Attribution method, apparatus, device, and storage medium

Abstract

The application provides a responsibility determination method, device, equipment and storage medium, relates to the technical field of vehicles, can improve the responsibility determination accuracy, and realizes the quantitative output of the responsibility proportion. The method comprises the following steps: obtaining accident associated data of an accident vehicle, wherein the accident associated data is used for indicating data associated with an accident from a plurality of data sources; performing feature extraction on the accident associated data to obtain accident associated features of the accident vehicle; constructing an association graph based on the accident associated features and the accident associated data; and calculating the responsibility proportion of the accident vehicle based on the association graph, wherein the responsibility proportion is used for indicating the responsibility proportion of each party involved in causing the accident of the accident vehicle.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and more particularly to a method, apparatus, device, and storage medium for determining liability. Background Technology

[0002] Determining liability in an accident is the core of car insurance claims and traffic accident handling. Its accuracy and efficiency directly affect the timeliness of claims processing, the rate of disputes with car owners, and the fairness of traffic law enforcement.

[0003] The relevant technologies mainly rely on the event video of the Advanced Driving Assistance System (ADAS) of the accident vehicle or the Controller Area Network (CAN) bus data to determine the responsibility for the accident.

[0004] However, the liability determination in the above-mentioned technical solutions is easily affected by the bias of a single data source, resulting in a low accuracy rate in liability determination. In addition, the above-mentioned technical solutions can only achieve qualitative determination of full responsibility, primary and secondary responsibility, or no responsibility, and cannot quantify the proportion of responsibility of the accident vehicle. Summary of the Invention

[0005] The purpose of this application is to provide a method, apparatus, device, and storage medium for determining liability, which can improve the accuracy of liability determination and achieve quantitative output of liability proportion.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application provides a method for determining liability, the method comprising: Obtain accident-related data for the accident vehicles. This data indicates data from multiple sources associated with the accident. Extract features from the accident-related data to obtain the accident-related features of the accident vehicles. Construct a correlation graph based on these features and the accident-related data. Calculate the responsibility percentage of the accident vehicles based on the correlation graph. This responsibility percentage indicates the proportion of responsibility of each party involved in the accident.

[0007] The technical solution provided in this application obtains accident-related data from multiple data sources, extracts features from the accident-related data to obtain accident-related features, constructs a correlation graph based on the accident-related features and accident-related data, and then calculates the responsibility ratio of the accident vehicle based on the correlation graph. By determining responsibility based on multi-source data, the determination of responsibility can avoid the deviation caused by data bias of single-source data, thus improving the accuracy of responsibility determination. In addition, the calculation of the responsibility ratio of the accident vehicle based on the accident-related features and the correlation graph can output a quantitative responsibility ratio of 0-100%.

[0008] One possible implementation, based on a correlation graph, calculates the responsibility percentage of the accident vehicle. Specifically, this can be achieved by calculating the correlation strength between nodes in the correlation graph based on their feature values. Then, based on this correlation strength, the responsibility percentage of the accident vehicle is calculated.

[0009] Another possible implementation is to calculate the association strength between nodes in the association graph based on the feature values ​​of each node. Specifically, this can be achieved by using a cosine similarity algorithm to calculate the association strength between nodes in the association graph based on the feature values ​​of each node.

[0010] Another possible implementation is to calculate the responsibility ratio of the accident vehicle based on the correlation strength between nodes in the correlation graph. Specifically, this can be achieved by calculating the responsibility ratio of the accident vehicle based on the correlation strength between nodes in the correlation graph using the first formula.

[0011] Another possible implementation, the first formula can be expressed as:

[0012] in, For the percentage of responsibility of the j-th party involved, Let be the association strength between the j-th involved party and the i-th type of node in the association graph. Let be the weight value of the i-th type of node.

[0013] Another possible implementation involves constructing a correlation graph based on accident correlation features and accident correlation data. Specifically, this can be achieved by: constructing nodes in the correlation graph based on the accident correlation data; and assigning accident correlation features to the nodes to obtain the correlation graph.

[0014] Another possible implementation is that the nodes in the association graph include: involved party nodes, behavior nodes, environment nodes, and device nodes.

[0015] Another possible implementation, the liability determination method provided in this application may further include: generating accident scene reconstruction information based on accident-related data, wherein the accident scene reconstruction information includes at least one of spatiotemporal reconstruction information, process reconstruction information, or state reconstruction information.

[0016] Another possible implementation, the liability determination method provided in this application, may further include: obtaining historical liability determination data; updating the weight values ​​of nodes in the association graph based on the historical liability determination data; and / or updating the node categories in the association graph.

[0017] Another possible implementation method provided in this application may include: data processing of accident-related data, including outlier removal, missing value filling, or standardization processing, at least one of the following.

[0018] Another possible implementation, the liability determination method provided in this application may also include: time alignment processing and / or spatial location alignment processing of accident-related data.

[0019] Secondly, a liability determination device is provided, the device comprising: The acquisition module is used to acquire accident-related data of the accident vehicle. The accident-related data is used to indicate data related to the accident from multiple data sources.

[0020] The processing module is used to extract features from accident-related data to obtain the accident-related features of the accident vehicles.

[0021] The module is used to build a correlation graph based on accident correlation features and accident correlation data.

[0022] The calculation module is used to calculate the responsibility ratio of the accident vehicle based on the correlation graph. The responsibility ratio is used to indicate the responsibility ratio of each party involved in the accident.

[0023] One possible implementation is that the calculation module is also used to: calculate the association strength between nodes in the association graph based on the feature values ​​of each node in the association graph; and calculate the responsibility ratio of the accident vehicle based on the association strength between nodes in the association graph.

[0024] Another possible implementation is that the calculation module is also used to: calculate the association strength between nodes in the association graph based on the feature values ​​of each node in the association graph using the cosine similarity algorithm.

[0025] Another possible implementation is that the calculation module is also used to: calculate the responsibility ratio of the accident vehicle based on the correlation strength between nodes in the correlation graph and through the first formula.

[0026] Another possible implementation, the first formula can be expressed as:

[0027] in, For the percentage of responsibility of the j-th party involved, Let be the association strength between the j-th involved party and the i-th type of node in the association graph. Let be the weight value of the i-th type of node.

[0028] Another possible implementation involves building a module that is also used to construct nodes in a correlation graph based on accident correlation data. Accident correlation features are then assigned to the nodes to obtain the correlation graph.

[0029] Another possible implementation is that the nodes in the association graph include: involved party nodes, behavior nodes, environment nodes, and device nodes.

[0030] Another possible implementation is that the processing module is also used to: generate accident scene reconstruction information based on accident-related data, wherein the accident scene reconstruction information includes at least one of spatiotemporal reconstruction information, process reconstruction information, or state reconstruction information.

[0031] In another possible implementation, the liability determination device provided in this application further includes an update module, which is used to acquire historical liability determination data, update the weight values ​​of nodes in the association graph based on the historical liability determination data, and / or update the node categories in the association graph.

[0032] Another possible implementation is that the acquisition module is also used to: perform data processing on accident-related data, including outlier removal, missing value filling, or at least one of standardization processing.

[0033] Another possible implementation is that the acquisition module is also used to perform time alignment processing and / or spatial location alignment processing on accident-related data.

[0034] The technical effects of any implementation method in the second aspect can be found in the technical effects of any implementation method in the first aspect mentioned above, and will not be repeated here.

[0035] Thirdly, a computer device is provided, comprising: a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the above-mentioned method of assigning responsibility.

[0036] Fourthly, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to implement the above-mentioned method of assigning responsibility.

[0037] The solutions provided in the third and fourth aspects above are used to implement the method provided in the first aspect above, and their specific implementations will not be elaborated further. The technical effects corresponding to any implementation method in the solutions provided in the third and fourth aspects above can be found in the technical effects corresponding to any implementation method in the first aspect above, and will not be elaborated further here.

[0038] It should be noted that any of the possible implementations of any of the above aspects can be combined, provided that the solutions do not contradict each other. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A schematic diagram of the system architecture for a liability determination method provided in an embodiment of this application; Figure 2 A flowchart illustrating a method for determining liability provided in an embodiment of this application; Figure 3 A flowchart illustrating another method for determining liability provided in an embodiment of this application; Figure 4 A flowchart illustrating another method for determining liability provided in an embodiment of this application; Figure 5 A flowchart illustrating another method for determining liability provided in an embodiment of this application; Figure 6 A flowchart illustrating another method for determining liability provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a liability determination system provided in an embodiment of this application; Figure 8 This is a schematic diagram of the structure of a liability determination device provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0041] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0042] In the description of this application, it should be understood that the terms "upper," "lower," "left," "right," "front," "rear," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or relative positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and for simplification, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Unless otherwise specified, the above-mentioned orientational descriptions can be flexibly set in practical applications, provided that the relative positional relationships shown in the accompanying drawings are satisfied.

[0043] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0044] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "communication" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection. They can refer to a direct connection or an indirect connection through an intermediate medium, or a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0045] In embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, article, or apparatus that includes that element.

[0046] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0047] In the embodiments of this application, at least one can also be described as one or more, and multiple can be two, three, four or more, and this application does not impose any restrictions.

[0048] In the description of this specification, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0049] It should be noted that all information (including but not limited to device information, personal information of the subject), data (including but not limited to data used for analysis, stored data, and displayed data), and signals involved in this application have been authorized by the subject or fully authorized by all parties, and the collection, use, and processing of related data must comply with relevant laws, regulations, and standards. For example, accident-related data involved in this application were obtained with full authorization.

[0050] The industry-standard method for determining liability is the traditional method, which will be briefly explained below.

[0051] Traditional methods of determining liability primarily rely on subjective judgment made by personnel based on accident scene photos, statements from parties involved, and simple investigation records by traffic police. Alternatively, they may use basic algorithms to analyze limited data (such as speed from a single source or collision time) to arrive at a determination of full liability, primary or secondary liability, equal liability, or no liability. Furthermore, traditional methods largely depend on textual descriptions or accident scene photos to reconstruct the accident scenario.

[0052] However, the aforementioned traditional methods of determining liability mainly rely on manual processes, making automation impossible and resulting in lengthy assessment cycles. Furthermore, the results can only classify liability as full, primary or secondary, equal, or no liability, failing to quantify the proportion of responsibility in an accident. Additionally, these technical solutions fail to provide a visual representation of the accident scenario.

[0053] Among related technologies, automated liability determination can also be achieved through a single-source data-driven automatic liability determination scheme for auto insurance accidents.

[0054] Specifically, the system collects ADAS event videos (driving recorder footage) or CAN bus data (such as basic vehicle data like speed and braking status) from the accident vehicle. It then performs simple analysis on the data from these single data sources (such as video frame analysis and comparison of basic driving parameters). In conjunction with preset basic traffic rules (such as whether a red light was run or whether a line was crossed), it makes a preliminary determination of liability for the accident.

[0055] However, the above-mentioned technical solutions rely on data from only a single data source for liability determination, making the results susceptible to data bias from that single source and resulting in low accuracy. Furthermore, these solutions do not consider external factors such as roadside facilities and third-party liability, and therefore cannot cover special scenarios such as chain collisions and accidents caused by roadside obstacles, thus limiting their application scope.

[0056] Based on this, this application provides a method for determining liability. By acquiring accident-related data from multiple data sources, feature extraction is performed on the accident-related data to obtain accident-related features. A correlation graph is constructed based on the accident-related features and accident-related data. Then, based on the correlation graph, the liability ratio of the accident vehicle is calculated. Determining liability based on multi-source data can avoid the liability determination bias caused by data bias of single-source data, thus improving the accuracy of liability determination. In addition, the calculation of the liability ratio of the accident vehicle based on the correlation graph can output a quantitative liability ratio of 0-100%.

[0057] The solutions provided by the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0058] Figure 1 A schematic diagram of the system architecture for a liability determination method provided in an embodiment of this application, as shown below. Figure 1 As shown, the system includes: a computer device 101, which can be implemented as a terminal or a server.

[0059] When the computer device 101 is implemented as a server, the computer device 101 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0060] When the computer device 101 is implemented as a terminal, the computer device 101 can be a smartphone, tablet computer, laptop computer, desktop computer, etc.

[0061] Optionally, the system described above includes one or more computer devices 101. This application embodiment does not limit the number of computer devices 101.

[0062] Specifically, in Figure 1 In the illustrated system architecture, computer device 101 acquires accident-related data of the accident-involved vehicles, extracts features from the accident-related data to obtain accident-related features of the accident-involved vehicles, then constructs a correlation graph based on the accident-related data and accident-related features, and finally calculates the responsibility ratio of the accident-involved vehicles based on the correlation graph. The specific implementation of computer device 101 is described in the following method embodiment.

[0063] Figure 2 This is a flowchart illustrating a method for determining liability provided in an embodiment of this application. The method can be executed by a computer device. The computer device can be, for example,... Figure 1 The computer device 101 shown is shown.

[0064] like Figure 2 As shown, the liability determination method provided in this application embodiment may include: Step S202: The computer equipment acquires the accident-related data of the accident vehicle.

[0065] Among them, accident-related data is used to indicate accident-related data from multiple data sources.

[0066] Optionally, the accident-related data includes at least one of vehicle-side data, roadside data, and third-party data.

[0067] Vehicle-side data refers to data collected from the terminal of the vehicle involved in the accident.

[0068] Optionally, vehicle-side data includes, but is not limited to, vehicle trajectory data, collision sensor data, dashcam video data, and intelligent driving system operation data.

[0069] Vehicle trajectory data refers to the driving trajectory data of the accident vehicle. For example, vehicle trajectory data may include GPS latitude and longitude, driving speed, acceleration, steering angle, and lane position. When collecting vehicle trajectory data, the collection frequency can be set according to the actual situation; for example, the collection frequency can be set to 10Hz. It should be noted that the collection frequency setting must ensure that it can capture the details of vehicle movement.

[0070] Collision sensing data refers to data related to the collision details of vehicles involved in an accident. For example, collision sensing data may include collision time, collision intensity, collision location, and the degree of vehicle body deformation. Specifically, computer equipment can trigger the collection of collision sensing data in real time using onboard collision sensors.

[0071] Dashcam video data refers to video footage recorded by a dashcam installed on the vehicle involved in an accident. For example, dashcam video data can be video recorded by the front, side, and rear cameras of the vehicle involved in the accident, with a video resolution of no less than 1080P and a frame rate of 30fps. The dashcam video data can completely record the surrounding environment of the vehicle before and after the accident.

[0072] Intelligent driving system operational data refers to the operational status information of the intelligent driving system of an accident vehicle in an intelligent driving scenario. For example, intelligent driving system operational data may include driving mode (manual, L3, L4), sensor operating status, algorithm decision logs, emergency braking trigger records, and decision delay time. It is understood that intelligent driving system operational data is applicable to liability determination in intelligent driving scenarios.

[0073] In some embodiments, the computer equipment collects data from terminals such as the vehicle's onboard telematics box (T-Box), onboard sensors (such as speed sensors, acceleration sensors, steering sensors, etc.), front and rear side view dashcams, and intelligent driving system controllers to obtain vehicle-side data of the accident vehicle.

[0074] Vehicle-side data can reflect the vehicle's own motion status, collision details, environmental observation, and the operation of the intelligent driving system. Vehicle-side data is the core foundational data for determining liability.

[0075] Roadside data refers to vehicle behavior data collected by sensors (such as cameras and lidar) installed at locations such as beside roads or intersections. Roadside data can also be called roadside perception data, but this application does not limit the name of roadside data.

[0076] Optionally, roadside data may include roadside camera video data, radar detection data, and traffic light status data.

[0077] Roadside camera video data refers to the video data collected by roadside cameras. Roadside camera video data is usually a panoramic view, which can cover the accident section and surrounding lanes, and make up for the blind spots of the vehicle-mounted image.

[0078] Radar detection data refers to data detected by roadside lidar. For example, radar detection data can include information such as the distance between vehicles involved, relative speed, trajectory deviation, and the timing of multi-vehicle interactions. Radar detection data can accurately capture the spatial positional relationships of multiple vehicles.

[0079] Traffic light status data is used to describe the status information of traffic lights at the accident site. For example, traffic light status data may include the light color and switching timestamp at the time of the accident. Traffic light status data is also used to define traffic rule constraints.

[0080] In some embodiments, computer equipment connects to an urban traffic sensing network and uses roadside facilities such as roadside high-definition cameras, millimeter-wave radar, and traffic light status sensors to acquire roadside data of accident vehicles.

[0081] Roadside data can provide a panoramic view of the accident scene and objective environmental data, thereby solving the problem of limited perspective of vehicle-side data and helping to reconstruct the multi-vehicle interaction process.

[0082] Third-party data refers to data from other data sources besides vehicle-side data and roadside data that is relevant to accident liability determination. Third-party data can also be referred to as third-party auxiliary data; however, this application does not limit the name of third-party data in its embodiments.

[0083] Optionally, third-party data may include electronic map data, real-time weather data, and traffic congestion index data.

[0084] Electronic map data is used to describe basic road information about the section of road where the accident occurred. For example, electronic map data may include lane distribution information, speed limits, and road type (urban road, highway). Electronic map data is used to clarify the basic road conditions at the time of the accident.

[0085] Real-time weather data is used to describe weather conditions at the time of an accident. For example, real-time weather data can include weather type (such as sunny, rainy, snowy), visibility, road surface dryness / wetness, and other information. Real-time weather data can quantify the impact of the environment on driving.

[0086] Traffic congestion index data is used to describe the traffic congestion situation on the road segment where the accident occurred. For example, traffic congestion index data may include the real-time congestion level and traffic flow of the accident segment, and it is used to supplement background information of the scene.

[0087] In some embodiments, computer devices connect to electronic map platforms, meteorological service platforms, and traffic management department data platforms through publicly available standardized application programming interfaces (APIs) to obtain third-party data on accident vehicles.

[0088] By introducing third-party data such as external environment and road infrastructure data, the problem of focusing only on the vehicle itself and ignoring external influencing factors when determining liability can be solved, making the factors considered in determining liability more comprehensive and thus improving the accuracy of liability determination.

[0089] In addition, computer equipment can also collect accident-related data in the following ways.

[0090] Specifically, computer equipment can replace traditional roadside equipment and some vehicle-side sensor data by collecting data from Vehicle to Everything (V2X), drone aerial photography, and mobile phone sensor data.

[0091] Among them, V2X directly acquires multi-vehicle interaction and roadside facility status data, drone aerial photography data supplements panoramic data for remote road sections and scenarios without basic equipment, and mobile phone sensor data (GPS, accelerometer) is used to fill the gap of insufficient sensing in old vehicles. The three work together to achieve full-dimensional data coverage.

[0092] Step S204: The computer equipment extracts features from the accident-related data to obtain the accident-related features of the accident vehicle.

[0093] Among them, accident-related characteristics can be understood as factors that have a core impact on accident liability determination.

[0094] Optionally, the accident-related features include at least one of the following: accident spatiotemporal features, vehicle motion features, collision interaction features, environmental impact features, and intelligent driving status features.

[0095] Among them, the spatiotemporal characteristics of an accident are used to reflect the basic spatiotemporal conditions in which the accident occurred and are the basic reference factors for determining liability. For example, the spatiotemporal characteristics of an accident may include the time of the accident (accurate to the second), geographical location (latitude and longitude, lane number), road speed limit standard, and lane distribution type.

[0096] In some embodiments, computer devices can extract spatiotemporal features of an accident from electronic maps and time-synchronized data.

[0097] Vehicle motion characteristics are used to reflect whether the driving behavior of the vehicle involved in the accident was compliant (such as whether it was speeding or changing lanes illegally). For example, vehicle motion characteristics may include instantaneous speed, acceleration change curve, steering angle, lane departure, and following distance within 30 seconds before the collision.

[0098] In some embodiments, a computer device can extract vehicle motion features from vehicle trajectory data.

[0099] Collision interaction features are the core features for determining the severity of an accident and its direct liability. For example, collision interaction features may include collision intensity (unit: g), collision location (front, rear, side), relative speed of the two vehicles at the time of collision, and collision angle.

[0100] In some embodiments, the computer device can extract collision interaction features from collision sensing data and roadside radar data.

[0101] Environmental impact characteristics are used to quantify the indirect impact of the external environment on the occurrence of an accident. For example, environmental impact characteristics may include weather type, visibility, road surface condition, traffic congestion level, and traffic light status at the time of the accident.

[0102] In some embodiments, computer equipment can extract environmental impact features from third-party auxiliary data and roadside data.

[0103] Intelligent driving status features are used for liability determination in L3 / L4 level intelligent driving scenarios. For example, intelligent driving status features may include driving mode (manual, L3, L4), intelligent driving sensor working status (normal, fault), algorithm decision log (such as whether an obstacle ahead is identified), decision delay time (the time from when the sensor detects an obstacle to when the system triggers braking), and emergency braking trigger status.

[0104] In some embodiments, a computer device can extract intelligent driving status features from the intelligent driving system's operating data.

[0105] By extracting features from accident-related data, the accident-related features of the vehicles involved in the accident are obtained, thereby transforming the accident-related data into key influencing factors that can be identified by the algorithm. This covers all dimensions of the accident, including time and space, vehicles, collision, environment, and intelligent driving, thus avoiding omission of core liability determination criteria.

[0106] Step S206: The computer equipment constructs an association map based on accident association features and accident association data.

[0107] The association graph can be understood as a relational network graph constructed based on accident association data. The association graph consists of edges and nodes.

[0108] The nodes in the association graph include involved party nodes, behavior nodes, environment nodes, and device nodes.

[0109] The parties involved refer to entities related to the accident. For example, the parties involved may include human drivers (vehicle owner A, vehicle owner B, etc.), intelligent driving systems (in L3 / L4 mode), and third-party responsible entities (such as road construction companies and traffic light maintenance companies).

[0110] Behavioral nodes can be understood as driving behavior factors that affect accidents. For example, behavioral nodes include speeding, illegal lane changing, failure to yield, following too closely, and failure to warn of sudden braking, among other illegal or improper driving behaviors.

[0111] Environmental nodes can be understood as environmental factors that affect accidents. For example, environmental nodes include environmental or road conditions such as rain, low visibility, traffic congestion, red lights, and road construction.

[0112] Equipment nodes can be understood as the status information of equipment that affects an accident. For example, equipment nodes include the working status (normal or faulty) of related equipment such as vehicle braking systems, intelligent driving sensors, and traffic lights.

[0113] Edges represent the interactive relationships between nodes. For example, "Car Owner A - Speeding Behavior" means that Car Owner A is speeding; "Intelligent Driving System - Sensor Failure" means that the intelligent driving system sensor is malfunctioning; "Speeding Behavior - Collision Accident" means that there is a causal relationship between speeding behavior and collision accident; "Traffic Light Failure - Failure to Avoid" means that traffic light failure indirectly caused failure to avoid.

[0114] In one possible implementation, the computer device constructs nodes in a correlation graph based on accident correlation data. Accident correlation features are then assigned to the nodes to obtain the correlation graph.

[0115] Specifically, the computer equipment first assigns weights to the incident-related features.

[0116] Since collision interaction features are the core influencing factors of accidents and are directly related to the direct causes of accidents; vehicle motion feature weights are key influencing factors of accidents and reflect the compliance of driving behavior; intelligent driving status features are only enabled in L3 / L4 intelligent driving modes and are adapted to the division of responsibility in intelligent driving scenarios; environmental impact features are indirect influencing factors of accidents and help determine the responsibility ratio of the accident vehicle; and accident spatiotemporal features are the basic reference factors for accidents.

[0117] Therefore, computer devices can assign corresponding feature weights to accident-related features in the following manner: collision interaction feature weight is 0.35, vehicle motion feature weight is 0.25, intelligent driving status feature weight is 0.2, environmental impact feature weight is 0.1, and accident spatiotemporal feature weight is 0.1.

[0118] By allocating the aforementioned feature weights, the responsibility contribution of core features (such as collision interaction and vehicle movement) is highlighted, while the interference of redundant information (such as environmental details unrelated to the accident) is suppressed. At the same time, the weight allocation of intelligent driving features ensures the relevance of responsibility determination in intelligent driving scenarios and provides accurate input for subsequent responsibility determination algorithms.

[0119] It should be noted that the weights of accident-related features are not fixed and can be adaptively adjusted according to the accident type (such as chain collisions or single-vehicle accidents). For example, in a chain collision accident, the weight of vehicle motion features can be increased to 0.3, and the weight of environmental impact features can be adjusted to 0.08.

[0120] Then, the computer device can use an attention mechanism algorithm (such as a self-attention mechanism model) to weight and fuse various accident-related features according to the aforementioned weights, generating a one-dimensional vector of accident core features. This vector contains quantitative information on all key influencing factors of the accident. The fused accident core feature vector is then assigned to the involved party nodes, behavior nodes, environment nodes, and equipment nodes in the association graph, becoming the core feature values ​​of the nodes. This provides a unified and standardized quantitative basis for subsequently calculating the association strength between nodes using the cosine similarity algorithm.

[0121] Furthermore, attention-based algorithms can be replaced by federated learning mechanisms. For data privacy-sensitive scenarios involving multiple institutions, distributed training aggregates model parameters without transmitting raw data, achieving multi-source data fusion while meeting data security and compliance requirements.

[0122] Alternatively, the Dempster-Shafer (DS) evidence theory can be adopted to improve the reliability of data fusion in complex environments by modeling uncertain information and using synthesis rules to integrate conflicting information from different data sources.

[0123] Furthermore, computer equipment can also use random forest algorithms and fuzzy comprehensive evaluation methods to replace graph neural networks. The random forest algorithm is used to screen key factors influencing accidents (such as speeding magnitude, collision angle, and intelligent driving decision delay); the fuzzy comprehensive evaluation method establishes a responsibility level membership function, calculates the responsibility proportion through multi-level weighted indicators, reduces reliance on computing power, and is suitable for low-configuration deployment scenarios.

[0124] By constructing a correlation graph, scattered characteristic factors are transformed into structured correlation logic, clearly presenting the causal chain of "who (the party involved) - what did (behavior) - in what environment (environment) - what is the equipment status (equipment) - causing the accident", providing logical support for quantifying responsibility.

[0125] In addition, computer equipment can use causal inference models (such as propensity score matching) to replace correlation graphs. By analyzing the strength of the causal relationship between the behavior of the parties involved and the accident outcome, it can eliminate the interference of confounding factors and accurately quantify the contribution of a single behavior to the accident, especially suitable for complex accidents with multiple superimposed causes.

[0126] Step S208: The computer equipment calculates the responsibility ratio of the accident vehicle based on the correlation graph.

[0127] The liability percentage of the accident vehicle is used to indicate the liability percentage of each party involved in the accident.

[0128] The parties involved refer to entities related to the accident involving the vehicle. For example, the parties involved may include human drivers (vehicle owner A, vehicle owner B, etc.), intelligent driving systems (in L3 / L4 mode), and third-party responsible entities (such as road construction companies and traffic signal maintenance companies).

[0129] In one possible implementation, the computer device calculates the association strength between nodes in the association graph based on the feature values ​​of each node. Then, based on the association strength between nodes in the association graph, it calculates the proportion of responsibility of the accident vehicle.

[0130] Among them, the correlation strength is used to indicate the tightness of the connection between two nodes in the correlation graph.

[0131] In some embodiments, the computer device calculates the association strength between nodes in the association graph based on the feature values ​​of each node in the association graph using a cosine similarity algorithm.

[0132] The formula for calculating cosine similarity can be expressed as: .

[0133] Where A and B are the feature vectors of the two nodes, and S is the association strength. The value of S ranges from 0 to 1. The closer S is to 1, the stronger the association between the two nodes.

[0134] For example, a computer device can calculate the correlation strength between speeding behavior and a collision accident based on speed data in vehicle motion characteristics and collision intensity data in collision interaction characteristics. The greater the speeding and the higher the collision intensity, the closer the correlation strength is to 1.

[0135] Computer equipment can calculate the correlation strength between sensor malfunctions and decision delays based on sensor operating status data and decision delay times in intelligent driving state characteristics. The more severe the sensor malfunction and the longer the decision delay time, the closer the correlation strength is to 1.

[0136] Computer equipment can calculate the correlation strength between red lights and non-yield behavior based on traffic light status in environmental nodes and non-yield behavior data in behavioral nodes. For non-yield behavior during a red light, the correlation strength is close to 1.

[0137] By calculating the correlation strength between nodes, we can quantify the correlation between each factor and the occurrence of the accident and the behavior of each party involved, providing a quantitative basis for calculating the proportion of responsibility.

[0138] In some embodiments, after calculating the correlation strength between nodes, the computer device calculates the responsibility ratio of the accident vehicle based on the correlation strength between each node in the correlation graph using a first formula.

[0139] The first formula can be expressed as: .

[0140] in, Let be the percentage of liability of the j-th party involved, ranging from 0 to 100%, and the sum of the percentages of liability of all parties involved is 100%. Let be the association strength between the j-th party involved and the i-th type of node. Let be the weight value of the i-th type of node.

[0141] Specifically, for each involved node, all behavioral nodes, environmental nodes, and device nodes associated with that involved node are traversed, and the association strength of each association is calculated.

[0142] In some embodiments, the computer device can uniquely categorize behavior nodes, environment nodes, and device nodes into five types of accident-related features: spatiotemporal accident features, vehicle motion features, collision interaction features, environmental impact features, and intelligent driving status features. For example, speeding behavior in behavior nodes can be categorized into vehicle motion features, environment nodes into environmental impact features, and intelligent driving sensor malfunctions in device nodes into intelligent driving status features.

[0143] Then, the computer equipment can use feature label encoding and mapping databases to bind the feature weights of corresponding accident-related features, i.e., the weight values ​​of the corresponding nodes, to each type of node. When calculating the proportion of responsibility for each party involved, the relevant data is automatically retrieved. This is used for calculating the proportion of responsibility.

[0144] When the vehicle involved in the accident is in L3 or L4 level intelligent driving mode at the time of the accident, the intelligent driving scenario responsibility determination submodule is automatically activated to determine the proportion of responsibility for the accident.

[0145] Specifically, computer equipment can collect decision logs from the intelligent driving system and determine the system's status based on these logs. If the intelligent driving system has sensor malfunctions, decision delays exceeding 0.5 seconds, or algorithmic defects, the correlation strength of the intelligent driving status characteristics will be considered when calculating the proportion of responsibility. The responsibility contribution of the intelligent driving system is increased by 1.2 times, and the responsibility contribution of the intelligent driving system shall not exceed 70% at most; if the intelligent driving system is completely normal (no sensor failure, decision delay less than or equal to 0.5 seconds, and no algorithm defects), the responsibility contribution shall be mainly allocated to the human driver or other parties involved.

[0146] When the number of parties involved in an accident is greater than two, the computer equipment can automatically identify all the parties involved in the accident (including multiple vehicles and third-party responsible parties) through the association graph, calculate the responsibility ratio of each party, and ensure that no responsible party is omitted.

[0147] When there are relationships such as "road construction - lane occupancy" or "traffic light malfunction - illegal passage" in the association graph, the computer equipment will automatically include the third-party responsible party in the involved party node, and the responsibility ratio allocation ranges from 0% to 30%. The specific value of the responsibility ratio of the three parties is calculated based on the association strength.

[0148] The above technical solutions can effectively address the issues of ambiguous responsibility in intelligent driving scenarios, difficulties in delineating responsibility among multiple parties, and lack of third-party liability identification, thereby improving the scenario adaptability of the technical solutions.

[0149] In addition, computer equipment can also quantify the proportion of responsibility in the following ways.

[0150] It employs a Transformer model and a rule engine. The Transformer model is used to capture the temporal correlation features of multiple stakeholders, behaviors, and environments. The rule engine is embedded in traffic regulations and intelligent driving safety standards (such as ISO 21448).

[0151] Specifically, the computer device first outputs the responsibility contribution ranking through the Transformer model, and then the rule engine quantifies it into a responsibility ratio of 0-100%. This method can adapt to fast reasoning in large-scale scenarios with multiple features.

[0152] In summary, the technical solution provided in this application obtains accident-related data from multiple data sources, extracts features from the accident-related data to obtain accident-related features, constructs a correlation graph based on the accident-related features and accident-related data, and then calculates the responsibility ratio of the accident vehicles based on the correlation graph. By determining responsibility based on multi-source data, it can avoid the deviation in responsibility determination caused by data bias of single-source data, thus improving the accuracy of responsibility determination. In addition, the calculation of the responsibility ratio of the accident vehicles based on the correlation graph can output a quantitative responsibility ratio of 0-100%.

[0153] Furthermore, after the computer device acquires the accident-related data of the accident vehicle, the computer device can also perform data preprocessing on the accident-related data, that is, step S2031 is also included between step S202 and step S204.

[0154] Figure 3 This is a flowchart illustrating another method for determining liability provided in an embodiment of this application. This method can be executed by a computer device.

[0155] Step S2031: The computer equipment performs data processing on the accident-related data, including at least one of outlier removal, missing value filling, or standardization.

[0156] Outlier removal refers to identifying and removing abnormal data from accident-related data.

[0157] In some embodiments, the computer device may employ an interquartile range (IQR) algorithm to identify and remove outliers (such as extreme speed values ​​caused by momentary sensor malfunctions) from collision sensing data and trajectory data. The formula for outlier removal using the IQR algorithm can be: remove data smaller than Q1 - 1.5IQR or greater than Q3 + 1.5IQR, thereby ensuring data authenticity.

[0158] Missing value imputation refers to supplementing missing data in accident-related data.

[0159] For example, for missing frames in dashcam footage (such as frame loss due to transmission interruption), computer devices can use video frame interpolation algorithms to ensure the continuity of the video data.

[0160] Video frame interpolation algorithms mainly analyze the pixel features between adjacent valid frames to generate high-quality intermediate images, thereby achieving smooth completion of missing frames.

[0161] Standardization can be understood as unifying the units, formats, and other aspects of accident-related data.

[0162] In some embodiments, the computer device may use the standard score (Z-score) standardization method to normalize continuous data such as velocity, acceleration, and collision intensity.

[0163] Specifically, the Z-score formula can be expressed as: .

[0164] Where X represents the original accident-related data collected. The mean, The standard deviation is denoted as .

[0165] Data standardization is used to eliminate the differences in the units of measurement of different data dimensions, which facilitates subsequent feature fusion.

[0166] After the computer device acquires the accident-related data of the accident vehicle, in addition to preprocessing the accident-related data, the computer device can also perform spatiotemporal alignment processing on the accident-related data. That is, step S2032 is also included between step S202 and step S204.

[0167] Figure 4 This is a flowchart illustrating another method for determining liability provided in an embodiment of this application. This method can be executed by a computer device.

[0168] Step S2032: The computer equipment performs time alignment processing and / or spatial location alignment processing on the accident-related data.

[0169] In some embodiments, the computer device may employ a collision sensor data timestamp master synchronization strategy to time-align accident-related data.

[0170] Among them, the collision sensor data timestamp master synchronization strategy refers to using the precise timestamp of the vehicle-side collision sensor data (accurate to the millisecond level) as a benchmark, and adjusting the time dimension of dashcam images, roadside radar data, and third-party data through timestamp mapping to ensure that the time error of all data is less than or equal to 10ms, thereby solving the problem of time synchronization of accident-related data.

[0171] In some embodiments, computer equipment may employ a strategy combining Global Positioning System (GPS)-assisted calibration and coordinate system transformation to spatially align accident-related data.

[0172] The strategy of combining GPS-assisted calibration and coordinate system transformation refers to first calibrating the position using vehicle GPS data and roadside equipment GPS data, and then converting all data from the WGS84 coordinate system to a planar coordinate system (such as the UTM coordinate system) to achieve spatial alignment of vehicle trajectory data, roadside radar detection data, and electronic map data, strictly controlling the spatial error within ±0.5 meters.

[0173] By aligning accident-related data in time and space, the temporal and spatial barriers of accident-related data are broken down, and a complete data link that is consistent in time and unified in space is constructed, laying the foundation for subsequent feature extraction and correlation analysis.

[0174] Furthermore, the accident-related data that has undergone data preprocessing and spatiotemporal alignment are integrated to generate a unified accident dataset containing trajectories, collisions, images, roadside data, and third-party data. Then, the accident dataset is standardized in terms of data format, transforming it into a structured format that can be directly used for algorithm processing.

[0175] In addition, computer equipment can also perform spatiotemporal alignment processing on accident-related data using the Laboratory Virtual Instrument Workbench (LabVIEW) virtual instrument technology. Through dual-buffer synchronization and programmable function interface (PFI) line-triggered calibration mechanisms, millisecond-level time synchronization and spatial calibration of multi-channel data can be achieved, supporting flexible hardware replacement and reducing system integration costs.

[0176] It should be noted that the above-described method for determining liability may include both steps S2031 and S2032, or it may include only one of steps S2031 and S2032. Furthermore, steps S2031 and S2032 may be executed in either the order of S2031 and S2032 or the order of S2032 and S2031; this embodiment does not impose specific limitations on this.

[0177] Furthermore, after the computer device calculates the proportion of responsibility of the accident vehicle, the computer device can also generate accident scene reconstruction information based on the accident-related data, that is, step S209 is included after step S208.

[0178] Figure 5 This is a flowchart illustrating another method for determining liability provided in an embodiment of this application. This method can be executed by a computer device.

[0179] Step S209: The computer equipment generates accident scene reconstruction information based on the accident-related data.

[0180] Among them, accident scene reconstruction information is used to describe the scene at the time of the accident. Accident scene reconstruction information includes at least one of spatiotemporal reconstruction information, process reconstruction information, or state reconstruction information.

[0181] Spatiotemporal reconstruction information can be understood as information that can reconstruct the time and space of an accident. For example, spatiotemporal reconstruction information can be a "visualized chart of spatiotemporal information of an accident".

[0182] Specifically, computer equipment can mark the precise location (latitude and longitude, lane number) of the accident on an electronic map platform, and overlay information such as the time of the accident (accurate to the second), road speed limit standards, lane distribution, and traffic light status to generate an "accident spatiotemporal information visualization chart".

[0183] Among them, the "Accident Spatiotemporal Information Visualization Chart" includes an accident scene map, key timeline nodes (such as collision time and braking trigger time), and a road basic information table, which can clearly present the basic spatiotemporal conditions of the accident, making the "scenario background" for determining liability clear at a glance.

[0184] Process reconstruction information can be understood as relevant information that can reconstruct the complete course of an accident. For example, process reconstruction information can be a "trajectory animation simulation".

[0185] Specifically, computer equipment can extract key video clips from 30 seconds before the accident to 10 seconds after the collision by fusing dashcam footage and roadside camera footage. Then, it can combine vehicle trajectory data to generate a "trajectory animation simulation" (using 3D animation rendering technology). The animation synchronously displays the movement trajectory, speed changes, and relative positions of multiple vehicles at the moment of collision.

[0186] The "Trajectory Animation Simulation" also supports functions such as animation speed adjustment, keyframe annotation (e.g., annotating key nodes such as "Vehicle Owner A begins speeding" and "Intelligent Driving System triggers braking"), and multi-view switching (vehicle view, roadside panoramic view). By generating "Trajectory Animation Simulation", the complete process of an accident can be intuitively recreated, making the key aspects of "who violated the rules first and how the collision occurred" traceable.

[0187] State restoration information can be understood as information that can reconstruct the state of the vehicle at the time of the accident. Optionally, state restoration information may include speed change curves, collision intensity distribution heatmaps, and intelligent driving system decision-making flowcharts.

[0188] Specifically, computer devices can generate various quantitative data charts through data visualization technology. The content presented in these quantitative data charts can include data charts, key parameter values ​​(such as collision intensity 4.2g, decision delay 0.8 seconds), and device status labels (such as "intelligent driving sensor normal" and "braking system triggered").

[0189] Among them, the speed change curve can show the speed changes of each vehicle involved before and after the collision.

[0190] Collision intensity distribution heatmaps can indicate the intensity distribution at the points of vehicle collision.

[0191] The intelligent driving system decision-making flowchart (intelligent driving scenario) can display the intelligent driving sensor data input, algorithm decision-making process, braking trigger time, etc. in the form of a time sequence diagram.

[0192] By generating various quantitative data charts, the core data in the accident process is presented in a quantitative manner, providing "data evidence" for determining the proportion of responsibility and making the logic of determining responsibility transparent.

[0193] Furthermore, after obtaining the spatiotemporal information visualization charts, "trajectory animation simulations," and various quantitative data charts, these charts and text descriptions are integrated to generate a standardized multimodal report. This multimodal report supports online viewing and offline PDF export.

[0194] The written description may include an overview of the accident, the proportion of responsibility of each party involved, the basis for determining responsibility (key features, correlation strength, weight allocation), and the conclusion of the scene reconstruction.

[0195] By generating multimodal reports, we can provide insurance claims adjusters, traffic police, and car owners with intuitive, easy-to-understand, and verifiable evidence for determining liability, thereby significantly reducing the dispute rate and enhancing the credibility of the liability determination results.

[0196] Furthermore, computer equipment can utilize digital twin and VR / AR technologies to construct a 1:1 three-dimensional twin of the accident scene, integrating multi-source data to dynamically reconstruct vehicle trajectories, collision processes, and intelligent driving system decision-making processes. Users can immerse themselves in key moments using VR devices or scan the scene with AR to reconstruct the accident sequence, improving the interpretability of liability determination results.

[0197] Computer equipment can also generate process reconstruction information by combining 3D reconstruction and physical simulation. Specifically, computer equipment can generate point clouds and solid mesh models of the accident scene through triangulation reconstruction algorithms, and combine them with physics engines (such as collision dynamics models) to simulate the accident evolution path, supporting "changing variables to see the results" (such as adjusting vehicle speed and braking timing), providing more intuitive evidence for liability determination.

[0198] In addition, after the computer device calculates the responsibility ratio of the accident vehicle, the computer device can also update the weight value of the node in the association graph and the node category in the association graph. That is, after step S208, steps S210 and S211 are also included.

[0199] Figure 6 This is a flowchart illustrating another method for determining liability provided in an embodiment of this application. This method can be executed by a computer device.

[0200] Step S210: The computer device acquires historical liability determination data.

[0201] Historical liability determination data is used to indicate relevant data on liability determinations that have been completed in past liability cases. For example, historical liability determination data may include automatic liability determination results, manual review feedback results, and accident type tags.

[0202] Specifically, computer devices can connect to the insurance company's system through an Application Programming Interface (API) to obtain historical liability determination data.

[0203] Step S211: The computer device updates the weight values ​​of nodes in the association graph based on historical responsibility data, and / or updates the node categories in the association graph.

[0204] In some embodiments, the computer device may update the feature weights corresponding to each incident-related feature in the following manner.

[0205] Specifically, the gradient descent algorithm is adopted to iteratively update the weight allocation of each accident-related feature with the goal of minimizing the error between the automatic liability determination result and the manual review result (such as increasing the weight of vehicle motion features when the proportion of chain collision accidents increases).

[0206] By dynamically updating the feature weights, the feature weights can be adapted to the distribution characteristics of accident types in different periods and regions, ensuring that the accuracy rate of liability determination remains stable at over 92%.

[0207] In some embodiments, a computer device may update the node categories in an association graph in the following manner.

[0208] Specifically, when L4 and above intelligent driving technology becomes widespread or the functions of intelligent driving systems are upgraded, "high-level intelligent driving system responsibility judgment factors" will be added, such as the algorithm decision delay threshold (which can be adjusted to 0.3 seconds for L4 intelligent driving) and the multi-sensor fusion fault judgment standard.

[0209] For new functions of intelligent driving systems (such as automatic lane changing and highway navigation), corresponding behavioral nodes (such as "automatic lane changing without observing blind spots") and associated rules can be added to ensure the comprehensiveness of liability determination in intelligent driving scenarios, continuously adapt to the development of intelligent driving technology, and avoid the failure of liability determination models due to upgrades in intelligent driving technology.

[0210] Furthermore, computer equipment can also analyze historical liability data to identify new accident scenarios (such as chain collisions, sudden appearance of roadside obstacles, and multi-vehicle interaction accidents).

[0211] For each new type of accident scenario, new exclusive accident-related features and rules have been added. For example, a multi-vehicle collision sequence feature has been added for the new type of chain-reaction collision accident scenario; a location and warning feature has been added for the new type of accident scenario where roadside obstacles suddenly appear; and a new association rule has been added for "no warning signs are set for obstacles - increased third-party liability ratio".

[0212] In addition, computer equipment can incorporate new accident scenario data into the training set of graph neural network models, fine-tune the graph neural network models, and improve the graph neural network models' ability to determine responsibility for abnormal scenarios.

[0213] By updating the weight values ​​and node categories in the association graph, the scenario coverage of the technical solution can be expanded, avoiding failure in liability determination or a decrease in accuracy due to special accident scenarios.

[0214] In addition, for intelligent driving systems, computer equipment can also construct a fault tree for the intelligent driving system based on Fault Tree Analysis (FTA) and the ISO 21448 functional safety standard (the top event is "accident occurs", the intermediate events are "sensor failure", "delay in decision-making", etc.), identify the fault source through minimal cut sets, and combine ISO 21448 to quantify the expected functional deficiencies of the intelligent driving system and clarify the human-machine responsibility boundary.

[0215] Computer equipment can also use Model Predictive Control (MPC) to analyze the rationality of intelligent driving system decisions. By reverse-engineering the optimal decision path of the intelligent driving system before the accident and comparing it with the actual decision, the impact of decision deviation on the accident can be quantified and used as a basis for liability division.

[0216] In summary, the technical solution provided in this application obtains accident-related data from multiple data sources, extracts features from the accident-related data to obtain accident-related features, constructs a correlation graph based on the accident-related features and accident-related data, and then calculates the responsibility ratio of the accident vehicles based on the correlation graph. By determining responsibility based on multi-source data, it can avoid the deviation in responsibility determination caused by data bias of single-source data, thus improving the accuracy of responsibility determination. In addition, the calculation of the responsibility ratio of the accident vehicles based on the correlation graph can output a quantitative responsibility ratio of 0-100%.

[0217] Furthermore, the technical solutions provided in this application embodiment can also be implemented in the following ways. These solutions can be executed by a computer device.

[0218] Method 1: Computer devices can determine the responsibility of accident vehicles through a collaborative model of rapid processing at the edge and precise determination of responsibility in the cloud.

[0219] Specifically, lightweight data acquisition and preprocessing modules are deployed at the edge (such as vehicle-side and roadside edge nodes). Drones and handheld LiDAR are used to quickly collect on-site data. Simultaneous Localization and Mapping (SLAM) technology is used to generate 3D real-world data with an accuracy of ±2mm. Outlier removal and basic time-series alignment are completed in real time, and only key features (such as collision time, core speed data, and intelligent driving fault indicators) are uploaded.

[0220] The cloud is used to receive key features from the edge, and Bayesian networks and transfer learning are used to quantify responsibility. Bayesian networks model the probabilistic relationship between behavior, environment, and responsibility, and transfer learning reuses the parameters of the trained model to quickly adapt to new regions or new intelligent driving levels. A simplified scene reconstruction report is generated through the digital twin platform, which supports online viewing and export.

[0221] Simple accident data is recorded at the edge and uploaded to the cloud regularly to update the model. The cloud uses reinforcement learning to iteratively optimize the probability parameters of the Bayesian network to ensure scene adaptability.

[0222] By assigning responsibility in the above manner, the responsibility assignment cycle can be compressed to the minute level, the edge processing latency is less than or equal to 500ms, the accuracy of responsibility division in intelligent driving scenarios is on par with the original solution, and the hardware cost is reduced by 30%-50%.

[0223] Method 2: Computer devices can also determine liability for accident vehicles through a hybrid mode of rule engine processing for simple scenarios and machine learning processing for complex scenarios.

[0224] Specifically, the rule engine layer has a large number of (e.g., 100+) basic traffic rules and simple accident liability determination templates (e.g., "Red light straight-through collision with green light left turn - straight-through party fully responsible", "Single-vehicle collision with fixed object - driver fully responsible") built in, which can directly handle more than 60% of simple accidents and output the liability percentage within 10 seconds.

[0225] The machine learning layer is used to determine liability in complex accidents (such as multi-vehicle interactions, intelligent driving scenarios, and third-party liability), specifically employing a model that combines XGBoost with causal inference.

[0226] XGBoost is used to extract the importance of multi-source features, while causal inference is used to eliminate interference from irrelevant factors and quantify the responsibilities of each party involved. For intelligent driving scenarios, the fault diagnosis logs of the intelligent driving system can also be accessed, and human and machine responsibilities can be divided through Failure Mode and Effects Analysis (FMEA).

[0227] In addition, computer equipment can also recreate scenes in the following ways: for simple accidents, output standardized text and key photo reports; for complex accidents, call the lightweight 3D reconstruction module to generate snapshots of key accident nodes and trajectory diagrams.

[0228] This method of determining liability for accidents can improve the efficiency of determining liability for simple accidents by 300%, ensure the accuracy of complex accidents, and has a low system deployment threshold. This method is suitable for small and medium-sized insurance companies and grassroots traffic law enforcement scenarios.

[0229] Method 3: Computer equipment can also use V2X data as the core data source, digital twins as the scene reconstruction carrier, and causal inference as the core of responsibility quantification to determine the responsibility of accident vehicles.

[0230] Specifically, computer equipment can acquire real-time data on the movement of multiple vehicles, the status of roadside traffic lights, and interactive data from intelligent driving systems via V2X networks, without needing to connect to a separate third-party platform, and can simultaneously access the real-time data interface of the meteorological department to supplement environmental information.

[0231] Then, computer equipment can use multi-camera spatiotemporal synchronization technology (PTP / GPS time synchronization and geometric calibration) to unify V2X multi-source data into the same three-dimensional coordinate system with an error of less than or equal to 1 cm, and generate a structured accident dataset.

[0232] Finally, based on causal inference models, such as structural causal modeling (SCM), a causal graph is constructed between the parties involved, behaviors, and accidents. The responsibility contribution of a single behavior is quantified through intervention analysis. In the intelligent driving scenario, an additional "intelligent driving decision intervention variable" is introduced to clarify the system responsibility ratio.

[0233] Computer equipment can also utilize a digital twin platform to reconstruct the dynamic process of an accident in real time, supporting multi-view switching and key parameter annotation. Furthermore, it can adapt to new intelligent driving levels through transfer learning and periodically import historical data to optimize the causal graph structure.

[0234] This method can increase the coverage of data collection by 20%, make the responsibility division logic of intelligent driving scenarios more transparent, and reduce the dispute rate to the same level as the technical solution provided in this application embodiment. It can also be seamlessly integrated with the smart city traffic management system.

[0235] Figure 7 This is a schematic diagram of the structure of a liability determination system provided in an embodiment of this application.

[0236] like Figure 7 As shown, the responsibility determination system includes a data acquisition module 701, a data preprocessing module 702, a spatiotemporal alignment module 703, a feature fusion module 704, a responsibility determination module 705, a scene restoration module 706, and an optimization module 707.

[0237] Specifically, the data acquisition module 701 is used to collect vehicle-side data, roadside data, and third-party data. The data acquisition module and the data preprocessing module are connected through a standardized API interface.

[0238] The data preprocessing module 702 is used to preprocess the vehicle-side data, roadside data, and third-party data collected by the data acquisition module. The data preprocessing includes at least one of outlier removal, missing value filling, and standardization.

[0239] The spatiotemporal alignment module 703 is used to perform spatiotemporal alignment processing on accident-related data, with time synchronization less than or equal to 10ms and spatial calibration less than or equal to 0.5m.

[0240] The feature fusion module 704 is used to extract accident-related features, and then, based on the accident-related features, the accident-related features are weighted and fused using an attention mechanism.

[0241] The liability determination module 705 is used to construct a correlation graph, which uses a graph neural network to build a four-dimensional correlation graph and calculate the liability ratio of each party involved in the accident.

[0242] The scene restoration module 706 is used to generate accident scene restoration information based on accident-related data. The accident scene restoration information includes at least one of spatiotemporal restoration information, process restoration information, or state restoration information, and outputs a multimodal liability determination report based on the accident scene restoration information.

[0243] The optimization module 707 is used to analyze historical liability determination data, update the weight values ​​of nodes in the association graph and the node categories in the association graph based on the analysis results, iterate the liability determination rules, and adapt to new accident scenarios.

[0244] The foregoing mainly describes the solution provided in this application. Accordingly, this application also provides a liability determination device for implementing the above-described method embodiments.

[0245] like Figure 8 The schematic diagram of the liability determination device shown indicates that the device may include an acquisition module 801, a processing module 802, a construction module 803, and a calculation module 804. The acquisition module 801 is used to execute... Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 The operation of step S201 in the illustrated method, Figure 3 The operation of step S2031 in the illustrated method and Figure 4 The illustrated method includes step S2032; the processing module 802 is used to execute... Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 The operation of step S202 in the illustrated method and Figure 5 The illustrated method includes step S209; the construction module 803 is used to execute... Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 The illustrated method includes step S203; the calculation module 804 is used to execute... Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 The illustrated method includes step S204.

[0246] In some embodiments, the determination device includes hardware structures and / or software modules corresponding to the execution of each function in order to achieve the above-described functions. Those skilled in the art will readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0247] This application embodiment can divide the responsibility determination device into functional modules according to the above method embodiment. For example, each function can be divided into a separate functional module, or two or more functions can be integrated into one responsibility determination module. The integrated module can be implemented in hardware or as a software functional module. It should be noted that the module division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0248] like Figure 9 As shown, the computer device provided in this application embodiment may include a processor 901, a bus 902, a communication interface 903, and a memory 904. The processor 901, memory 904, and communication interface 903 communicate with each other via the bus 902. It should be understood that this application does not limit the number of processors and memories in the network device.

[0249] The 902 bus can be a PCI bus, an Extended Industry Standard Architecture (EISA) bus, or a UB bus, etc. Buses can be divided into address buses, data buses, control buses, etc. For ease of representation, Figure 9 The bus 902 may be represented by a single line, but this does not mean that there is only one bus or one type of bus. The bus 902 may include a path for transmitting information between various components of the network device (e.g., memory 904, processor 901, communication interface 903).

[0250] Processor 901 may include any one or more processors such as CPU, graphics processing unit (GPU), microprocessor (MP), or digital signal processor (DSP).

[0251] The memory 904 may include volatile memory, such as random access memory (RAM). The processor 901 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0252] The communication interface 903 uses transceiver modules, such as, but not limited to, network interface cards and transceivers, to enable communication between network devices and other devices or communication networks.

[0253] The memory 904 stores executable program code, and the processor 901 executes the executable program code to implement the functions of the aforementioned method embodiments. That is, the memory 904 stores instructions for executing the above-described liability determination method.

[0254] In another aspect, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor to implement the liability determination method provided in the above-described method embodiments.

[0255] In another aspect, a computer program product is provided, which includes a computer program or instructions that, when executed by a processor, implement the liability determination method provided in the above-described method embodiments.

[0256] Through the above description of the implementation methods, those skilled in the art will clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the module can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, modules, and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0257] Since the liability determination module, computer-readable storage medium, and computer program product in the embodiments of the present invention can be applied to the above methods, the technical effects they can achieve can also be referred to the above method embodiments. The embodiments of the present invention will not be described again here.

[0258] The method steps in this embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. Alternatively, the ASIC can reside in a network device. Of course, the processor and storage medium can also exist as discrete components in the network device.

[0259] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer programs or instructions. When a computer program or instruction is loaded and executed on a computer, the processes or functions of the embodiments of this application are performed, in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user equipment, or other programmable module. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, a computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; it can also be an optical medium, such as a digital video disc (DVD); or it can be a semiconductor medium, such as a solid-state drive (SSD).

[0260] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations 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. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for determining liability, characterized in that, The method includes: Obtain accident-related data of the accident vehicle, wherein the accident-related data is used to indicate accident-related data from multiple data sources; Feature extraction is performed on the accident-related data to obtain the accident-related features of the accident vehicle; Based on the accident association features and the accident association data, an association map is constructed; Based on the correlation graph, the responsibility percentage of the accident vehicle is calculated, and the responsibility percentage is used to indicate the responsibility percentage of each party involved in the accident.

2. The method according to claim 1, characterized in that, The calculation of the responsibility percentage of the accident vehicle based on the correlation graph includes: Based on the feature values ​​of each node in the association graph, the association strength between each node in the association graph is calculated; Based on the correlation strength between nodes in the correlation graph, the responsibility percentage of the accident vehicle is calculated.

3. The method according to claim 2, characterized in that, The step of calculating the association strength between nodes in the association graph based on the feature values ​​of each node in the association graph includes: Based on the feature values ​​of each node in the association graph, the association strength between each node in the association graph is calculated using the cosine similarity algorithm.

4. The method according to claim 2, characterized in that, The calculation of the responsibility percentage of the accident vehicle based on the correlation strength between nodes in the correlation graph includes: Based on the correlation strength between nodes in the correlation graph, the responsibility percentage of the accident vehicle is calculated using the first formula.

5. The method according to claim 4, characterized in that, The first formula is expressed as: in, Let j be the percentage of responsibility of the j-th party involved. The strength of the association between the j-th party involved and the i-th type of node. Let be the weight value of the i-th type of node.

6. The method according to claim 1, characterized in that, The construction of the association graph based on the accident association features and the accident association data includes: Based on the accident-related data, the nodes in the association graph are constructed; The accident-related features are assigned to the nodes to obtain the association graph.

7. The method according to claim 6, characterized in that, The nodes in the association graph include: involved party nodes, behavior nodes, environment nodes, and device nodes.

8. The method according to any one of claims 1 to 7, characterized in that, The method further includes: Based on the accident-related data, accident scene reconstruction information is generated, which includes at least one of spatiotemporal reconstruction information, process reconstruction information, or state reconstruction information.

9. The method according to any one of claims 1 to 7, characterized in that, The method further includes: Obtain historical liability determination data; Based on the historical responsibility data, the weight values ​​of nodes in the association graph are updated, and / or the node categories in the association graph are updated.

10. The method according to any one of claims 1 to 7, characterized in that, The method further includes: The accident-related data is processed, including at least one of outlier removal, missing value filling, or standardization.

11. The method according to any one of claims 1 to 7, characterized in that, The method further includes: The accident-related data is subjected to time alignment and / or spatial location alignment.

12. A liability determination device, characterized in that, The device includes: The acquisition module is used to acquire accident-related data of the accident vehicle, wherein the accident-related data is used to indicate accident-related data from multiple data sources; The processing module is used to extract features from the accident-related data to obtain the accident-related features of the accident vehicle. The construction module is used to construct an association map based on the accident association features and the accident association data; The calculation module is used to calculate the responsibility ratio of the accident vehicle based on the correlation map, and the responsibility ratio is used to indicate the responsibility ratio of each party involved in the accident.

13. A computer device, characterized in that, The computer device includes a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor to implement the liability determination method as described in any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to implement the liability determination method as described in any one of claims 1 to 11.

Images