Simple traffic accident liability intelligent affirmation system and method based on AI large model

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a simple traffic accident liability intelligent affirmation system and method based on an AI large model, and the method comprises the steps: collecting an accident scene image and an accident description text of a simple traffic accident scene, and carrying out the feature extraction of the accident scene image and the accident description text, obtaining accident key parameters and accident semantics; analyzing the accident space-time field quantity of the simple traffic accident scene, and constructing an accident space-time panorama by using the accident space-time field quantity; constructing an accident association map of a simple traffic accident scene, and analyzing an accident logic chain by using the accident association map; performing accident liability analysis on the simple traffic accident scene based on the accident space-time panorama and the accident logic chain to obtain an accident analysis text; and performing accident judgment optimization and intelligent conclusion calibration on the accident liability text to obtain a target accident text, and constructing an accident liability report by using the target accident text. According to the invention, the efficiency of simple traffic accident liability determination can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a simple traffic accident responsibility intelligent identification system and method based on an AI large model. BACKGROUND

[0002] A large number of minor traffic accidents occur in China every year, such as rear-end collisions and scratches, and a considerable part of them meet the conditions for "simple procedure handling". However, the traditional accident handling method still relies on on-site disposal by traffic police or negotiation between parties, which has problems such as low efficiency, complex process, and strong subjectivity in responsibility identification, resulting in a high rate of reconsideration. With the continuous development of artificial intelligence technology, it is possible to extract key information from accident scene photos and text descriptions and conduct intelligent analysis.

[0003] Currently, traffic accident responsibility identification mainly relies on on-site investigation by traffic police, witness testimony, and evidence materials such as monitoring video, and then comprehensive judgment is made according to traffic regulations and experience. However, the traditional traffic accident responsibility identification method has a long identification process, resulting in low efficiency of traffic accident responsibility identification. SUMMARY

[0004] The present application provides a simple traffic accident responsibility intelligent identification system and method based on an AI large model, which aims to improve the efficiency of simple traffic accident responsibility identification.

[0005] To achieve the above purpose, the simple traffic accident responsibility intelligent identification system based on an AI large model provided by the present application comprises a feature extraction module, a scene panoramic map construction module, an accident logic analysis module, an accident responsibility analysis module, and a simple accident report module. The feature extraction module is used to collect accident scene images and accident description texts of a simple traffic accident scene, and to extract features from the accident scene images and the accident description texts using a trained AI large model to obtain accident key parameters and accident semantics. The scene panoramic map construction module is used to analyze the accident space-time field of the simple traffic accident scene using the accident key parameters, and to construct an accident space-time panoramic map of the simple traffic accident scene using the accident space-time field. The accident logic analysis module is used to construct an accident correlation graph of the simple traffic accident scene using the accident semantics, and to analyze the accident logic chain of the simple traffic accident scene using the accident correlation graph. The accident responsibility analysis module is used to analyze the accident responsibility of the simple traffic accident scene based on the accident space-time panoramic map and the accident logic chain, and to obtain an accident analysis text. The simple accident report module is used for querying historical accident judgment data corresponding to the simple traffic accident scene, performing accident judgment optimization on the accident responsibility text based on the historical accident judgment data, obtaining an optimized accident text, performing intelligent conclusion calibration on the optimized accident text, obtaining a target accident text, and constructing an accident responsibility report of the simple traffic accident scene by using the target accident text.

[0006] Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters.

[0007] Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters.

[0008] Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters.

[0009] Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. Optionally, the accident key parameters are subjected to spatio-temporal coordinate system unification processing to obtain standardized parameters. performing critical path screening on the complete path set to obtain a critical causal path; performing attribution analysis on the critical causal path to obtain a responsibility distribution path; performing natural language conversion on the responsibility distribution path to obtain an accident logic chain.

[0010] Optionally, the performing critical path screening on the complete path set to obtain a critical causal path comprises: calculating a node importance score of a path node in the complete path set by using the following formula: ; wherein, denotes the node importance score of node i, denotes the total number of nodes in the complete path set, and d denotes a damping coefficient, denotes the jth node in the path node, denotes a set of nodes pointing to node i in the path node, denotes the importance weight of node j, denotes the number of outlinks of node j; performing critical path screening on the complete path set based on the node importance score to obtain a critical causal path.

[0011] Optionally, based on the accident spatiotemporal panorama and the accident logic chain, performing accident responsibility analysis on the simple traffic accident scene to obtain an accident analysis text comprises: performing feature fusion on the accident spatiotemporal panorama and the accident logic chain to obtain a joint feature vector; performing multi-factor weighting on the joint feature vector to obtain a responsibility intensity vector; constructing a responsibility distribution map of the simple traffic accident scene by using the responsibility intensity vector; performing legal annotation on the responsibility distribution map to obtain an annotated responsibility distribution map; constructing an accident analysis text of the simple traffic accident scene by using the annotated responsibility distribution map.

[0012] Optionally, the performing multi-factor weighting on the joint feature vector to obtain a responsibility intensity vector comprises: querying a behavior vector in the joint feature vector; calculating a behavior responsibility score of the behavior vector by using the following formula: ; wherein, denotes the behavior responsibility score, denotes the number of behavior vectors, a responsibility weight of a behavior vector k, a behavior credibility of a behavior vector k, a causal correlation degree of a behavior vector k. based on the behavior responsibility score, multi-factor weighting is performed on the joint feature vector to obtain a responsibility intensity vector.

[0013] The behavior responsibility score refers to a quantitative index that comprehensively measures the influence degree of each behavior on responsibility determination in an accident.

[0014] Optionally, based on the historical accident judgment data, the accident responsibility text is optimized for accident judgment to obtain an optimized accident text, including: aligning the historical accident judgment data and the accident responsibility text by features to obtain a comparable feature set; detecting responsibility deviation of the comparable feature set to obtain a responsibility deviation report; based on the responsibility deviation report, the accident responsibility text is corrected by responsibility weight to obtain an optimized accident text.

[0015] The simple traffic accident responsibility intelligent identification method based on an AI large model, characterized in that the method comprises: collecting accident scene images and accident description texts of a simple traffic accident scene, using a trained AI large model to extract features of the accident scene images and the accident description texts to obtain accident key parameters and accident semantics; using the accident key parameters to analyze the accident space-time field quantity of the simple traffic accident scene, and using the accident space-time field quantity to construct an accident space-time panoramic view of the simple traffic accident scene; using the accident semantics to construct an accident correlation graph of the simple traffic accident scene, and using the accident correlation graph to analyze an accident logic chain of the simple traffic accident scene; based on the accident space-time panoramic view and the accident logic chain, the simple traffic accident scene is analyzed for accident responsibility to obtain an accident analysis text; querying historical accident judgment data corresponding to the simple traffic accident scene, based on the historical accident judgment data, the accident responsibility text is optimized for accident judgment to obtain an optimized accident text, the optimized accident text is calibrated for intelligent conclusion to obtain a target accident text, and the target accident text is used to construct an accident responsibility report of the simple traffic accident scene.

[0016] Compared with the prior art, the application first uses the trained AI large model to collect accident scene images and description texts, and extracts key parameters and semantics, to realize digital collection and analysis of accident information, and to convert unstructured data such as images and texts into structured information such as vehicle coordinates, collision angles and illegal behaviors, to provide accurate data basis for subsequent analysis; the application analyzes the space-time field quantity based on the key parameters of the accident, and then constructs a space-time panoramic map, to present the spatial layout of the accident scene, the vehicle motion trajectory and the mechanical action process in a visual and structured form, thereby helping users to intuitively understand the physical process of the accident and providing intuitive visual support for responsibility analysis; the application further constructs a correlation graph based on the semantics of the accident and analyzes the logic chain, to convert semantic information such as behavior elements and causal relationships into a structured network graph, thereby clarifying the internal logic of the occurrence and development of the accident; the application further analyzes the responsibility by combining the space-time panoramic map and the logic chain, combines the physical process and the causal logic, calculates the responsibility strength vector through feature fusion and multi-factor weighting, constructs a responsibility distribution map and labels the legal basis, and finally generates an accident analysis text, realizes the quantification and visualization of responsibility determination, comprehensively considers the physical characteristics and behavior semantics of the accident, scientifically and accurately determines the responsibility proportion of each participant, and provides intuitive and scientific basis for responsibility determination; further, the application optimizes and calibrates the accident responsibility text by querying historical accident judgment data, corrects the responsibility deviation by comparing with historical cases, generates an optimized accident text, and obtains a target accident text through intelligent calibration, finally constructs an accident responsibility report, improves the accuracy of responsibility determination by using historical experience, ensures that the report logic is rigorous and the expression is standard, forms a standardized file with legal effect, and provides reliable support for accident handling and subsequent tracing. Therefore, the application can improve the efficiency of simple traffic accident responsibility determination. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A functional module diagram of an AI large model-based simple traffic accident responsibility intelligent determination system provided by an embodiment of the application is shown in the figure. Figure 2 A flowchart of an AI large model-based simple traffic accident responsibility intelligent determination method provided by an embodiment of the application is shown in the figure. Figure 3 An accident correlation graph diagram in an AI large model-based simple traffic accident responsibility intelligent determination method provided by an embodiment of the application is shown in the figure. The implementation, functional characteristics and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0018] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0019] In addition, the step sequence in each of the following method embodiments is only an example and is not strictly limited.

[0020] In fact, the server device deployed by the simple traffic accident liability intelligent identification system based on the AI large model may be composed of one or more devices. The simple traffic accident liability intelligent identification system based on the AI large model can be implemented as a business instance, a virtual machine, or a hardware device. For example, the simple traffic accident liability intelligent identification system based on the AI large model can be implemented as a business instance deployed on one or more devices in a cloud node. In short, the simple traffic accident liability intelligent identification system based on the AI large model can be understood as a software deployed on a cloud node, which is used to provide the simple traffic accident liability intelligent identification service based on the AI large model for each user terminal. Alternatively, the simple traffic accident liability intelligent identification system based on the AI large model can also be implemented as a virtual machine deployed on one or more devices in a cloud node. The virtual machine has an application software installed therein for managing each user terminal. Alternatively, the simple traffic accident liability intelligent identification system based on the AI large model can also be implemented as a server composed of a plurality of same or different types of hardware devices, and one or more hardware devices are arranged to provide the simple traffic accident liability intelligent identification service based on the AI large model for each user terminal.

[0021] In terms of implementation form, the simple traffic accident liability intelligent identification system based on the AI large model and the user terminal are mutually adaptive. That is, the simple traffic accident liability intelligent identification system based on the AI large model is installed as an application on a cloud service platform, and the user terminal is a client that establishes a communication connection with the application; or the simple traffic accident liability intelligent identification system based on the AI large model is implemented as a website, and the user terminal is implemented as a webpage; or the simple traffic accident liability intelligent identification system based on the AI large model is implemented as a cloud service platform, and the user terminal is implemented as an applet in an instant messaging application.

[0022] Referring to Figure 1 Fig. 1 shows a functional module diagram of the simple traffic accident liability intelligent identification system based on the AI large model according to an embodiment of the present application.

[0023] The simple traffic accident liability intelligent identification system 100 based on the AI large model can be set in a cloud server, and in terms of implementation, can be used as one or more service devices, can be installed as an application on the cloud (such as a server for monitoring the intestinal nutrition of a critically ill patient, a server cluster, etc.), or can also be developed as a website. According to the functions implemented, the simple traffic accident liability intelligent identification system 100 based on the AI large model comprises a feature extraction module 101, a scene panoramic map construction module 102, an accident logic analysis module 103, a responsibility analysis module 104, and a simple accident report module 105.

[0024] In the embodiment of the present application, the simple traffic accident liability intelligent identification based on the AI large model can be independently implemented and called by other modules. The calling here can be understood as that a module can be connected to multiple modules of another type and provide corresponding services for the connected multiple modules. In the simple traffic accident liability intelligent identification system based on the AI large model provided by the embodiment of the present application, the application range of the simple traffic accident liability intelligent identification architecture based on the AI large model can be adjusted by increasing modules and directly calling without modifying program codes, the cluster type horizontal expansion is realized, and the purpose of quickly and flexibly expanding the simple traffic accident liability intelligent identification system based on the AI large model is achieved. In actual application, the above modules can be set in the same device or different devices, or can be set in a virtual device, such as a service instance in a cloud server.

[0025] The following will describe the components and specific work flow of the simple traffic accident liability intelligent identification system based on the AI large model with reference to specific embodiments.

[0026] The feature extraction module 201 is used to collect accident scene images and accident description texts of a simple traffic accident scene, extract features of the accident scene images and the accident description texts by using a trained AI large model, and obtain accident key parameters and accident semantics.

[0027] In the embodiment of the present application, the visual information (such as vehicle position, collision form, and road environment) and the text information (such as accident process and behavior description) of the accident scene can be obtained by collecting the accident scene images and the accident description texts of the simple traffic accident scene, the digital collection of the accident scene information is realized, and the basic data support for accident liability is provided.

[0028] The simple traffic accident scene refers to an accident scene that meets the "simple procedure processing" in the "Road Traffic Accident Handling Procedure", and usually has the following characteristics: only property damage or minor personal injury, no death or serious injury, no dispute over facts and causes, and the accident form is mostly rear-end, scratching, minor collision and other types that can be quickly handled. The accident scene image refers to visual data of the accident scene collected by cameras, mobile phones and other devices, such as the overall environment of the accident scene (such as road direction, traffic signs, relative position of vehicles), collision part details, license plate number and road marks, etc. The accident description text refers to the written information recording the accident, such as the operation behavior at the time of the accident (such as "changing lanes", "braking"), subjective feelings (such as "the other party suddenly rushed out"); the accident brief, environmental conditions such as weather / light, and the preliminary observation of the responsibility tendency when handling the scene.

[0029] Optionally, the accident scene image is obtained by using a mobile phone, camera or other device to take multi-angle photos of the accident scene panorama, collision part close-up and road environment, or by automatically capturing real-time images of the accident scene by intelligent traffic monitoring equipment; the accident scene image and the accident description text can be obtained by manually inputting the accident information, time and location by the on-site personnel through the accident handling APP, or by automatically extracting the voice data of the vehicle data recorder, the on-site recording text of the police and other structured descriptions.

[0030] Further, the embodiment of the present application extracts features from the accident scene image and the accident description text by using a trained AI large model to obtain accident key parameters and accident semantics, which can convert the original abstract unstructured data into accident key parameters (such as vehicle coordinates, collision angle) and semantic information (such as behavior description, cause and effect relationship) that can be understood by intelligent systems, thereby realizing digital analysis of accident information.

[0031] The trained AI large model refers to a multi-modal artificial intelligence model pre-trained based on massive traffic accident data (including images, texts and historical cases, etc.), which has cross-modal feature extraction and semantic understanding capability. Common types include: multi-modal large model (such as CLIP, ALBEF, which can realize joint representation of image visual features and text semantics), large language model (such as GPT series, BERT, which can analyze the semantic logic and event elements in the text). The accident key parameters refer to the structured physical characteristics and spatial information extracted from the accident image, such as vehicle characteristic parameters (position coordinates, collision points and vehicle attitude, etc.), road environment parameters (traffic sign types, marking attributes and road surface conditions, etc.) and space-time parameters (image shooting time, vehicle trajectory key points, etc.). The accident semantics refer to the semantic level event elements and logical relationships extracted from the accident description text, such as behavior semantics: "changing lanes", "exceeding speed", "running a red light" and other driving behavior descriptions, cause and effect semantics: "collision due to not yielding", "brake failure due to slippery road in rainy weather" and other cause and effect logic, time sequence semantics: "3 seconds before the collision, car A began to decelerate", "after car B passed through the intersection, a scratch occurred" and other time sequence relationships.

[0032] Optionally, the accident key parameters can be extracted by computer vision algorithm (such as YOLO) to extract vehicle position, collision trace, road sign and other key parameters from accident scene image, and natural language processing technology (such as NER) is used to analyze accident cause, responsible party and violation behavior and other accident semantics from accident description text.

[0033] The on-site panoramic graph construction module 202 is used for analyzing the accident space-time field quantity of the simple traffic accident scene by using the accident key parameters, and constructing the accident space-time panoramic graph of the simple traffic accident scene by using the accident space-time field quantity.

[0034] By analyzing the accident space-time field quantity of the simple traffic accident scene by using the accident key parameters, the physical modeling of the vehicle position, motion trajectory and time sequence relationship of the accident scene can be performed, and the digital field quantity representation containing spatial topology and time evolution is formed, so that the user can analyze the physical variable relationship of the accident field.

[0035] The accident space-time field quantity refers to a set of physical quantities extracted from the accident key parameters, which fuse spatial dimensions and time dimensions, such as spatial field quantity (vehicle three-dimensional coordinates, collision point spatial vector and road environment topological relationship) and time field quantity (accident behavior timestamp, vehicle motion time sequence and trajectory change rate, etc.).

[0036] As an embodiment of the present application, the accident space-time field quantity of the simple traffic accident scene is analyzed by using the accident key parameters, including: The accident key parameter is subjected to space-time coordinate systematization processing to obtain a standardized parameter; A kinematic field modeling is performed on the simple traffic accident scene by using the standardized parameter to obtain a velocity field; A stress field of the simple traffic accident scene is calculated based on the velocity field; The velocity field and the stress field are fused to obtain an accident space-time field quantity.

[0037] The velocity field refers to a vector field reflecting the vehicle motion velocity distribution constructed on the simple traffic accident scene after processing the accident key parameter, and the stress field refers to a physical field describing the stress distribution of the vehicle and related objects in the collision process in the simple traffic accident scene calculated based on the velocity field.

[0038] In the implementation process, the pixel coordinates (such as 2D image coordinates captured by a mobile phone) of the accident image can be mapped to the world coordinate system through a perspective transformation matrix, and then the GPS coordinates and the image coordinates are robustly matched to eliminate the coordinate deviation of the multi-source data and obtain the standardized parameters. According to the position information of the vehicle at different times in the key parameters of the accident, modeling is carried out in combination with the kinematic principle. Taking the linear acceleration motion of the vehicle as an example, it is assumed that at time t1, the vehicle is located at position (x1, y1), and at time t2, it is located at position (x2, y2). According to the definition of velocity, v = (x2 - x1) / (t2 - t1), the average speed in this time period can be calculated. In actual operation, the position data obtained can be processed by using the Kalman filtering algorithm to remove noise interference, and then a continuous position-time function is constructed based on the known discrete position points by using an interpolation algorithm such as cubic spline interpolation, and then the derivative of the function is obtained to obtain the instantaneous speed of the vehicle at any time. The speed information of different vehicles at different positions and times is integrated to form a speed field of the accident scene. According to the law of conservation of momentum, the impact force at the moment of collision is calculated in combination with the vehicle mass and other information. Assuming that the masses of the two vehicles are m1 and m2, the velocities before collision are v1 and v2, and the common velocity after collision is v, according to m1v1 + m2v2 = (m1 + m2) v, the velocity after collision can be calculated, and then the velocity change Δv during the collision process is obtained. According to the impulse theorem FΔt = mΔv, the collision impact force F can be estimated. For the mechanical properties of the vehicle material, the elastic modulus, Poisson's ratio and other parameters can be obtained from the material manual, and the finite element analysis software such as ANSYS is used to divide the vehicle structure into numerous small units, and the calculated impact force is applied to the corresponding unit to simulate the propagation and distribution of stress in the vehicle structure, and finally the stress field of the accident scene is generated. By using tensor fusion technology, the velocity field data in the form of vector (including speed size and direction) and the scalar form of stress field data (stress value) are set according to the importance of the two in describing the physical process of the accident, such as setting the weight of the velocity field to 0.6 and the weight of the stress field to 0.4. The velocity field vector and the stress field scalar are weighted and summed in the four-dimensional space-time dimension to form a comprehensive space-time field matrix Further, by using the accident space-time field quantity, the accident space-time panoramic graph of the simple traffic accident scene can be constructed, which can convert the field quantity containing physical information such as velocity distribution and stress state into a visual and structured accident scene expression, and facilitate users or systems to intuitively understand the spatial layout of the accident scene, vehicle motion trajectory and mechanical action process.

[0039] The accident space-time panoramic graph is a comprehensive graph formed by integrating the spatial layout of the accident scene, the vehicle motion trajectory, the mechanical action at the moment of collision and other information through visualization technology based on the accident space-time field quantity.

[0040] As an embodiment of the application, the accident space-time panorama of the simple traffic accident scene is constructed by using the accident space-time field quantity, including: The accident space-time field quantity is spatio-temporally sliced to obtain a key frame field quantity subset; Based on the key frame field quantity subset, the simple traffic accident scene is subjected to multi-physical field fusion to obtain a composite field quantity; By using the composite field quantity, the simple traffic accident scene is subjected to three-dimensional scene reconstruction to obtain a dynamic scene model; The dynamic scene model is subjected to interactive rendering to obtain an accident space-time panorama.

[0041] The key frame field quantity subset refers to a field quantity data set containing key physical states of accidents, which is filtered out from continuous accident space-time field quantities by spatio-temporal slicing technology. The composite field quantity refers to a comprehensive physical field quantity data obtained by fusing multiple types of physical field quantities (such as velocity field, stress field and acceleration field, etc.) in the key frame field quantity subset by tensor superposition, weighted combination and other technologies. The dynamic scene model refers to a three-dimensional digital model of an accident scene with time sequence variation characteristics.

[0042] In the implementation process, continuous accident space-time field quantities can be divided into multiple time segments according to time sequence, a clustering algorithm (such as DBSCAN) is used to identify representative key frames in each segment, and a subset of field quantities containing key information such as speed mutation and stress peak is screened out. For example, in a two-vehicle collision accident, the speed field and stress field data of key time points such as deceleration before collision, collision moment and sliding after collision are extracted through time-space slicing to quickly lock the core physical state of accident evolution; the tensor superposition technology is used to align the dimensions and weight-merge the speed field vector (direction and size), stress field scalar (stress intensity) and other physical fields (such as acceleration field) in the key frame, to generate a composite field quantity containing multiple physical attributes, for example, in a curve rollover accident, the lateral speed vector of the vehicle at the rollover moment, the friction stress value between the tire and the ground, and the acceleration data of the body inclination are fused to form a composite data volume that completely describes the accident mechanics process; the three-dimensional modeling engine (such as Unity or Unreal Engine) is used to map the physical information in the composite field quantity to scene parameters, the spatial data such as vehicle position and road environment are converted into three-dimensional models through point cloud reconstruction technology, and the models are given dynamic change attributes according to time sequence, for example, in a rear-end collision accident, three-dimensional models before and after the collision of the two vehicles are constructed based on the composite field quantity data, accurately restoring details such as vehicle deformation and fragment scattering trajectory; the ray tracing rendering technology (such as NVIDIA RTX) is used to simulate the real environment of the accident site in combination with HDR lighting, and an interactive function (such as time axis dragging and view switching) is added to generate an accident space-time panorama that can be viewed in all directions.

[0043] The accident logic analysis module 203 is configured to construct an accident correlation graph of the simple traffic accident scene by using the accident semantics, and analyze an accident logic chain of the simple traffic accident scene by using the accident correlation graph.

[0044] By constructing the accident correlation graph of the simple traffic accident scene by using the accident semantics, the semantic information such as behavior elements, cause-effect relationships and legal basis extracted from the text can be converted into structured nodes and edges, the logical correlation and visual expression of the accident-related information are realized, and the internal relationship between elements in the accident is intuitively presented.

[0045] The accident correlation graph refers to a structured network graph constructed by abstracting the behavior elements (such as “changing lanes” and “exceeding speed”) and cause-effect relationships (such as “collision caused by not decelerating”) extracted from the accident description text as nodes.

[0046] As an embodiment of the present application, constructing the accident correlation graph of the simple traffic accident scene by using the accident semantics comprises: Element structure processing is performed on the accident semantics to obtain structured semantics; An accident relationship chain of the simple traffic accident scene is constructed using the structured semantics; Dynamic event marking is performed on the accident relationship chain to obtain a marked accident chain; The marked accident chain is visually encapsulated to obtain an accident correlation graph.

[0047] The structured semantics refer to standardized and computable semantic data obtained by organizing scattered text information in accident semantics according to a fixed data structure (such as a triple form), and the accident relationship chain refers to a chain clearly showing the sequence and cause-effect correlation of events in an accident.

[0048] In the specific implementation process, named entity recognition (NER) combined with dependency syntax analysis technology is used to extract elements such as behavior subjects (such as "car A" and "pedestrian"), actions (such as "speeding" and "courtesy"), times (such as "18:00"), and cause-effect relationship words (such as "cause" and "because") from accident description texts, to construct a triple structure (subject-predicate-object), for example, in the case of "car A running a red light causes car B to avoid not in time", the entities "car A" and "car B", the actions "running a red light" and "avoiding", and the cause-effect relationship "cause" are identified by the BERT model to generate structured triples: [car A - running a red light - empty], [running a red light - causes - car B to avoid]; based on the structured triples, a graph database (such as Neo4j) is used to establish logical relationships between entities, and nodes are connected through relationship types such as "trigger", "influence", and "cause" to form a time-sequenced event chain, for example, the "car A" node is connected to the "running a red light" event node through the "execute" edge, and the "running a red light" node is connected to the "car B avoiding" node through the "cause" edge, to construct a relationship chain of "violation behavior -> consequence"; in combination with key frame data (such as sudden change of speed at the collision moment) in the accident space-time field, physical properties (such as speed value and stress value) and time stamps are marked for event nodes in the relationship chain, to form dynamic marking, for example, the collision node is marked with "speed 15 m / s at t=3s" and "stress peak value 120 MPa", and the "car B avoiding" node is marked with "start decelerating at t=2s"; a visualization engine is used to render the marked accident chain into an interactive graph, to generate a scalable accident graph through node size (indicating severity), edge thickness (indicating cause-effect strength), and color mapping (distinguishing behavior types).

[0049] Further, the accident correlation graph is used to analyze the accident logic chain of the simple traffic accident scene, which can integrate scattered accident information into a clear and coherent cause-effect chain by combing the correlation between element nodes in the graph, to intuitively present the internal logic and evolution process of accident occurrence and development.

[0050] As an embodiment of the present application, the accident correlation graph is used to analyze the accident logic chain of the simple traffic accident scene, including: The accident correlation graph is used to identify the starting event set of the simple traffic accident scene; The starting event set is used to perform causal path tracking on the simple traffic accident scene to obtain a complete path set; The complete path set is subjected to key path screening to obtain a key causal path; The key causal path is subjected to attribution analysis to obtain a responsibility distribution path; The responsibility distribution path is subjected to natural language conversion to obtain an accident logic chain.

[0051] The starting event set refers to the beginning of the entire accident development process, the complete path set refers to all possible causal chains in the accident development process, for example, in a chain rear-end collision accident, all different combinations of event development paths from the initial behavior of different starting vehicles to the final multi-vehicle collision result jointly form the complete path set, the key causal path refers to the path that best reflects the core causal logic of the accident and plays a major role in the accident, and the responsibility distribution path refers to the path formed after matching the event nodes in the key causal path with a traffic regulation knowledge base and calculating and distributing the corresponding responsibility proportion according to the violation behavior and severity of each behavior subject by using a rule engine.

[0052] In the implementation process, the graph traversal algorithm (such as breadth-first search BFS) can be used to find the nodes with no incoming edges or only containing "time starting point" type incoming edges in the accident correlation graph, and define them as starting events, for example, in the rear-end collision graph, identify the nodes of "vehicle A not keeping a safe distance" and "vehicle B suddenly braking" as starting events, which usually correspond to the initial conditions or triggering behaviors of the accident; starting from the starting event node, perform a depth-first search along the "cause" and "impact" directed edges in the graph, record all reachable event node sequences, generate complete causal paths, for example, starting from "vehicle B suddenly braking", trace to "vehicle A braking not enough" → "two vehicles collide" → "vehicle damaged" and other nodes, form a complete accident development chain; match the event nodes in the critical path with the traffic regulation knowledge base, calculate the violation degree of each behavior subject (such as the weight of running a red light is 0.8) through the rule engine, and allocate the responsibility proportion according to the contribution degree, for example, in the path of "vehicle A running a red light (0.8) → vehicle B not avoiding enough (0.2)", automatically generate the responsibility proportion: vehicle A 80%, vehicle B 20%; using template generation technology, map the nodes and relationships in the responsibility path to natural language templates, fill in the specific event description, and generate coherent accident logic text, such as converting the responsibility path to "due to vehicle A running a red light (primary responsibility), leading to vehicle B not avoiding enough to cause a collision (secondary responsibility)", and automatically associate Article XX of the Road Traffic Safety Law.

[0053] Preferably, the complete path set is screened for a critical path to obtain a critical causal path, comprising: The node importance score of the path node in the complete path set is calculated using the following formula: ; Wherein, represents the node importance score of node i, represents the total number of nodes in the complete path set, and d represents the damping coefficient, represents the jth node in the path node, represents the set of nodes pointing to node i in the path node, represents the importance weight of node j, represents the outlink frequency of node j; Based on the node importance score, the complete path set is screened for a critical path to obtain a critical causal path.

[0054] In the implementation process, a node importance score threshold value can be set, nodes with scores higher than the threshold value in the complete path set and their connection edges are retained, nodes with low scores and redundant paths are removed, and then the remaining nodes are connected into a coherent path through path merging and optimization to determine the key causal path. If the threshold value is set to 0.6, the secondary nodes with scores lower than the threshold value such as “the normal driving of the B car” can be removed, and important nodes such as “the overspeed driving of the A car”, “the failure of the A car to observe the rearview mirror” and “the occurrence of scratching” are retained. These nodes are connected by using a graph algorithm to obtain the key causal path “the overspeed driving of the A car and the failure of the A car to observe the rearview mirror lead to scratching with the B car”.

[0055] It should be further explained that the principle of the node importance score calculation formula is as follows: (1) Importance propagation mechanism: the core idea of the formula is similar to “importance transfer”. The importance score of node i is determined by the importance weight of all nodes j pointing to it. The higher the weight of node j and the more direct the relationship pointing to node i, the greater the contribution of node j to the importance of node i. For example, in a causal path, if the weight of “user clicks on an advertisement” (node j) is high and it directly points to “purchase behavior” (node i), the score of node i will be improved due to the contribution of node j.

[0056] (2) Normalization effect of out-link times: the “out-link times ” in the denominator is used to balance the importance allocation of node j. If node j points to multiple nodes (with many out-link times), the importance transferred by each pointing relationship will be diluted. For example, node j points to i and k at the same time, and its weight will be evenly distributed to the two nodes, avoiding the virtual increase of the importance of a single node due to being pointed by multiple relationships.

[0057] (3) Damping coefficient and global balance: the damping coefficient d (usually set to about 0.85) is used to simulate the “random jump” behavior to avoid excessive concentration of importance in local nodes. The “ ” term in the formula represents the basis importance of global uniform distribution, ensuring that all nodes at least obtain a certain score to prevent isolated nodes from being ignored.

[0058] The formula can quantify the “hub role” of a node in a causal path. For example, in a user behavior analysis scenario, if the “add to cart” node is pointed by multiple high-weight pre-action behaviors (such as “browse detail page” and “get discount coupon”) and the out-link of these pre-action behaviors is less, the score of “add to cart” will be higher, and it will be determined as a key conversion node, and then redundant paths are filtered.

[0059] The accident responsibility analysis module 204 is configured to perform accident responsibility analysis on the simple traffic accident site based on the accident space-time panoramic graph and the accident logic chain, and obtain an accident analysis text.

[0060] The embodiment of the present application can combine the visual accident physical process and the structured causal logic to quickly and accurately determine the responsibility proportion of each participant in the accident and output the accident analysis text by analyzing the accident responsibility of the simple traffic accident scene based on the accident space-time panoramic diagram and the accident logic chain, thereby providing a scientific and intuitive basis for responsibility determination.

[0061] The accident analysis text refers to a written document finally generated by comprehensively analyzing the accident scene physical process (such as vehicle motion trajectory, collision stress distribution) and the causal relationship (such as the responsibility path of the violation behavior leading to the accident), and containing the basic information of the accident, the description of the occurrence process, the responsibility determination result and the basis.

[0062] As an embodiment of the present application, the accident responsibility of the simple traffic accident scene is analyzed based on the accident space-time panoramic diagram and the accident logic chain to obtain the accident analysis text, which comprises: The feature fusion is performed on the accident space-time panoramic diagram and the accident logic chain to obtain a joint feature vector; The joint feature vector is subjected to multi-factor weighting to obtain a responsibility intensity vector; The responsibility distribution diagram of the simple traffic accident scene is constructed by using the responsibility intensity vector; The responsibility distribution diagram is subjected to legal annotation to obtain an annotated responsibility distribution diagram; The accident analysis text of the simple traffic accident scene is constructed by using the annotated responsibility distribution diagram.

[0063] The joint feature vector refers to a multi-dimensional data structure presented in the form of a vector after aligning and integrating the physical features (such as vehicle motion trajectory, collision speed, stress distribution value) contained in the accident space-time panoramic diagram and the semantic features (such as violation behavior, causal relationship, regulation association information) contained in the accident logic chain in time or event occurrence order. The responsibility intensity vector refers to a one-dimensional numerical vector obtained by weighting and summing the joint feature vector by using techniques such as the analytic hierarchy process according to the influence degree of each feature on the accident responsibility determination. The responsibility distribution diagram refers to a visualized graph formed by mapping the responsibility intensity numerical value of each participant in the accident onto the spatial layout of the accident scene through visual elements such as color depth, graphic area and icon size by using the GIS geographic information system or the Echarts visualization tool based on the responsibility intensity vector.

[0064] In the implementation process, physical features (such as vehicle trajectory coordinates, collision speed, stress distribution value) can be extracted from the accident space-time panoramic map, and semantic features (such as illegal behavior, causal relationship weight) can be extracted from the accident logic chain. After aligning the two types of features in time sequence, they are combined into joint features in matrix form. For example, in a two-car scratch accident, the speed curve and displacement coordinates of the car A when changing lanes in the panoramic map are integrated with semantic features such as "illegal lane changing" and "not observing the following car" in the logic chain into a two-dimensional matrix containing space-time data and behavior description. Based on the responsibility intensity vector, the responsibility values of the accident participants are mapped into spatial visualization graphics using visualization drawing tools (such as Echarts), and the responsibility proportion is distinguished by color depth, graphic area and other visual elements, such as in a crossroad collision accident, the red area is used to mark the responsibility intensity of car A as 0.7 (with a larger area), and the blue area is used to mark the responsibility intensity of car B as 0.3 (with a smaller area), to intuitively present the responsibility distribution. The responsibility elements in the responsibility distribution map are matched with the traffic regulation knowledge base, and the relevant legal provisions, penalty basis and responsibility identification standards are marked at the corresponding positions in the map, and the marked map with text and graphics is formed. In the area where the car A caused the collision by speeding, the provisions of Article 42 of the Road Traffic Safety Law about speed limit and the identification of "speeding behavior bearing main responsibility" are marked, forming a marked map with text and graphics. Through natural language generation (NLG) technology, the visualization information, responsibility value and legal basis in the marked responsibility distribution map are converted into structured text, such as "On X month X day, 2025, car A drove at a speed of 20% on XX road (in violation of Article 42), and collided with car B. After analysis, car A bears 70% of the main responsibility, and car B bears 30% of the secondary responsibility".

[0065] Preferably, the joint feature vector is subjected to multi-factor weighting to obtain a responsibility intensity vector, including: querying the behavior vector in the joint feature vector; calculating the behavior responsibility score of the behavior vector using the following formula: ; wherein, represents the behavior responsibility score, represents the number of behavior vectors, represents the responsibility weight of the behavior vector k, represents the behavior credibility of the behavior vector k, represents the causal correlation degree of the behavior vector k; based on the behavior responsibility score, the joint feature vector is subjected to multi-factor weighting to obtain a responsibility intensity vector.

[0066] wherein, the behavior responsibility score refers to a quantitative index for comprehensively measuring the influence degree of each behavior in the accident on responsibility identification.

[0067] In a specific implementation, the feature vectors of behaviors related to the accident are screened out from the joint feature vectors, such as the vectors corresponding to behaviors such as "excessive speed", "running a red light", "not maintaining a safe distance", and the like; all features in the joint feature vectors are weighted according to the responsibility scores of the behaviors, and the behavior factors and other factors (such as vehicle trajectory and collision strength) are comprehensively calculated to finally generate a one-dimensional responsibility strength vector, quantifying the accident responsibility degree of each participant, such as after calculating the responsibility score of the "excessive speed" and "not maintaining a safe distance" behaviors of the B vehicle, combining other features such as the vehicle collision instantaneous speed, stress field data, and the like, and through a weighted average algorithm, a responsibility strength vector of the B vehicle responsibility strength of 0.8 and the A vehicle responsibility strength of 0.2 is obtained, providing a quantitative basis for responsibility determination.

[0068] It should be further pointed out that the principle of the behavior responsibility score calculation formula is as follows: (1) Multi-factor weighted fusion: the formula integrates the "responsibility weight", "behavior credibility" and "causal correlation degree" in the form of product. The responsibility weight reflects the severity of the violation, the credibility ensures the reliability of the data, and the causal correlation quantifies the direct impact of the behavior on the result. The multiplication of the three can comprehensively reflect the actual contribution of the behavior in the responsibility determination.

[0069] (2) Logic of weight distribution: the responsibility weight is a core parameter set a priori (such as the regulation stipulates that the weight of "running a red light" is higher than that of "not turning on the turn signal"), and the credibility and causal correlation degree are dynamically assigned based on evidence and logical analysis. The combination of the two can achieve adjustment according to the specific scene, achieving the dual constraint effect of "rules + evidence".

[0070] The simple accident report module 205 is configured to query historical accident judgment data corresponding to the simple traffic accident scene, optimize the accident judgment of the accident responsibility text based on the historical accident judgment data, obtain an optimized accident text, calibrate an intelligent conclusion of the optimized accident text, obtain a target accident text, and construct an accident responsibility report of the simple traffic accident scene by using the target accident text.

[0071] By querying the historical accident judgment data corresponding to the simple traffic accident scene, the embodiments of the present application can provide valuable references for the responsibility determination, processing flow and prevention strategies of the current accident by means of the similar experience of past accidents in terms of causes, responsibility determination and processing methods, thereby helping to improve the efficiency and accuracy of accident handling.

[0072] Optionally, the historical accident judgment data can be obtained from the database of the traffic safety comprehensive service management platform.

[0073] Further, the embodiment of the present application optimizes the accident judgment of the accident responsibility text based on the historical accident judgment data, and obtains an optimized accident text. The accident responsibility text can be optimized based on the historical accident judgment data. By comparing the responsibility determination logic, similar case judgment standards and processing results of historical cases, deviations or omissions in the current accident responsibility text are corrected, the accuracy and rationality of the responsibility determination are improved, and a more scientific and rigorous optimized accident text is generated.

[0074] The optimized accident text refers to a text formed by correcting and improving the original accident responsibility text in terms of responsibility determination logic and responsibility proportion allocation through multidimensional comparison and analysis with historical accident judgment data.

[0075] As an embodiment of the present application, the accident judgment of the accident responsibility text is optimized based on the historical accident judgment data, and an optimized accident text is obtained, comprising: aligning features of the historical accident judgment data and the accident responsibility text to obtain a comparable feature set; detecting responsibility deviation of the comparable feature set to obtain a responsibility deviation report; correcting responsibility weight of the accident responsibility text based on the responsibility deviation report to obtain an optimized accident text.

[0076] The comparable feature set refers to a structured data set obtained by aligning features of historical accident judgment data and accident responsibility text. The responsibility deviation report refers to a file generated by analyzing the comparable feature set, which is used to reveal the differences between the current accident responsibility text and the historical accident judgment data in terms of responsibility determination.

[0077] In the implementation process, first, the named entity recognition (NER) technology is used to extract key features such as accident type, responsible party, violation behavior, and casualty from historical accident judgment data and accident liability text. Then, the dynamic time warping (DTW) algorithm is used to align feature sequences of different text lengths in time or logical order to form a unified structure of comparable feature sets. For example, in a two-car scratch accident, the features such as "turning at intersection and scratching" and "full responsibility of car A" are extracted from historical cases, and the features such as "scratching caused by overtaking on a curve" and "main responsibility of car B" are extracted from the current accident liability text. Through the DTW algorithm, the dimensions such as accident type and responsibility division are aligned to generate a structured feature comparison table. The cosine similarity algorithm is used to calculate the difference between the current accident and the historical case in each feature dimension, and a threshold is set to determine whether there is a deviation in the responsibility determination. Then, combined with the association rule mining technology, the responsibility deviation report is generated by analyzing the responsibility allocation patterns corresponding to similar feature combinations in historical data and comparing them with the current text. If, in the historical case of "collision caused by running a red light at an intersection", the vehicle running a red light is usually fully responsible, but in the current accident liability text, the vehicle running a red light (car C) is only responsible for 60%, and the other vehicle (car D) that is driving normally but speeding is responsible for 40%, the cosine similarity is calculated for the features such as accident type, violation behavior, and responsibility division, and it is found that the difference between the current responsibility determination and the historical case exceeds the preset threshold. The responsibility deviation report points out that the current accident responsibility determination is inconsistent with the full responsibility of the vehicle running a red light in the historical mode, prompting that the responsibility division needs to be reviewed again. According to the deviation prompt in the responsibility deviation report, referring to the responsibility weight distribution of similar features in historical cases, the weighted average method is used to recalculate the weight of each responsible party in the current accident. Then, the corrected responsibility information is automatically filled into the accident liability text framework to generate an optimized accident text. For example, historical data shows that in the "collision caused by not showing courtesy to pedestrians at an intersection" scenario, the motor vehicle usually needs to bear more than 80% responsibility. In a current accident, the original responsibility determination text determines that the pedestrian is 60% responsible for running a red light, and the driver of car A is 40% responsible for not reducing speed. After comparing historical data, it is found that the current determination is significantly different from past cases. Referring to historical cases, the responsibility weight of car A can be increased from 40% to 85%, and the responsibility weight of the pedestrian can be decreased from 60% to 15%. The optimized responsibility determination text is as follows: After investigation, car A did not reduce speed and did not show courtesy to the pedestrian crossing the intersection, violating the relevant provisions of the "Road Traffic Safety Law"; although the pedestrian ran a red light, car A did not take effective evasive measures, which was the main cause of the accident. According to the historical responsibility determination experience of similar accidents, it is determined that car A bears 85% of the main responsibility and the pedestrian bears 15% of the secondary responsibility.

[0078] The embodiment of the present invention performs intelligent conclusion calibration on the optimized accident text to obtain a target accident text, which can detect and correct responsibility identification contradictions, logical loopholes and expression ambiguities in the text, and output a target accident text with more rigorous conclusions and more standardized expressions.

[0079] Among them, the target accident text refers to the final document of simple traffic accident responsibility identification formed after multiple rounds of processing and optimization.

[0080] Optionally, the optimized accident text is subjected to intelligent conclusion calibration, and the target accident text obtained can be verified by manual review.

[0081] Furthermore, the embodiment of the present invention utilizes the target accident text to construct an accident responsibility report for the simple traffic accident scene, which can convert the accurate and rigorous accident responsibility information after intelligent calibration into a standardized, complete and legally effective written document, providing a standardized basis for accident handling, responsibility determination and subsequent tracing.

[0082] During the specific implementation process, the precise spatiotemporal information, physical process description, responsibility determination results and legal basis in the target accident text can be used as the core content. According to the standardized report template, the basic information of the accident, the analysis process and the handling conclusion can be structured and integrated to generate a simple traffic accident scene accident responsibility report with standardized format and detailed content. For example, on October 15, 2024, two vehicles rear-ended each other on XX Avenue in XX City. After investigation, it was found that vehicle B was speeding and did not maintain a safe distance. It hit vehicle A, which was decelerating, at 65km / h. The responsibility intensity was calculated to account for 80%. Vehicle B was determined to be primarily responsible and must bear compensation and accept punishment.

[0083] like Figure 2 FIG. 1 is a flow chart of a simplified intelligent traffic accident responsibility determination method based on an AI large model according to an embodiment of the present invention. In this embodiment, the simplified intelligent traffic accident responsibility determination method based on an AI large model includes: Collect accident scene images and accident description text of a simple traffic accident scene, use the trained AI large model to extract features from the accident scene images and the accident description text, and obtain key accident parameters and accident semantics; Analyzing the accident space-time field quantity of the simple traffic accident scene by using the accident key parameters, and constructing a space-time panoramic view of the accident scene by using the accident space-time field quantity; Using the accident semantics, construct an accident association graph of the simple traffic accident scene, and using the accident association graph, analyze the accident logic chain of the simple traffic accident scene; Based on the accident spatiotemporal panorama and the accident logic chain, an accident responsibility analysis is performed on the simple traffic accident scene to obtain an accident analysis text; Inquire the historical accident judgment data corresponding to the simple traffic accident scene, optimize the accident judgment of the accident responsibility text based on the historical accident judgment data, obtain the optimized accident text, calibrate the intelligent conclusion of the optimized accident text, obtain the target accident text, and use the target accident text to construct the accident responsibility report of the simple traffic accident scene.

[0084] In the several embodiments of the present application, it should be understood that the system and method provided can be implemented in other manners. For example, the division of the system embodiments is merely an example, and the division of the modules can be different, for example, the division of the modules can be one division, two divisions, three divisions or four divisions, or the division of the modules can be one division, two divisions, three divisions or four divisions.

[0085] In addition, each function module in each embodiment of the present application can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware, or in the form of a combination of hardware and software function modules.

[0086] Finally, it should be noted that the above embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A simple intelligent traffic accident responsibility identification system based on AI big model, characterized by: The system includes: a feature extraction module, a scene panorama construction module, an accident logic analysis module, a fault responsibility analysis module and a simple accident reporting module; The feature extraction module is used to collect accident scene images and accident description text of a simple traffic accident scene, and use the trained AI large model to extract features from the accident scene images and the accident description text to obtain key accident parameters and accident semantics; The on-site panoramic map construction module is used to analyze the accident time-space field quantity of the simple traffic accident scene using the accident key parameters, and construct the accident time-space panoramic map of the simple traffic accident scene using the accident time-space field quantity; The accident logic analysis module is used to construct an accident association map of the simple traffic accident scene using the accident semantics, and analyze the accident logic chain of the simple traffic accident scene using the accident association map; The accident responsibility analysis module is used to perform accident responsibility analysis on the simple traffic accident scene based on the accident spatiotemporal panorama and the accident logic chain to obtain an accident analysis text; The simple accident report module is used to query the historical accident judgment data corresponding to the simple traffic accident scene, optimize the accident judgment of the accident responsibility text based on the historical accident judgment data, obtain the optimized accident text, perform intelligent conclusion calibration on the optimized accident text, obtain the target accident text, and use the target accident text to construct the accident responsibility report of the simple traffic accident scene.

2. The AI ​​large model-based simple traffic accident responsibility intelligent identification system according to claim 1, wherein the use of the key accident parameters to analyze the accident time and space field quantities at the simple traffic accident scene includes: Performing time-space coordinate system normalization on the key accident parameters to obtain standardized parameters; Using the standardized parameters, a kinematic field model is performed on the simple traffic accident scene to obtain a velocity field; Calculating the stress field at the simple traffic accident scene based on the velocity field; The velocity field and the stress field are fused to obtain the accident space-time field quantity.

3. The simplified intelligent traffic accident responsibility determination system based on AI large model as claimed in claim 1 is characterized in that: Using the accident space-time field quantity, a space-time panoramic view of the accident scene is constructed, including: Performing spatiotemporal slicing on the accident spatiotemporal field quantity to obtain a key frame field quantum set; Based on the key frame field quantum set, multi-physics field fusion is performed on the simple traffic accident scene to obtain composite field quantities; Using the composite field quantity, a three-dimensional scene reconstruction is performed on the simple traffic accident scene to obtain a dynamic scene model; The dynamic scene model is interactively rendered to obtain a spatiotemporal panoramic view of the accident.

4. The simplified intelligent traffic accident responsibility determination system based on AI large model as claimed in claim 1 is characterized in that: Using the accident semantics, an accident association graph of the simple traffic accident scene is constructed, including: Performing element-structuring processing on the accident semantics to obtain structured semantics; Using the structured semantics, constructing an accident relationship chain of the simple traffic accident scene; Performing dynamic event marking on the accident relationship chain to obtain a marked accident chain; The marked accident chain is visually encapsulated to obtain an accident correlation graph.

5. The simplified intelligent traffic accident responsibility determination system based on AI large model as claimed in claim 1 is characterized in that: Utilizing the accident association graph, analyzing the accident logic chain of the simple traffic accident scene includes: Using the accident association graph, identifying a set of starting events at the scene of the simple traffic accident; Using the initial event set, causal path tracing is performed on the simple traffic accident scene to obtain a complete path set; Performing critical path screening on the complete path set to obtain a critical causal path; Perform attribution analysis on the key causal path to obtain the responsibility distribution path; The responsibility distribution path is converted into natural language to obtain the accident logic chain.

6. The simplified intelligent traffic accident responsibility determination system based on the AI ​​large model as claimed in claim 5 is characterized in that: The critical path screening of the complete path set to obtain the critical causal path includes: The node importance score of the path nodes in the complete path set is calculated using the following formula: ； in, represents the node importance score of node i, represents the total number of nodes in the complete path set, d represents the damping coefficient, represents the jth node in the path node, Represents the set of path nodes pointing to node i, represents the importance weight of node j, represents the number of outlinks of node j; Based on the node importance scores, critical paths are screened for the complete path set to obtain critical causal paths.

7. The simplified intelligent traffic accident responsibility determination system based on AI large model as claimed in claim 1 is characterized in that: Based on the accident spatiotemporal panorama and the accident logic chain, an accident responsibility analysis is performed on the simple traffic accident scene to obtain an accident analysis text, including: Performing feature fusion on the accident spatiotemporal panorama and the accident logic chain to obtain a joint feature vector; Performing multi-factor weighting on the joint feature vector to obtain a responsibility intensity vector; Using the responsibility intensity vector, constructing a responsibility distribution map of the simple traffic accident scene; Performing legal annotation on the responsibility distribution map to obtain an annotated responsibility distribution map; The annotated responsibility distribution map is used to construct an accident analysis text of the simple traffic accident scene.

8. The simplified intelligent traffic accident responsibility determination system based on the AI ​​large model as claimed in claim 7 is characterized in that: The multi-factor weighting of the joint feature vector to obtain the responsibility intensity vector includes: querying the behavior vector in the joint feature vector; The behavior responsibility score of the behavior vector is calculated using the following formula: ； in, represents the behavioral responsibility score, represents the number of behavior vectors, represents the responsibility weight of the behavior vector k, represents the behavior credibility of behavior vector k, Represents the causal correlation degree of behavior vector k; Based on the behavior responsibility score, the joint feature vector is weighted by multiple factors to obtain a responsibility intensity vector; Among them, the behavioral responsibility score refers to a quantitative indicator that comprehensively measures the impact of each behavior in the accident on the determination of responsibility.

9. The simplified intelligent traffic accident responsibility determination system based on AI large model as claimed in claim 1 is characterized in that: Based on the historical accident judgment data, the accident responsibility text is optimized to obtain an optimized accident text, including: Aligning features of the historical accident judgment data and the accident responsibility text to obtain a comparable feature set; Performing responsibility deviation detection on the comparable feature set to obtain a responsibility deviation report; Based on the responsibility deviation report, the responsibility weight of the accident responsibility text is corrected to obtain an optimized accident text.

10. A simple intelligent traffic accident responsibility determination method based on AI big model is characterized by: The method comprises: Collect accident scene images and accident description text of a simple traffic accident scene, use the trained AI large model to extract features from the accident scene images and the accident description text, and obtain key accident parameters and accident semantics; Analyzing the accident space-time field quantity of the simple traffic accident scene by using the accident key parameters, and constructing a space-time panoramic view of the accident scene by using the accident space-time field quantity; Using the accident semantics, construct an accident association graph of the simple traffic accident scene, and using the accident association graph, analyze the accident logic chain of the simple traffic accident scene; Based on the accident spatiotemporal panorama and the accident logic chain, an accident responsibility analysis is performed on the simple traffic accident scene to obtain an accident analysis text; Query the historical accident judgment data corresponding to the simple traffic accident scene, optimize the accident judgment of the accident responsibility text based on the historical accident judgment data to obtain an optimized accident text, perform intelligent conclusion calibration on the optimized accident text to obtain a target accident text, and use the target accident text to construct an accident responsibility report for the simple traffic accident scene.

Images