entrances and exits, and bus lanes to compensate for blind spots in vehicle-mounted perception, provide overall traffic situation information, and support decision-making for autonomous vehicles.

[0126] Based on the same inventive concept, this application also provides an accident report generation device corresponding to the accident report generation method. Since the principle of the device in this application is similar to the accident report generation method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0127] Reference Figure 9 As shown, Figure 9 This is a schematic diagram of an accident report generation device provided in an embodiment of this application. It is applied to the server side of a vehicle information processing system. The vehicle information processing system includes: a server side, multiple vehicles, and multiple roadside devices. The device includes: a determination module 901, a relationship reasoning module 902, a task reasoning module 903, and a report generation module 904. The determination module 901 is used to determine the target time period in which the accident occurred, and to acquire vehicle-side data reported by the accident vehicle and roadside data reported by the roadside equipment during the target time period. The vehicle-side data includes driving log data, driving image data and point cloud data, and the roadside data includes video data and traffic light data. The relation reasoning module 902 is used to generate at least one relation vector based on vehicle-side data, road-side data and a pre-trained relation reasoning model. Each relation vector is used to indicate the dependency relationship between the participating entities when an accident occurs. The task reasoning module 903 is used to input the relation vectors into the pre-trained task prediction model to perform task reasoning and generate task results. The task prediction model is based on a multi-task prediction head, and the task results include: collision results, behavior results, and attribution results. The report generation module 904 is used to input the task results into the pre-trained report generation model, which then generates and pushes the accident report.

[0128] Optionally, the relational reasoning module 902 is specifically used for: Vehicle-side data and roadside data are fused to obtain a fused feature sequence, which includes multiple fused feature vectors arranged in chronological order. The fused feature sequence is input into the pre-trained relation reasoning model to generate at least one relation vector.

[0129] Optionally, the relational reasoning module 902 is specifically used for: The driving log data is processed using a pre-defined text processing model to extract text features and generate structured behavior labels. The driving image data is processed by a preset first image encoder to generate vehicle-side image features; The video data is processed by a preset second image encoder to generate roadside image features; Point cloud data is encoded using a preset point cloud encoder to generate point cloud feature vectors. Structured behavior labels, vehicle-side image features, roadside image features, point cloud feature vectors, and traffic light data are fused in chronological order to obtain a fused feature sequence.

[0130] Optionally, the relational reasoning model includes multiple relational reasoning modules 902 connected in sequence, each relational reasoning module 902 being implemented based on a multi-head self-attention mechanism and a feedforward neural network; the relational reasoning module 902 is specifically used for: Each relation reasoning module 902 sequentially performs self-attention mechanism relation reasoning on the fused feature sequence to generate at least one relation vector.

[0131] Optionally, the task prediction model includes: a collision prediction head, a behavior classification head, an attribution classification head, and a fusion module; the relation vectors are input into the pre-trained task prediction model for task inference to generate task results, including: Each relation vector is input into the collision prediction head to predict the collision probability and generate collision results. The collision results are used to indicate the probability of a collision. Each relation vector is input into the behavior classification head to perform behavior classification prediction and generate behavior results. The behavior results are used to indicate the behavior classification of the accident vehicle. Each relation vector is input into the attribution classification head to predict the cause of the accident and generate attribution results. The attribution results are used to indicate the cause of the accident. The collision results, behavior results, and attribution results are input into the fusion module to generate the task results.

[0132] Optionally, the report generation model includes: a scenario construction module, a responsibility weight assessment module, and a text generation module; the report generation module 904 is specifically used for: The task results are input into the scenario construction module to construct a scenario and generate an accident semantic graph. Input the task results and the accident semantic graph into the responsibility weight assessment module to generate the responsibility weight assessment results; Input the accident semantic graph and the responsibility weight assessment results into the text generation module to generate and push the accident report.

[0133] Optionally, before determining the target time period in which the accident occurred, the following steps are also included: Construct a sensing network and data channels.

[0134] Optionally, the process of collecting vehicle-side data includes: Point cloud data of each vehicle is collected using the vehicle's lidar. The vehicle's camera unit collects driving image data for each vehicle; The driving log data of each vehicle is collected through the vehicle's electronic control unit; The point cloud data, driving image data, and driving log data of each vehicle are transmitted to the server through the data channel.

[0135] Optionally, the process of collecting roadside data includes: Video data and traffic light data of each vehicle during its journey are collected by multiple roadside devices, each of which includes millimeter-wave radar, lidar, and high-definition cameras.

[0136] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0137] This application also provides a server-side implementation, such as... Figure 10 As shown, Figure 10 A schematic diagram of a server architecture provided in an embodiment of this application includes: a processor 1001, a memory 1002, and optionally, a bus 1003. The memory 1002 stores machine-readable instructions executable by the processor 1001 (e.g., ...). Figure 9 The device in the middle determines the execution instructions corresponding to the module 901, the relationship reasoning module 902, the task reasoning module 903, and the report generation module 904. When the server is running, the processor 1001 and the memory 1002 communicate through the bus 1003. When the machine-readable instructions are executed by the processor 1001, the steps of the above-mentioned accident report generation method are executed.

[0138] This application also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the above-described incident report generation method.

[0139] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here. In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces; the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms.

[0140] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0141] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for generating accident reports, characterized in that, A server-side application in a vehicle information processing system, the vehicle information processing system comprising: the server-side application, multiple vehicles, and multiple roadside devices, the method comprising: Determine the target time period in which the accident occurred, and acquire vehicle-side data reported by the accident vehicle and roadside data reported by the roadside equipment during the target time period. The vehicle-side data includes driving log data, driving image data, and point cloud data, while the roadside data includes video data and traffic light data. Based on the vehicle-side data, the road-side data, and the pre-trained relational reasoning model, at least one relation vector is generated, and each relation vector is used to indicate the dependency relationship between the participating entities when an accident occurs. Each of the aforementioned relation vectors is input into a pre-trained task prediction model for task inference to generate task results. The task prediction model is based on a multi-task prediction head, and the task results include: collision results, behavior results, and attribution results. The task results are input into a pre-trained report generation model, which then generates and pushes an accident report.

2. The accident report generation method according to claim 1, characterized in that, The step of generating at least one relation vector based on the vehicle-side data, the roadside data, and a pre-trained relation reasoning model includes: The vehicle-side data and road-side data are fused to obtain a fused feature sequence, which includes multiple fused feature vectors arranged in chronological order. The fused feature sequence is input into a pre-trained relational reasoning model to generate at least one relational vector.

3. The accident report generation method according to claim 2, characterized in that, The process of fusing the vehicle-side data and roadside data to obtain a fused feature sequence includes: The driving log data is processed using a pre-defined text processing model to extract text features and generate structured behavior tags. The driving image data is image encoded by a preset first image encoder to generate vehicle-side image features. The video data is processed by a preset second image encoder to generate roadside image features; The point cloud data is encoded using a preset point cloud encoder to generate a point cloud feature vector. The structured behavior labels, vehicle-side image features, roadside image features, point cloud feature vectors, and traffic light data are fused in chronological order to obtain the fused feature sequence.

4. The accident report generation method according to claim 2, characterized in that, The relational reasoning model includes multiple relational reasoning modules connected in sequence, and each relational reasoning module is implemented based on a multi-head self-attention mechanism and a feedforward neural network. The step of inputting the fused feature sequence into a pre-trained relation inference model to generate at least one relation vector includes: Each of the aforementioned relation reasoning modules sequentially performs self-attention mechanism relation reasoning on the fused feature sequence to generate at least one relation vector.

5. The accident report generation method according to claim 1, characterized in that, The task prediction model includes: a collision prediction head, a behavior classification head, an attribution classification head, and a fusion module; The step of inputting each of the aforementioned relation vectors into a pre-trained task prediction model for task inference and generating task results includes: Each of the aforementioned relation vectors is input into the collision prediction head to predict the collision probability and generate a collision result, which is used to indicate the probability of a collision occurring. Each of the aforementioned relation vectors is input into the behavior classification head to perform behavior classification prediction and generate behavior results, which are used to indicate the behavior classification of the accident vehicle. Each of the aforementioned relation vectors is input into the attribution classification head to perform accident attribution prediction and generate attribution results, which are used to indicate the cause of the accident. The collision results, the behavior results, and the attribution results are input into the fusion module to generate task results.

6. The accident report generation method according to claim 1, characterized in that, The report generation model includes: a scenario construction module, a responsibility weight assessment module, and a text generation module; The step of inputting the task results into a pre-trained report generation model, and having the report generation model generate and push an accident report, includes: The task results are input into the scenario construction module to construct a scenario and generate an accident semantic graph. The task results and the accident semantic graph are input into the responsibility weight assessment module to generate responsibility weight assessment results; The accident semantic graph and the responsibility weight assessment results are input into the text generation module to generate and push the accident report.

7. The accident report generation method according to claim 1, characterized in that, Before determining the target time period in which the accident occurred, the method further includes: Construct a sensing network and data channels.

8. The accident report generation method according to claim 1, characterized in that, The process of collecting vehicle-side data includes: Point cloud data of each vehicle is collected using the vehicle's lidar. Driving image data of each vehicle is acquired through the vehicle's camera unit; The driving log data of each vehicle is collected through the vehicle's electronic control unit; The point cloud data, driving image data, and driving log data of each vehicle are transmitted to the server via a data channel.

9. The accident report generation method according to claim 1, characterized in that, The process of collecting roadside data includes: Video data and traffic light data of each vehicle during its journey are collected by multiple roadside devices, each of which includes millimeter-wave radar, lidar, and high-definition cameras.

10. A vehicle information processing system, characterized in that, include: Server-side equipment, multiple vehicles, and multiple roadside devices; The server is used to execute the accident report generation method according to any one of claims 1-9.

Images