AI intelligent vehicle property damage assessment method and system supporting multi-mode interaction

Abstract

This application provides an AI-powered intelligent vehicle and property damage assessment method and system supporting multi-mode interaction, applicable to damage assessment management of non-motorized vehicles and urban public facilities. The method acquires first and second interaction data of the damaged object before and after the damage, analyzes and forms baseline feature information and damage description information, and constructs a damage assessment correlation graph based on preset association rules. Through the baseline feature information and damage description information, difference factors are generated, and preset damage assessment indicators are dynamically adjusted accordingly to obtain target damage assessment indicators. Subsequently, the target damage assessment indicators are input into a preset damage assessment model to generate damage assessment data, resulting in damage assessment results including damage level, damage range, damaged components, and repair costs. Simultaneously, it supports the integration of damage assessment items and corrections by damage assessors, generating a final damage assessment report, thus achieving efficient damage assessment management for multiple types of vehicles and property.

Description

Technical Field

[0001] This application relates to the field of damage assessment technology, and in particular to an AI-powered intelligent vehicle and property damage assessment method and system that supports multi-mode interaction. Background Technology

[0002] With the improvement of digitalization in urban transportation facility management, the demand for damage assessment of non-motorized vehicles, shared mobility tools, and urban public infrastructure is constantly increasing. The management of multiple types of damage assessment objects has gradually become a core task in accident handling, operation and maintenance management, and asset risk control.

[0003] Existing intelligent damage assessment technologies mainly focus on establishing a standardized parts system and damage level system around the structure of motor vehicles, and use image recognition models to locate and assess damaged parts.

[0004] However, these methods are difficult to apply directly to non-motorized vehicles and various third-party facilities. In existing application scenarios, damage identification and level assessment of different damage assessment objects usually rely on human experience for judgment. Various damage assessment objects lack a unified way of expressing component structure and a damage description framework. Therefore, existing technologies lack a unified basic feature system that can be built based on multiple types of damage assessment objects, which limits the level of automation of the damage assessment process and results in low damage assessment efficiency. Summary of the Invention

[0005] This application provides an AI-powered intelligent vehicle and property damage assessment method and system that supports multi-modal interaction. Its core lies in: acquiring multi-modal interaction data of the damaged object before and after damage; constructing comparable baseline feature information and damage description information; forming a damage assessment correlation graph using preset association rules; generating quantitative difference factors based on damage assessment edges; dynamically calibrating the indicators based on preset damage assessment indicators to obtain target damage assessment indicators; subsequently inputting the target damage assessment indicators into a preset damage assessment model to generate damage assessment data; and performing structured matching of the damage assessment data through the damage assessment nodes of the damage assessment correlation graph to form a damage assessment result including damage level, damage range data, damaged component data, and repair costs. This enables damage assessment management of multiple types of vehicles and property, improving damage assessment efficiency.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] This application provides an AI-powered intelligent vehicle and property damage assessment method that supports multi-modal interaction. The method may include:

[0008] The first interaction data of the damaged object is obtained before the damage is assessed. The first interaction data is analyzed to form a first damage assessment information set, which may include benchmark feature information.

[0009] The second interactive data after the damage to the object to be assessed is obtained, and the damage is identified by combining the second interactive data with the first damage assessment information set to form a second damage assessment information set, which may include damage description information.

[0010] According to the preset association rules, the first damage assessment information set and the second damage assessment information set are associated to obtain a damage assessment association graph, which may include damage assessment nodes and damage assessment edges.

[0011] Based on the damage assessment edge, the baseline feature information in the first damage assessment information set is compared with the damage description information in the second damage assessment information set to generate a difference factor;

[0012] Obtain the preset loss assessment indicators for the loss assessment object, adjust the preset loss assessment indicators according to the difference factors to obtain the target loss assessment indicators, input the target loss assessment indicators into the preset loss assessment model, perform loss assessment, and generate loss assessment data;

[0013] The damage assessment nodes in the damage assessment correlation graph are used as index items. The index items are associated and matched with the damage assessment data to obtain the damage assessment results. The damage assessment results may include damage level, damage range data, damaged component data, and repair costs.

[0014] In some possible implementations, the first damage assessment information set and the second damage assessment information set are associated according to preset association rules to obtain a damage assessment association map, which may include:

[0015] According to the preset association rules, the benchmark feature information in the first loss assessment information set is matched with the damage description information in the second loss assessment information set item by item to establish corresponding association records;

[0016] The associated records are integrated, and a loss assessment association map is generated based on the preset map structure.

[0017] In some possible implementation methods, the preset association rules may include category matching rules, component association rules, and weight allocation rules. Based on these preset association rules, the baseline feature information in the first damage assessment information set is matched item by item with the damage description information in the second damage assessment information set to establish corresponding association records. This may include:

[0018] Based on category matching rules, the baseline feature information and the damage description information are combined according to category to generate category matching records;

[0019] Based on component association rules, the baseline feature information and the damage description information are associated and matched according to the component structural relationship, component spatial location relationship and component functional relationship to generate component association records;

[0020] Based on the weighting rules, the baseline feature information and the damage description information are weighted and combined according to the risk impact weight, structural importance weight, and cost sensitivity weight to generate weighted association records.

[0021] In some possible implementations, integrating related records and generating a loss assessment correlation map based on a preset map structure may include:

[0022] Based on the preset map structure, multiple damage assessment nodes are constructed for the benchmark feature information and damage description information in the category matching record, component association record, and weight association record, respectively.

[0023] Construct corresponding loss assessment edges for each loss assessment node to generate a loss assessment association graph.

[0024] In some possible implementations, based on the damage assessment edge, the baseline feature information in the first damage assessment information set is compared with the damage description information in the second damage assessment information set to generate a difference factor, which may include:

[0025] Traverse the loss assessment edges in the loss assessment association graph and obtain the loss assessment nodes connected by the loss assessment edges;

[0026] The baseline feature information in the damage assessment node is compared with the damage description information item by item to generate comparison results;

[0027] Based on the comparison results, numerical or hierarchical difference indicators are generated to form difference factors, which are used to represent the degree of change in the state of the damaged object before and after the damage.

[0028] In some possible implementation methods, the preset loss assessment indicators of the loss assessment object are obtained, and the preset loss assessment indicators are adjusted according to the difference factors to obtain the target loss assessment indicators, which may include:

[0029] Obtain preset damage assessment indicators, which may include damage level, damaged range, component identification, and repair costs;

[0030] The damage level, damaged area, component identification, and repair cost are adjusted according to the difference factors to obtain the target damage assessment index. The target damage assessment index may include the target damage level, the target damaged area, the target component identification, and the target repair cost.

[0031] In some possible implementation methods, the target loss assessment indicators are input into a pre-defined loss assessment model to perform loss assessment and generate loss assessment data. This may include:

[0032] The target damage level, target damage range, target component identification, and target repair cost are used as input parameters and input into a preset damage assessment model. The preset damage assessment model is then used to integrate the input parameters and generate the corresponding damage assessment data.

[0033] In some possible implementations, loss assessment nodes in the loss assessment correlation graph are used as index items. These index items are then correlated and matched with the loss assessment data to obtain the loss assessment result. This can include:

[0034] Using the loss assessment nodes in the loss assessment correlation graph as index items, the corresponding loss assessment data is retrieved based on the index items to obtain the first loss assessment data;

[0035] The first loss assessment data is associated and matched with the index items to generate loss assessment entries for the loss assessment objects;

[0036] The damage assessment items are integrated according to the preset format to generate the damage assessment results.

[0037] In some possible implementation methods, after integrating the damage assessment items according to a preset format to obtain the damage assessment result, it may also include:

[0038] Trigger loss assessment correction, responding to the loss assessor's correction of the loss assessment results, and generate a loss assessment report.

[0039] An AI-powered intelligent vehicle and property damage assessment system that supports multi-mode interaction may include: an interaction module, a damage recognition module, a damage assessment map module, and an AI damage assessment module.

[0040] The interaction module is used to obtain the first interaction data before the damage to the damaged object and the second interaction data after the damage.

[0041] The damage identification module is used to analyze the first interactive data to form a first damage assessment information set, and to combine the first damage assessment information set to identify the damage to the second interactive data to form a second damage assessment information set.

[0042] The damage assessment graph module may include a graph generation unit and a difference analysis unit;

[0043] The map generation unit is used to associate the first damage assessment information set with the second damage assessment information set according to preset association rules to obtain a damage assessment association map.

[0044] The difference analysis unit is used to compare the baseline feature information in the first loss assessment information set with the damage description information in the second loss assessment information set based on the loss assessment edge, and generate difference factors.

[0045] The AI ​​damage assessment module may include an indicator generation unit and an AI damage assessment unit;

[0046] The indicator generation unit is used to obtain the preset loss assessment indicators of the loss assessment object, and adjust the preset loss assessment indicators according to the difference factors to obtain the target loss assessment indicators.

[0047] The AI ​​damage assessment unit is used to input target damage assessment indicators into a preset damage assessment model, perform damage assessment, generate damage assessment data, use damage assessment nodes in the damage assessment correlation graph as index items, associate and match index items with damage assessment data to obtain damage assessment results, which may include damage level, damaged range data, damaged component data, and repair costs.

[0048] As can be seen from the above technical solution, this application has the following beneficial effects:

[0049] 1. This application analyzes and compares multi-modal interaction data of the damaged object before and after the damage, forms benchmark feature information and damage description information, and constructs a damage assessment correlation map and difference factors, so that the damage assessment results can quantitatively reflect the state changes of the damaged object before and after the damage, thereby improving the accuracy of damage assessment.

[0050] 2. This application generates target loss assessment indicators by dynamically calibrating preset loss assessment indicators based on difference factors, and inputs them into a preset loss assessment model to generate loss assessment data, thereby automating and data-driven the loss assessment process and improving loss assessment efficiency and processing capacity.

[0051] 3. This application achieves systematic output of information such as damage level, damage range, damaged parts and repair costs by structured matching of damage assessment association map nodes and damage assessment data, supports unified management of multiple types of vehicles and property, and improves the usability and traceability of damage assessment results. Attached Figure Description

[0052] The present application will be further described below with reference to the accompanying drawings.

[0053] Figure 1 A flowchart illustrating an AI-powered intelligent vehicle and property damage assessment method supporting multi-modal interaction, provided for this application;

[0054] Figure 2 A flowchart of another AI-powered intelligent vehicle and property damage assessment method supporting multi-modal interaction provided in this application;

[0055] Figure 3 An example diagram of an AI-powered intelligent vehicle and property damage assessment system that supports multi-modal interaction, provided for this application. Detailed Implementation

[0056] The terms "first," "second," and "third," etc., used in this application specification, claims, and drawings are used to distinguish different objects, not to limit a specific order.

[0057] 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.

[0058] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the related technologies is given first:

[0059] In existing technologies, damage assessment usually relies on manual or semi-automated methods. For motor vehicles, existing methods mainly establish a standardized parts system and damage level system, match the damaged parts of the vehicle with the accessory feature data in the database, and combine image recognition models to locate and evaluate the damaged areas of appearance or structure, thereby determining the damaged parts and repair costs. These methods are usually based on two-dimensional images or sensor data and generate damage assessment results according to human experience or fixed rules.

[0060] Research has found that existing damage assessment systems are mainly geared towards motor vehicles, relying on standardized parts systems and damage level systems to identify damaged areas. They use two-dimensional images or sensor data combined with human experience or fixed rules to complete the damage assessment. These methods can accurately reflect the damage situation on motor vehicles because the structure of motor vehicle parts is relatively standardized and the database is mature. However, when applied to non-motorized vehicles, shared mobility tools, or urban public facilities, existing methods have obvious limitations.

[0061] Because motor vehicles and shared mobility tools have diverse structural types and flexible component combinations, different brands and models of vehicles or equipment vary significantly in appearance, structure, and function. The existing standardized parts system is difficult to cover all objects. Urban public facilities such as streetlights, guardrails, and shared bicycle parking racks suffer from various forms of damage, including both cosmetic damage and functional damage. Existing technologies lack a unified way of expressing components and a damage description framework, making it difficult to uniformly encode or quantitatively characterize the damage to these objects.

[0062] Furthermore, due to the lack of a unified feature system and damage description framework, the identification and assessment of different damage assessment objects still heavily rely on manual judgment. Damage assessors need to observe, compare, and infer based on experience, which leads to a lack of standardization in the damage assessment process and makes it difficult to automate through computational models, resulting in low damage assessment efficiency.

[0063] Example 1

[0064] To address the aforementioned issues, this application proposes an AI-powered intelligent vehicle and property damage assessment method that supports multi-modal interaction. Please refer to [link to relevant documentation]. Figure 1 .

[0065] S101, acquire interactive data and form a damage assessment information set.

[0066] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the relevant terms is given first:

[0067] The interaction data includes first interaction data and second interaction data.

[0068] First-level interactive data refers to relevant information collected through multi-modal interactive methods before the damaged object is damaged. Multi-modal interactive methods include, but are not limited to, image acquisition (still photos, videos), sensor acquisition (GPS positioning, accelerometers, tilt sensors), physical property measurement (dimensions, materials, structural parameters), and historical maintenance or usage records.

[0069] The second interactive data refers to the relevant information collected after damage identification of the specified damaged object, using the aforementioned multi-modal interactive methods. The second interactive data may include images of the damaged component, abnormal sensor data, changes in functional status, and structural deformation parameters.

[0070] The damage assessment information set includes a first damage assessment information set and a second damage assessment information set.

[0071] The first damage assessment information set refers to the collection formed after systematic analysis, feature extraction, and structured organization of the first interactive data. This information set includes benchmark feature information, which is mainly used to describe the complete original state of the damage assessment object. It may include the location of key components, appearance features, functional status, structural layout, and spatial parameters, providing a benchmark for subsequent damage comparison and quantification.

[0072] The second damage assessment information set refers to the collection after analyzing and structuring the second interactive data. It includes damage description information, which describes the state of the damaged object after the damage. This information may include the damaged location, degree of deformation, functional abnormality, structural displacement, and damage range.

[0073] In some possible implementation methods, the first interaction data and the second interaction data are acquired, and the two types of interaction data are respectively structured and feature extracted. The specific processing may include the following technical means.

[0074] Image data processing: Preprocess the acquired static photos or videos (such as noise reduction, color correction, scale normalization), and use computer vision algorithms (such as edge detection, deep learning object detection, key point recognition, segmentation algorithms) to extract the contours, shape features, damaged areas and structural offset parameters of key components.

[0075] Sensor data analysis: Filtering, smoothing, and anomaly detection are performed on data collected by sensors such as GPS, accelerometers, and tilt sensors. Parameters such as position changes, tilt angles, and vibration amplitudes are extracted, and time series analysis is combined to identify abnormal states or damage signals.

[0076] Physical property data processing: Normalize and standardize the dimensions, materials, and structural parameters, and convert the measured values ​​into a unified code or feature vector to facilitate fusion with other pattern data.

[0077] Historical maintenance and usage record analysis: Natural language processing, keyword extraction, and structured encoding are performed on unstructured text information such as historical maintenance records, usage frequency, and fault logs to form comparable status features.

[0078] By fusing images, sensor data, physical attributes, and historical data in multiple modes, and generating a loss assessment information set in a unified format through feature vector concatenation, weighted combination, or dimensionality reduction, a loss assessment information set can be generated.

[0079] The first damage assessment information set obtained through the above methods serves as a reference for the baseline state, while the second damage assessment information set reflects the damage state, providing basic data support for subsequent damage assessment correlation map generation, dynamic indicator calibration, and AI damage assessment model input. The multi-mode interactive acquisition method can cover different data types such as images, sensors, and physical measurements, improving information integrity and recognition accuracy, and is applicable to different types of damage assessment objects (such as non-motorized vehicles, electric bicycles, and urban public facilities).

[0080] For example, in a non-motorized vehicle damage assessment scenario, the operator can use a mobile terminal to take panoramic photos of the damage assessment object, collect static photos and videos (image acquisition), and at the same time obtain GPS positioning, vehicle tilt angle and acceleration changes through on-board or external IoT sensors (sensor acquisition), measure wheel diameter, vehicle length and material information (physical property measurement), and read historical usage records and maintenance logs (historical maintenance records).

[0081] The system performs denoising, color correction, and scale normalization on the acquired image data. It then uses deep learning object detection and key point recognition algorithms to extract the contours, positions, and shape features of key components such as handlebars, wheels, and frames. For sensor data, the system performs filtering and smoothing, extracting positional offsets, tilt angle changes, and vibration amplitudes, and combines this with time series analysis to identify signs of anomalies or damage. Physical attribute data is standardized and encoded to form a unified feature vector. Historical maintenance records are transformed into structured status features through natural language processing and keyword extraction.

[0082] By fusing the aforementioned data from various sources and employing feature vector concatenation or weighted combination, a first damage assessment information set is generated. This set describes the complete state of the object before damage. After a collision or damage occurs, the multi-mode data acquisition and processing steps are repeated to generate a second damage assessment information set, recording information such as damaged components, degree of deformation, functional abnormalities, and structural offsets. The generated first and second damage assessment information sets provide the data foundation for subsequent damage assessment correlation mapping, difference factor calculation, and AI damage assessment model input, enabling automated evaluation of different types of damage assessment objects.

[0083] S102, Based on the preset association rules, construct a loss assessment association map based on the first and second loss assessment information sets.

[0084] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the relevant terms is given first:

[0085] A damage assessment correlation graph is a graph structure composed of nodes and edges formed by mapping and associating baseline feature information from a first damage assessment information set with damage description information from a second damage assessment information set. It is used to represent the damage status and interrelationships of various components of the damage assessment object. In this graph, damage assessment nodes represent baseline feature information or damage description information, and damage assessment edges represent the correlation relationships between nodes, including information such as category, structure, and weight.

[0086] Preset association rules refer to the rules used to establish the correspondence between the first loss assessment information set and the second loss assessment information set. These rules include category matching rules, component association rules, and weight allocation rules. Category matching rules are used to match similar components or features; component association rules are used to establish associations between nodes based on spatial location, structural relationships, or functional relationships; and weight allocation rules are used to assign different weights to associated nodes according to risk impact, structural importance, and cost sensitivity.

[0087] S201, Establish the correspondence between baseline feature information and damage description information.

[0088] In some possible implementation methods, this correspondence is established through multi-dimensional information fusion. Based on component or feature categories, the baseline feature information in the first damage assessment information set is initially matched with the damage description information in the second damage assessment information set to ensure that components of the same or similar categories can be correctly paired. For example, in a non-motorized vehicle damage assessment scenario, the system first marks the baseline nodes of key components such as handlebars, wheels, and frames in the first damage assessment information set, and performs preliminary category matching with the corresponding damaged nodes in the second damage assessment information set; for example, mapping the handlebar baseline node to the damaged handlebar node, and the wheel baseline node to the damaged wheel node.

[0089] By combining the spatial layout or three-dimensional structural information of the damaged object, the spatial position of the initially matched nodes is corrected. For example, the relationship between the front and rear wheel spacing and the handlebar position is corrected to ensure that the relative positions of the baseline feature nodes and the damage description nodes are consistent in the overall structure, thereby eliminating possible ambiguities in category matching.

[0090] The correspondence between components is verified and adjusted by analyzing their structural connections and functional dependencies, generating component association records. For example, the accuracy of the correspondence of a damaged handlebar node is determined by the directional control relationship between the handlebars and the front wheel, while excluding other damaged nodes unrelated to that function.

[0091] Each pair of corresponding nodes is assigned a weight to reflect its importance or risk impact in the overall loss assessment result. This weight can be allocated based on the structural importance, functional criticality, and maintenance cost sensitivity of the component, so as to adjust the impact on the loss assessment indicators in the subsequent difference factor calculation.

[0092] By integrating category matching, spatial location association, structural function verification, and weight information, a structured association record is generated, including category matching record, component association record, and weight association record. This forms a complete correspondence between baseline feature information and damage description information, providing basic data support for subsequent damage assessment node generation and map construction.

[0093] S202, integrate related records and generate loss assessment nodes.

[0094] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the relevant terms is given first:

[0095] Damage assessment nodes include baseline feature nodes and damage description nodes.

[0096] A baseline feature node is a damage assessment node constructed from baseline feature information in the first damage assessment information set. Baseline feature nodes describe the original structural state, appearance, component functions, spatial layout, and key attribute parameters of the damage assessment object before the damage occurred. Baseline feature nodes serve as the benchmark reference in the damage assessment correlation map.

[0097] Damage description nodes refer to damage assessment nodes constructed from damage description information in the second damage assessment information set. Damage description nodes are used to represent the damage state, deformation displacement, functional failure, material fracture degree, and damage range of the damage assessment object after the damage occurs.

[0098] In some possible implementation methods, based on a pre-defined graph structure, category matching records, component association records, and weight association records are uniformly converted into a programmable data structure to generate damage assessment nodes. Each node object contains a node ID, node type, component identifier, status attribute, spatial location parameter, and weight information. Feature vector concatenation, weighted combination, or normalization are used to integrate category matching, spatial location correction, functional dependency, and weight allocation information into each node object, achieving a comprehensive representation of the node. Uniqueness is determined by node ID, component category, spatial location, and status attribute. Duplicate or highly similar nodes are merged using hash mapping or an index table to avoid redundant nodes and ensure graph clarity and computational efficiency. Node entities are established in a graph database or graph structure data, and the type of damage assessment node (baseline feature node or damage description node) is labeled according to component category and status attribute. Weight information, functional importance, and structural location parameters are appended as node attributes, providing direct references for subsequent damage assessment edge construction and difference factor calculation. Based on the rule engine or constraints (such as spatial continuity and functional dependency constraints), the generated node set is checked for consistency, and abnormal or mismatched nodes are corrected to ensure that the damage assessment nodes can accurately reflect the actual component status of the damage assessment object and its related relationships.

[0099] For example, in a non-motorized vehicle damage assessment scenario, the system creates graph database node objects for the baseline handlebar node, damaged handlebar node, and baseline wheel node, respectively. Each node records its category, status, location coordinates, and functional importance weight. Category matching information and spatial location correction results are integrated into the node attributes, and the dependencies between nodes are used to construct subsequent damage assessment edges. Duplicate or redundant nodes are merged based on their unique node IDs and spatial locations, forming a structured set of damage assessment nodes, providing a foundation for generating the damage assessment association graph.

[0100] S203, construct the loss assessment edge to form a complete loss assessment correlation map.

[0101] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the relevant terms is given first:

[0102] The damage assessment edge includes structurally related edges, functionally related edges, damaged corresponding edges, and impact edges.

[0103] Structural association edges refer to the damage assessment edges used to represent the structural connection relationships between different components of the damage assessment object. Structural association edges can describe the physical connection methods, connection strength levels, spatial adjacency relationships, and structural topological features between components, and are used to reflect the overall structural composition of the damage assessment object in its original state.

[0104] Functional correlation edges refer to damage assessment edges used to describe the functional dependencies between different components. Functional correlation edges can represent the action transmission relationship, control dependency relationship or energy transmission path between components, and are used to reflect the functional logic chain of the damage assessment object, providing a basis for identifying functional damage.

[0105] Damage correspondence edge refers to the damage assessment edge used to match the baseline feature information in the first damage assessment information set with the damage description information in the second damage assessment information set. Damage correspondence edge can describe the degree of damage, deformation offset, damage type and damage propagation direction, and is an important data source for the calculation of difference factor.

[0106] The impact margin refers to the damage assessment margin used to represent the degree of impact of component damage on the overall damage assessment result. The impact margin includes structural importance weight, functional criticality weight, maintenance cost sensitivity weight, etc., which are used to reflect the impact level of different components in damage assessment analysis and provide a basis for dynamic adjustment of assessment indicators.

[0107] In some possible implementation methods, structural damage assessment edges are established between benchmark feature nodes using the structural layout data, three-dimensional spatial relationship data, and component connection diagram of the damage assessment object in the first damage assessment information set. The connection direction and strength between components are automatically determined using a structural adjacency list, topological matrix, or spatial connectivity judgment algorithm, thereby forming structural association edges that reflect the overall structural constraints.

[0108] Based on the functional relationships of the components of the damage assessment object (such as control relationships, load-bearing relationships, and driving relationships), and using preset functional dependency rules, it is determined whether there are functional interactions or linkages between nodes. For node pairs with critical functional dependencies, functional association edges are generated, and the functional dependency level, scope of effect, and probability of functional reduction are written as functional attributes into the edge attributes.

[0109] The correspondence between baseline feature nodes and damage description nodes established in step S201 is used to connect the corresponding nodes in the form of damage mapping edges. The damage mapping edges contain parameters such as difference type attributes (deformation difference, fracture difference, functional abnormality difference, spatial offset difference), damage probability, and estimated damage level, providing an intuitive path for the calculation of difference factors.

[0110] For the damage assessment nodes assigned weights by S201 and S202, the node weight, structural importance weight, cost sensitivity weight, etc., are comprehensively calculated into edge weights, and an influence edge is generated through a linear weighted model or a normalized risk assessment model. This type of edge can be used for importance ranking and priority repair determination in the damage assessment model.

[0111] The various types of loss assessment edges mentioned above are formatted to ensure a consistent representation of all edges in the loss assessment correlation graph, facilitating subsequent difference factor calculation, graph inference, and model computation. A rule-based validator or... Figure 1 The consistency detection algorithm performs a validity check on the generated edge set, such as avoiding erroneous edges between unconnected components, avoiding duplicate edges, and avoiding cyclical erroneous edges. Edges that do not conform to structural or functional logic are filtered or adjusted. All structurally related edges, functionally related edges, damaged mapping edges, and impact edges are written into a graph database or graph structure cache to form a complete damage assessment correlation graph structure. The system can generate a visual graph based on the graph visualization component for subsequent difference factor calculation path analysis and damage assessment model feature extraction.

[0112] S103, based on the loss determination edge, forms a difference factor.

[0113] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the relevant terms is given first:

[0114] A difference factor is a quantitative indicator used to measure the degree of difference between two data segments, two types of structural parameters, or two state characteristics. Difference factors are typically constructed based on the comparative relationship of multi-source parameters. By comprehensively characterizing the numerical offset, morphological difference, and temporal variation of input parameters, the degree of difference is converted into a quantitative value that can be processed by computational models. The value range of the difference factor can be normalized or scaled according to the needs of a specific model, enabling it to serve as a core input parameter for subsequent classification, aggregation, risk assessment, or loss determination decisions. In this example, the difference factor is mainly used to distinguish the degree of difference in key features of a target object under different states, providing a basis for deriving quantitative judgment results.

[0115] In some possible implementations, structural parameters, such as length, angle, planar profile, curvature, and geometric deformation vector, are extracted from the baseline feature node and the damage description node based on the structural association attributes contained in the damage assessment edge. These structural parameters are then differentially calculated to obtain structural difference quantities, such as length change, tilt offset, or the magnitude of the three-dimensional deformation vector, to characterize the degree of geometric offset of the component before and after damage.

[0116] Based on the spatial location attributes in the damage assessment edge, spatial coordinates, attitude parameters or three-dimensional positioning information are obtained from the damage assessment node. Spatial location difference index is obtained through spatial distance calculation, attitude difference calculation or point cloud overlap rate calculation, which reflects the overall offset, distortion or positional abnormality of the component.

[0117] Based on the damage attributes in the damage assessment edge, appearance feature parameters such as crack length, dent depth, surface damage area, and wear level are extracted from the damage description node. These features are then compared with the healthy appearance parameters in the baseline feature node. Appearance difference indices are generated using methods such as feature similarity calculation, texture residual analysis, or depth feature difference measurement to describe the severity of visible damage.

[0118] Based on the functional association attributes in the damage assessment edge, functional status parameters such as rotational resistance, braking response, electrical conductivity or load capacity are extracted from the baseline feature node and the damage description node. Functional difference indicators are generated by ratio calculation, threshold detection or status level mapping to reflect whether the component function has degraded or failed.

[0119] Using a pre-defined difference mapping model, the aforementioned multiple difference indicators are mapped to single or hierarchical difference factors. The difference mapping model can include a threshold segmentation mapping model, a continuous numerical normalization model, or a hybrid hierarchical mapping model. Specifically, the threshold segmentation mapping model maps the numerical range of the difference indicator vector to the corresponding hierarchical factor; the continuous numerical normalization model compresses or expands the difference indicator to a standardized range (e.g., the 0-1 range) based on the magnitude changes of the difference indicator in different dimensions; the hybrid hierarchical mapping model aggregates the difference trends in different dimensions, converting the difference indicator into a hybrid difference factor that combines hierarchical expression and continuous measurement characteristics. Through this mapping process, the difference indicator vector is transformed into a difference factor with a unified expression form, achieving a standardized expression from multi-dimensional parameters to a single quantitative result.

[0120] S104, based on the difference factor calibration evaluation index, input the evaluation index into the preset loss setting model to generate loss setting data.

[0121] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the relevant terms is given first:

[0122] Pre-defined damage assessment indicators are a set of benchmark evaluation parameters established before damage assessment begins, based on industry experience, historical damage data, or standardized assessment systems. They typically include basic evaluation elements such as initial damage level, initial damage extent, initial component identification, and initial repair costs, representing the default state of the damage object before differential quantification. They serve as the input baseline for subsequent differential factor calibration processes.

[0123] Target damage assessment indicators are a set of targeted evaluation parameters generated after the application of difference factors and the calibration, correction, or reconstruction of preset damage assessment indicators. The set includes the target damage level, target damage range, target component identification, and target repair cost, reflecting the true evaluation results of the damage assessment object under actual damage conditions. Target damage assessment indicators can be directly input into preset damage assessment models to drive the model to generate the final damage assessment data.

[0124] A pre-defined damage assessment model is a model structure used to receive target damage assessment indicators and perform damage assessment inference, damage assessment calculation, or damage assessment classification. Pre-defined damage assessment models can include rule-based damage assessment calculation models, machine learning-based damage assessment prediction models, or deep learning-based feature fusion models. The pre-defined damage assessment model can generate damage assessment data based on input parameters such as target damage level, target damaged area, target component identification, and target repair cost. The damage assessment data contains key quantitative information representing the final damage assessment result.

[0125] By acquiring the pre-set damage assessment indicators corresponding to the damage assessment object and combining them with the difference factors generated in the previous steps, the pre-set damage assessment indicators are dynamically calibrated to form target damage assessment indicators that can accurately reflect the true damage state of the damage assessment object. These indicators include target damage level, target damage range, target component identification, and target repair cost, which are used as inputs to the pre-set damage assessment model to generate the final damage assessment data.

[0126] In some possible ways, obtaining pre-set damage assessment indicators can be achieved through the following technical means. Based on the category of the damaged object (e.g., non-motorized vehicles, electric bicycles, urban public facilities, etc.), relevant indicator templates can be retrieved from a pre-set damage assessment indicator database or standardized evaluation system. These templates include information such as the baseline damage level, baseline damage range, component identification, and reference repair costs. Furthermore, by combining the historical maintenance records, usage conditions, and previous damage cases of the damaged object, preliminary pre-set damage assessment indicators can be dynamically adjusted or generated to provide a more realistic baseline evaluation.

[0127] When calibrating using difference factors, these factors are used as calibration weights or correction factors to adjust various indicators in the pre-defined damage assessment index. For example, based on difference factors, damage levels can be adjusted by adding or subtracting grades or numerical mapping corrections; the damaged area can be adjusted by area or volume ratio adjustments; damaged parts can be added, removed, or their markings updated; and repair costs can be weighted or corrected. In practice, methods such as linear weighting, weighted normalization, or function mapping can be used to integrate multidimensional difference factors and map them to adjustment coefficients for various damage assessment indicators, thereby generating the target damage assessment index.

[0128] After the target damage assessment indicators are input into the pre-defined damage assessment model, the model integrates and calculates the data based on the input parameters to generate damage assessment data. The damage assessment model can use methods such as rule calculation, machine learning prediction, or deep learning inference to perform fusion analysis on the target damage level, target damaged range, target component identification, and target repair cost, and output damage assessment data, including quantitative results such as the final damage level, damaged range data, damaged component information, and corresponding repair cost data.

[0129] S105 outputs the damage assessment results based on the map and damage assessment data.

[0130] In some possible implementation methods, each loss assessment node in the constructed loss assessment correlation map is used as an index item to retrieve the corresponding loss assessment data.

[0131] The specific operations include: traversing all damage assessment nodes in the damage assessment correlation graph, obtaining the target damage assessment indicators corresponding to each node and the damage assessment data generated through a pre-defined damage assessment model; associating and matching each damage assessment node with its corresponding damage assessment data to form a damage assessment entry for a single component or feature, including information such as damage level, damaged area, damaged component identification, and repair costs; and integrating all damage assessment entries according to a pre-defined output format (such as a structured table or database records) to generate a structured damage assessment result.

[0132] This embodiment collects relevant data on the damage assessment object using a multi-modal interaction method. The collected data is structured into a first damage assessment information set and a second damage assessment information set. A damage assessment association graph is then constructed based on preset association rules, connecting damage assessment nodes and edges. Structural differences, spatial differences, appearance differences, and functional differences are extracted from the damage assessment edges to generate difference factors. Based on these difference factors, preset damage assessment indicators are dynamically calibrated, forming target damage assessment indicators that are input into a preset damage assessment model to automatically generate damage assessment data. This achieves a unified representation of component structures and standardized damage descriptions for different types of damage assessment objects, improving damage assessment efficiency and cross-object scalability.

[0133] Example 2

[0134] This application relates to an AI-powered intelligent vehicle and property damage assessment system that supports multi-modal interaction. Please refer to [link / reference needed]. Figure 3 The system modules correspond to the specific implementation steps in Embodiment 1, including an interaction module, a damage identification module, a damage assessment map module, and an AI damage assessment module.

[0135] In some possible implementations, the interaction module acquires first interaction data of the damaged object before the damage and second interaction data after the damage. The first interaction data may include image acquisition (still photographs, videos), sensor data (GPS, accelerometer, tilt sensor), physical property measurements (dimensions, materials, structural parameters), and historical usage or maintenance records. The second interaction data is used to record the state of the damaged component, including images, functional anomaly data, and structural deformation information.

[0136] The damage identification module analyzes and extracts features from the first interactive data to form a first damage assessment information set. It then combines this first damage assessment information set to identify damage to the second interactive data, generating a second damage assessment information set. This module achieves accurate characterization of the baseline state and the damaged state through image processing, sensor data parsing, physical property analysis, and historical data structuring.

[0137] The functions of the two modules mentioned above correspond to step S101 in Embodiment 1.

[0138] The damage assessment map module includes a map generation unit and a difference analysis unit, corresponding to steps S102 and S103 in Example 1.

[0139] The map generation unit generates a damage assessment association map by associating the first damage assessment information set with the second damage assessment information set based on preset association rules.

[0140] The difference analysis unit compares the structural, spatial, appearance, and functional attributes of the damage assessment edge, calculates multidimensional difference indicators, and generates a uniformly expressed difference factor to quantify the changing characteristics of the damage assessment object under different states.

[0141] The AI ​​damage assessment module includes an indicator generation unit and an AI damage assessment unit, corresponding to steps S104 and S105 in Embodiment 1.

[0142] The indicator generation unit obtains the preset loss assessment indicators of the loss assessment object and performs dynamic calibration based on the difference factor to obtain the target loss assessment indicator.

[0143] The AI ​​damage assessment unit inputs the target damage assessment indicators into the preset damage assessment model, generates damage assessment data, and uses the damage assessment nodes in the damage assessment correlation graph as index items to correlate and match with the damage assessment data, outputting the final damage assessment results, including damage level, damage range, damaged component information, and repair costs.

[0144] Through the above system design, this embodiment realizes a unified representation and difference quantification of the state of the damage assessment object before and after the damage. It uses the difference factor to automatically calibrate the preset evaluation index and combines it with the AI ​​damage assessment model to generate standardized damage assessment results, thereby improving damage assessment efficiency and cross-object adaptability.

[0145] The foregoing has shown and described the basic principles, main features, and advantages of this application. Those skilled in the art should understand that this application is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this application. Various changes and modifications can be made to this application without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of this application as claimed. The scope of protection of this application is defined by the appended claims and their equivalents.

Claims

1. An AI-powered intelligent vehicle and property damage assessment method supporting multi-mode interaction, characterized in that: The method includes: First interaction data of the damaged object is obtained before the damage is assessed. The first interaction data is analyzed to form a first damage assessment information set, which includes benchmark feature information. The second interactive data after the damage to the damaged object is obtained, and the damage is identified by combining the second interactive data with the first damage assessment information set to form a second damage assessment information set, which includes damage description information. According to preset association rules, the first damage assessment information set and the second damage assessment information set are associated to obtain a damage assessment association graph, which includes damage assessment nodes and damage assessment edges; Based on the damage assessment edge, the baseline feature information in the first damage assessment information set is compared with the damage description information in the second damage assessment information set to generate a difference factor; Obtain the preset loss assessment index of the loss assessment object, adjust the preset loss assessment index according to the difference factor to obtain the target loss assessment index, input the target loss assessment index into the preset loss assessment model, perform loss assessment, and generate loss assessment data; The damage assessment nodes in the damage assessment correlation map are used as index items. The index items are matched with the damage assessment data to obtain the damage assessment results. The damage assessment results include damage level, damage range data, damaged component data, and repair costs.

2. The method according to claim 1, characterized in that, The step of associating the first damage assessment information set with the second damage assessment information set according to a preset association rule to obtain a damage assessment association map includes: According to the preset association rules, the benchmark feature information in the first damage assessment information set is matched with the damage description information in the second damage assessment information set item by item to establish corresponding association records; The associated records are integrated, and a loss assessment association map is generated according to the preset map structure.

3. The method according to claim 2, characterized in that, The preset association rules include category matching rules, component association rules, and weight allocation rules. The step of establishing corresponding association records by matching the baseline feature information in the first damage assessment information set with the damage description information in the second damage assessment information set item by item according to the preset association rules includes: Based on category matching rules, the baseline feature information and the damage description information are combined according to category to generate category matching records; Based on component association rules, the baseline feature information and the damage description information are associated and matched according to the component structural relationship, component spatial position relationship and component functional relationship to generate component association records; Based on the weighting rules, the baseline feature information and the damage description information are weighted and combined according to the risk impact weight, structural importance weight, and cost sensitivity weight to generate a weighted association record.

4. The method according to claim 3, characterized in that, The step of integrating the associated records and generating a loss assessment association map based on a preset map structure includes: Based on the preset map structure, multiple damage assessment nodes are constructed for the benchmark feature information and damage description information in the category matching record, the component association record, and the weight association record, respectively. Construct corresponding loss assessment edges between the loss assessment nodes to generate a loss assessment association graph.

5. The method according to claim 1, characterized in that, Based on the damage assessment edge, the benchmark feature information in the first damage assessment information set is compared with the damage description information in the second damage assessment information set to generate a difference factor, including: Traverse the loss assessment edges in the loss assessment association graph to obtain the loss assessment nodes connected by the loss assessment edges; The baseline feature information and the damage description information in the damage assessment node are compared item by item to generate a comparison result; Based on the comparison results, numerical or graded difference indicators are generated to form difference factors, which are used to represent the degree of change of the state of the damaged object before and after the damage.

6. The method according to claim 1, characterized in that, The process of obtaining the preset loss assessment indicators for the loss assessment object, and adjusting the preset loss assessment indicators according to the difference factor to obtain the target loss assessment indicators includes: Obtain preset damage assessment indicators, which include damage level, damaged range, component identification, and repair cost; The damage level, the damaged area, the component identification, and the repair cost are adjusted according to the difference factors to obtain the target damage assessment index, which includes the target damage level, the target damaged area, the target component identification, and the target repair cost.

7. The method according to claim 6, characterized in that, The step of inputting the target loss assessment index into a preset loss assessment model, performing loss assessment, and generating loss assessment data includes: The target damage level, the target damage range, the target component identification, and the target repair cost are used as input parameters and input into a preset damage assessment model. The preset damage assessment model is then used to integrate the input parameters to generate corresponding damage assessment data.

8. The method according to claim 1, characterized in that, The step of using the loss assessment nodes in the loss assessment correlation map as index items, and associating and matching the index items with the loss assessment data to obtain the loss assessment result includes: Using the loss assessment nodes in the loss assessment correlation graph as index items, the corresponding loss assessment data is retrieved according to the index items to obtain the first loss assessment data; The first damage assessment data is associated and matched with the index item to generate a damage assessment entry for the damage assessment object; The damage assessment items are integrated according to a preset format to generate the damage assessment result.

9. The method according to claim 8, characterized in that, After integrating the damage assessment items according to a preset format to obtain the damage assessment result, the process further includes: Trigger a damage assessment correction, which generates a damage assessment report in response to the damage assessor's correction of the damage assessment result.

10. An AI-powered intelligent vehicle and property damage assessment system supporting multi-mode interaction, characterized in that: The system includes: an interaction module, a damage identification module, a damage assessment map module, and an AI damage assessment module; The interaction module is used to acquire first interaction data of the damaged object before the damage and second interaction data of the damaged object after the damage. The damage identification module is used to analyze the first interactive data to form a first damage assessment information set, and combine the first damage assessment information set to identify damage to the second interactive data to form a second damage assessment information set. The damage assessment map module includes a map generation unit and a difference analysis unit; The graph generation unit is used to associate the first damage assessment information set and the second damage assessment information set according to a preset association rule to obtain a damage assessment association graph, wherein the damage assessment association graph includes damage assessment nodes and damage assessment edges; The difference analysis unit is used to compare the baseline feature information in the first loss assessment information set with the damage description information in the second loss assessment information set based on the loss assessment edge, and generate a difference factor. The AI ​​damage assessment module includes an indicator generation unit and an AI damage assessment unit; The indicator generation unit is used to obtain the preset loss assessment indicators of the loss assessment object, and adjust the preset loss assessment indicators according to the difference factor to obtain the target loss assessment indicators. The AI ​​damage assessment unit is used to input the target damage assessment indicators into a preset damage assessment model, perform damage assessment, generate damage assessment data, use the damage assessment nodes in the damage assessment correlation graph as index items, and associate and match the index items with the damage assessment data to obtain damage assessment results. The damage assessment results include damage level, damage range data, damaged component data, and repair costs.

Images