Pedestrian head collision injury prediction model construction method and system, and pedestrian head collision injury prediction method and system

Abstract

The invention discloses a pedestrian head collision injury prediction model construction and prediction method and system, and the method comprises the steps: firstly obtaining the simulation data of a plurality of groups of pedestrian head collision simulation tests, and extracting the HIC value and collision point data of a collision point, and the point cloud data of a vehicle body structure within a certain range around the collision point from the simulation data; and constructing a data set by taking the HIC value, the collision point data and the point cloud data corresponding to each group of simulation tests as samples. And expanding the sample size of the data set through a data enhancement method based on linear interpolation. And then fusing the point cloud feature extraction model and the uncertainty quantification model to construct a deep learning fusion model, and training the deep learning fusion model through a data set to obtain an HIC prediction model. And then extracting collision point data and point cloud data of the collision point from the to-be-predicted simulation result data. And finally, inputting the data into an HIC prediction model, and obtaining an HIC prediction value of the simulation test and the uncertainty degree of the HIC prediction value through the HIC prediction model.

Description

Technical Field

[0001] The present invention relates to the technical field of computer-aided design of automobiles, and in particular to the construction of a pedestrian head collision injury prediction model, a prediction method and a system. Background Art

[0002] According to the World Health Organization's "Pedestrian Safety: A Road Safety Handbook for Policymakers and Practitioners," pedestrians account for an average of approximately 23% of traffic fatalities worldwide. The head, due to its biomechanical fragility and lack of effective impact cushioning, is the most commonly injured area. To reduce the risk of head injuries, several countries and regions have established mandatory pedestrian protection standards and regulations, such as China's national standard GB 24550-2024 and the Economic Commission for Europe regulation ECE R127. Furthermore, new car evaluation systems such as the European New Car Assessment Program (Euro NCAP), the China New Car Assessment Program (C-NCAP), and the China Insurance Automobile Safety Index (C-IASI) all include pedestrian protection testing based on the Head Injury Criteria (HIC) assessment.

[0003] To meet these standards, regulations, and evaluation procedures, global automakers have incorporated pedestrian head protection performance development into their vehicle passive safety performance development systems. In the early stages of vehicle front-end structural development, HIC values ​​are primarily obtained through Computer-Aided Engineering (CAE). Due to the complex front-end structure and the need for CAE simulation analysis of hundreds of impact points, this evaluation process typically consumes thousands of CPU core hours, making it difficult to rapidly verify the pedestrian protection performance of the vehicle's front-end structure.

[0004] In order to meet this challenge, relevant technologies have made certain progress. For example, in the patent application with publication number CN117272511A previously applied for by the applicant, a method for predicting the HIC results of pedestrian protection head shape based on deep learning was proposed. By extracting feature data and building a database, and using BP neural network and random forest algorithm for training, the rapid prediction of HIC value was achieved, which significantly improved the analysis efficiency. However, this technical solution cannot accurately capture the three-dimensional structural information around the collision point, and the prediction accuracy and applicability for complex collision scenarios are low. It cannot be applied to the rapid prediction of HIC for different vehicle models and collision scenarios. In addition, the previous method does not have the ability to quantify uncertainty, and cannot provide an objective quantitative and risk-controllable decision-making basis for the pedestrian protection passive safety performance development system based on AI prediction, and it is difficult to meet the needs of engineering applications. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention proposes a pedestrian head impact injury prediction model, prediction method, and system that can meet the HIC value prediction requirements of different vehicle models. The specific technical solution is as follows:

[0006] In a first aspect, a method for constructing a pedestrian head impact injury prediction model is provided. In a first possible implementation of the first aspect, the method includes:

[0007] Acquire multiple sets of simulated data from pedestrian head impact tests to construct a data set. The simulated data includes the HIC value and impact point data of each impact point, as well as the point cloud data of the vehicle body within a certain range around the impact point.

[0008] A deep learning fusion model is constructed based on the point cloud feature extraction model and the uncertainty quantification model, and the deep learning fusion model is trained using the data set to obtain a HIC prediction model.

[0009] In conjunction with the first implementable manner of the first aspect, in a second implementable manner of the first aspect, obtaining simulation data of a pedestrian head impact simulation test includes:

[0010] Calculating the HIC value ratio between the collision point and the nearest collision points, and comparing the calculated HIC value ratio with a preset threshold;

[0011] In response to the HIC value ratio not exceeding a preset threshold, a linear interpolation data enhancement algorithm is used to calculate the HIC value corresponding to each target point between the collision point and the other collision points.

[0012] In conjunction with the first implementable manner of the first aspect, in a third implementable manner of the first aspect, obtaining simulation data of a pedestrian head impact simulation test includes:

[0013] Taking the collision point as the center of the sphere, the point cloud data of the vehicle body structure around the collision point is extracted according to the set radius.

[0014] In combination with the first feasible method of the first aspect, in the fourth feasible method of the first aspect, both the point cloud data and the collision point data include three-dimensional coordinate information of the finite element mesh nodes, as well as the corresponding component names, material properties and material thicknesses.

[0015] In combination with the first implementable manner of the first aspect, in a fifth implementable manner of the first aspect, the point cloud feature extraction model uses a hierarchical feature extraction network to output the HIC prediction value.

[0016] In combination with the fifth implementable manner of the first aspect, in a sixth implementable manner of the first aspect, the uncertainty quantification model uses multiple shallow regression models for parallel prediction based on the intermediate layer features output by the hierarchical feature extraction network;

[0017] The mean and standard deviation of the prediction results of all shallow regression models were calculated, and the standard deviation was compared with the preset threshold. The degree of uncertainty of the HIC prediction value was determined based on the comparison results.

[0018] In a second aspect, a method for predicting pedestrian head impact injuries is provided. In a first possible implementation of the second aspect, the method includes:

[0019] A HIC prediction model is constructed using the pedestrian head impact injury prediction model construction method described in any one of the first to sixth possible implementations of the first aspect;

[0020] Obtain the collision point data of the collision point in the pedestrian protection head collision simulation test, as well as the point cloud data within a certain range around the collision point;

[0021] According to the collision point data and point cloud data, the HIC prediction value and its corresponding uncertainty degree are obtained through the trained HIC prediction model.

[0022] In conjunction with the first possible implementation of the second aspect, in the second possible implementation of the second aspect, the method further includes:

[0023] The trained deep learning fusion model is used to obtain the HIC prediction value and the corresponding uncertainty degree of the prediction value. Finally, a head shape score map is constructed based on these HIC prediction values ​​and their uncertainty degrees.

[0024] In conjunction with the first implementable manner of the second aspect, in a third implementable manner of the second aspect, constructing the head shape score map includes:

[0025] Obtain the HIC predicted value and HIC true value corresponding to each collision point in the head shape collision area, and determine the two-dimensional coordinates of the collision point based on the point information of each collision point;

[0026] According to the two-dimensional plane coordinates, the HIC predicted value, the HIC true value, and the error interval corresponding to each collision point are respectively filled into the zero matrix corresponding to the head shape collision area to form corresponding predicted value score spectrum matrix, true value score spectrum matrix, and error interval spectrum matrix respectively;

[0027] Based on the predicted value score spectrum matrix, the true value score spectrum matrix and the error interval spectrum matrix respectively, the corresponding predicted value score spectrum, true value score spectrum and error interval spectrum are drawn.

[0028] In a third aspect, a pedestrian head collision injury prediction system is provided, comprising:

[0029] a model building module configured to build a HIC prediction model using the pedestrian head impact injury prediction model building method described in any one of the first to fifth possible implementations of the first aspect;

[0030] a data acquisition module configured to acquire collision point data of a collision point in a pedestrian protection head collision working condition simulation test, and point cloud data within a certain range around the collision point;

[0031] The indicator prediction module is configured to obtain the HIC prediction value and its corresponding uncertainty degree through the trained HIC prediction model according to the collision point data and the point cloud data.

[0032] Beneficial Effects: The pedestrian head impact injury prediction model, prediction method, and system of the present invention enable more accurate extraction of three-dimensional geometric structural information around the impact point through the constructed pedestrian head impact injury prediction model, significantly improving the prediction model's adaptability and prediction accuracy for complex collision scenarios. Furthermore, the introduction of an uncertainty quantification mechanism enables the model to not only predict HIC values ​​but also assess the reliability of the prediction results, providing risk-controlled decision support for design optimization and effectively reducing the design risks associated with prediction uncertainty. This makes the prediction model more applicable to different vehicle models and collision scenarios, better meeting the diverse needs of the automotive industry and further enhancing the practicality and versatility of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the specific embodiments of the present invention, the following briefly introduces the drawings required for use in the specific embodiments. In all the drawings, each element or part is not necessarily drawn according to the actual scale.

[0034] Figure 1 A flowchart of a method for constructing a human head impact injury prediction model provided by one embodiment of the present invention;

[0035] Figure 2 A flowchart of a method for predicting pedestrian head impact injuries provided by one embodiment of the present invention;

[0036] Figure 3 To construct the prediction value score map;

[0037] Figure 4 To construct a true value score map;

[0038] Figure 5 is the constructed prediction error interval map;

[0039] Figure 6This is the flow chart of the data enhancement algorithm;

[0040] Figure 7 This is the model structure diagram of the deep learning fusion model;

[0041] Figure 8 Flowchart for uncertainty quantification model to quantify the degree of uncertainty in prediction results;

[0042] Figure 9 Schematic diagram for simulation data extraction. DETAILED DESCRIPTION

[0043] The following embodiments of the technical solution of the present invention will be described in detail with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and are therefore only examples and are not intended to limit the scope of protection of the present invention.

[0044] like Figure 1 The flowchart of the method for constructing a pedestrian head impact injury prediction model is shown, and the method includes:

[0045] Step 1: Acquire multiple sets of simulated data from pedestrian head impact simulation tests to construct a dataset. The simulated data includes the HIC value and impact point data of each impact point, as well as the point cloud data of the vehicle body within a certain range around the impact point.

[0046] Step 2: A deep learning fusion model is constructed based on the point cloud feature extraction model and the uncertainty quantification model, and the deep learning fusion model is trained using the data set to obtain a HIC prediction model.

[0047] Specifically, multiple pedestrian head impact simulation tests can be conducted to obtain simulation data for each test. HIC values ​​and collision point data for all collision points can be extracted from the simulation data. This collision point data includes the vehicle body collision point location information and its associated attribute data, as well as point cloud data of the vehicle body structure within a certain range around the collision point. A dataset is constructed using the HIC values, collision point data, and point cloud data corresponding to each simulation test as samples. A deep learning fusion model can then be constructed by fusing the point cloud feature extraction model with the uncertainty quantification model. This deep learning fusion model can then be trained using this constructed dataset to produce a HIC prediction model.

[0048] Through the point cloud feature extraction model in the HIC prediction model, more detailed three-dimensional geometric structure information around the collision point can be accurately extracted, significantly improving the predictive model's adaptability and prediction accuracy for complex collision scenarios, as well as its ability to generalize across vehicle models. Furthermore, through the uncertainty quantification model, the reliability of the prediction results can be quantitatively evaluated, providing risk-controlled decision support for design optimization and effectively reducing the design risks caused by prediction uncertainty. This enables the HIC prediction model to be applicable to different vehicle models and collision scenarios. After introducing the uncertainty quantification mechanism, the HIC prediction model can not only output the prediction value, but also simultaneously give the degree of uncertainty of the prediction result, thereby providing stable and reliable prediction results when facing different vehicle structure and variable collision conditions, effectively enhancing the generalization ability and applicability of the model, and further improving the practicality and versatility of the present invention in automotive safety design.

[0049] A deep learning fusion model is constructed based on the point cloud feature extraction model and uncertainty quantification model, such as Figure 7 As shown in Figure 1, this deep learning fusion model consists of a point cloud feature extraction module and other feature extraction components. The point cloud feature extraction module uses three cascaded feature extraction units (sa1, sa2, and sa3) to extract local and global features, and then outputs global features after performing corresponding operations. Other feature extraction components process key point coordinates, thickness information, component category, material category, and age group to generate corresponding feature vectors. All features are concatenated to form a comprehensive feature vector, which is then input into the MLP to output the predicted HIC value. The model parameters are then updated through backpropagation using loss functions commonly used in regression problems, such as mean squared error.

[0050] In this embodiment, optionally, obtaining simulation data of a pedestrian head impact simulation test includes:

[0051] Calculating the HIC value ratio between the collision point and the nearest collision points, and comparing the calculated HIC value ratio with a preset threshold;

[0052] In response to the HIC value ratio not exceeding a preset threshold, a linear interpolation data enhancement algorithm is used to calculate the HIC value corresponding to each target point between the collision point and the other collision points.

[0053] Specifically, when acquiring simulation data, the HIC value of a head impact point and the HIC value of the nearest other impact point can be used to estimate the HIC values ​​of other impact points between these two impact points with known HIC values. This not only improves computational efficiency but also ensures the accuracy of the results.

[0054] For example, Figure 6As shown, the HIC values ​​of the known collision point A and its nearest collision point B are HIC A and HIC B When calculating the HIC values ​​of other collision points between collision point A and collision point B, the HIC value ratio between collision point A and collision point B can be calculated first. The specific calculation formula is as follows:

[0055]

[0056] The calculated HIC value ratio is compared with the preset threshold. If the HIC value ratio is greater than the preset threshold, it can be determined that the HIC values ​​of other collision points between collision point A and collision point B show nonlinear changes, and the linear interpolation data enhancement algorithm cannot be used to calculate the HIC value of collision point C between collision point A and collision point B. Otherwise, the linear interpolation algorithm can be used to calculate the HIC values ​​of other collision points. The specific calculation formula is as follows:

[0057]

[0058] Among them, d A d B These are the distances between collision points A, B, and C, respectively. Linear interpolation is an effective simplification method. By setting a reasonable threshold, it can significantly improve computational efficiency while ensuring the accuracy of the samples augmented by data augmentation methods. This allows the sample size of the training set to be expanded without additional CAE simulations, helping to shorten vehicle development cycles and provide a scientific basis for vehicle safety design.

[0059] In this embodiment, optionally, obtaining simulation data of a pedestrian head impact simulation test includes:

[0060] Taking the collision point as the center of the sphere, the point cloud data of the vehicle body structure around the collision point is extracted according to the set radius.

[0061] Specifically, if Figure 9 As shown in the figure, when extracting point cloud data, the position coordinates of the collision point are used as the center of the sphere, and the point cloud data of the vehicle body structure in the spherical area around the collision point is extracted according to the set radius, providing richer feature information for HIC value prediction, thereby improving the HIC prediction accuracy and cross-vehicle generalization ability under pedestrian head collision conditions.

[0062] Specifically, the spatial coordinate information of all collision points is first extracted from the simulation model's master control file, and the component number (PID) corresponding to each collision point is recorded. Then, based on the preset collision point coordinates, a radius R is set with each collision point as the center of a sphere, and point cloud data within this radius is intercepted in three-dimensional space. Key component attribute information, including component type, material grade, and thickness, is then manually or using auxiliary tools to be marked. The component attribute table is then indexed based on the PID of each point to obtain information such as the material, type, and thickness of the component to which each point belongs. Finally, the collision point, point cloud coordinates, PID, and matched material properties are organized into an analytical data table for subsequent model training or result interpretation. This effectively extracts geometric and material information about the structural area surrounding the collision point, providing accurate input features for the subsequent point cloud-based HIC prediction model while avoiding full exposure of the entire vehicle structural model, ensuring excellent engineering applicability and data security.

[0063] In this embodiment, both the point cloud data and the impact point data optionally include the 3D coordinates of the finite element mesh nodes, as well as the corresponding component names, material properties, and thicknesses. This data can comprehensively characterize the structural and material properties of the collision area, providing richer input features for the HIC prediction model.

[0064] In this embodiment, optionally, the point cloud feature extraction model uses a hierarchical feature extraction network to output HIC prediction values.

[0065] Specifically, a hierarchical feature extraction network is employed, capable of extracting local and global geometric features step by step through a multi-layered structure, adapting to the sparsity and irregularity of point cloud data. This network demonstrates strong spatial perception in capturing the complex morphology of vehicle body structures, enabling more precise identification of detailed features in the area surrounding the collision point, thereby improving the accuracy and robustness of HIC predictions. This is particularly applicable to collision scenarios involving significant structural variations or large vehicle model differences.

[0066] In this embodiment, optionally, the uncertainty quantification model uses multiple shallow regression models for parallel prediction based on the intermediate layer features output by the hierarchical feature extraction network;

[0067] The uncertainty of the HIC prediction value is determined by calculating the mean and standard deviation of the prediction results of all shallow regression models and comparing the standard deviation with the preset threshold.

[0068] Specifically, if Figure 8As shown in Figure 2, the hierarchical feature extraction network generates multiple intermediate-layer feature vectors during the point cloud feature extraction process. These intermediate representations contain comprehensive information such as local geometry, material characteristics, and overall structural distribution, and have good expressive power. Using them as inputs for uncertainty assessment helps capture potential ambiguity and risk in the prediction process.

[0069] When training an uncertainty quantification model, a pre-trained deep learning fusion model can be used to extract features and corresponding HIC values ​​from the penultimate layer of the hierarchical feature extraction network, and then train multiple shallow models. When a test sample is input, the pre-trained neural network model is loaded, the features and corresponding HIC values ​​from the penultimate layer are extracted, and the pre-trained shallow model is used for inference to obtain multiple prediction results. The prediction results of all shallow models are collected, and the mean and standard deviation are calculated. If the standard deviation is greater than a preset threshold, the sample is marked as high uncertainty; otherwise, it is marked as a reliable prediction. Finally, model training is completed, and prediction results that include uncertainty are obtained, providing a more reliable basis for decision-making in practical applications.

[0070] like Figure 2 The flowchart of the pedestrian head collision injury prediction method shown in FIG. 1 includes:

[0071] Step S1: constructing a HIC prediction model using the above-mentioned pedestrian head impact injury prediction model construction method;

[0072] Step S2: Obtaining collision point data of the collision point in the pedestrian head collision simulation test, as well as point cloud data within a certain range around the collision point;

[0073] Step S3: Obtain an HIC prediction value and its corresponding uncertainty level through a trained HIC prediction model based on the collision point data and the point cloud data.

[0074] Specifically, the HIC prediction model can be constructed using the aforementioned construction method to obtain a trained HIC prediction model. Then, the impact point data and the point cloud data surrounding the impact point can be extracted from the pedestrian head impact simulation test results. Finally, the impact point data and point cloud data can be input into the trained HIC prediction model. The HIC prediction model can then be used to obtain the HIC prediction value corresponding to the simulation test, as well as the corresponding degree of uncertainty of the HIC prediction value.

[0075] In this embodiment, optionally, the following is further included:

[0076] The HIC prediction values ​​and their corresponding uncertainty levels obtained multiple times by the HIC prediction model are obtained, and a head shape score map is constructed according to all the HIC prediction values ​​and their corresponding uncertainty levels.

[0077] Specifically, after obtaining predictions from the HIC prediction model, pedestrian head impact simulations can be repeated and the model can be used to obtain corresponding predictions. A head type score map is then constructed based on the prediction results from multiple simulations, visually presenting the HIC prediction values ​​and uncertainty distribution.

[0078] In this embodiment, optionally, constructing a head shape score map includes:

[0079] Obtain the HIC predicted value and HIC true value corresponding to each collision point in the head shape collision area, and determine the two-dimensional coordinates of the collision point based on the point information of each collision point;

[0080] According to the two-dimensional plane coordinates, the HIC predicted value, the HIC true value, and the error interval corresponding to each collision point are respectively filled into the zero matrix corresponding to the head shape collision area to form corresponding predicted value score spectrum matrix, true value score spectrum matrix, and error interval spectrum matrix respectively;

[0081] Based on the predicted value score spectrum matrix, the true value score spectrum matrix and the error interval spectrum matrix respectively, the corresponding predicted value score spectrum, true value score spectrum and error interval spectrum are drawn.

[0082] Specifically, first, the file containing the head impact test data can be read, the predicted HIC value and the actual HIC value of each test point can be extracted, and the two-dimensional plane coordinates of the impact point can be determined based on the coordinate information of each impact point;

[0083] Then, zero matrices are created to represent the head shape collision area. The predicted HIC value, true HIC value, and error interval (such as standard deviation) of each collision point are filled into the corresponding matrix position according to the coordinates, thereby forming the predicted value score spectrum matrix, the true value score spectrum matrix, and the error interval spectrum matrix. In addition, data is prepared for subsequent color mapping;

[0084] Finally, according to the contents of different matrices, the predicted value score map, the true value score map and the error interval map are drawn respectively, such as Figure 3-5 As shown in the figure, the color of each collision point is determined by its corresponding value, and the specific value is marked at the center of the color block. To enhance readability, the module draws grid lines to clearly distinguish the location of each test point and sets x-axis and y-axis labels to make the coordinates of the map more intuitive. Finally, the module adds titles such as "Prediction Value Score Map," "True Value Score Map," and "Prediction Error Range Map" to clarify the content of each map.

[0085] A pedestrian head collision injury prediction model construction system, comprising:

[0086] A data set construction module is configured to obtain multiple sets of simulated data from pedestrian head impact simulation tests to construct a data set, wherein the simulated data includes the HIC value and impact point data of each impact point, as well as point cloud data of the vehicle body within a certain range around the impact point;

[0087] The model training module is configured to build a deep learning fusion model based on the point cloud feature extraction model and the uncertainty quantification model, and train the deep learning fusion model through the data set to obtain the HIC prediction model.

[0088] Specifically, the construction system includes a dataset construction module and a model training module. The dataset construction module can obtain simulation data from multiple pedestrian head impact simulation tests, and extract the HIC value and impact point data of the collision point, as well as the point cloud data of the vehicle body structure within a certain range around the collision point from the simulation data. The dataset is constructed using the HIC value, impact point data, and point cloud data corresponding to each simulation test as samples. The model training module can fuse the point cloud feature extraction model and the uncertainty quantification model to construct a deep learning fusion model, and then train the deep learning fusion model using the constructed dataset to obtain a HIC prediction model.

[0089] A pedestrian head collision injury prediction system, comprising:

[0090] A model building module is configured to use the above-mentioned pedestrian head impact injury prediction model building method to build a HIC prediction model;

[0091] a data acquisition module configured to acquire collision point data of a collision point in a pedestrian head collision simulation test, and point cloud data within a certain range around the collision point;

[0092] The indicator prediction module is configured to obtain the HIC prediction value and its corresponding uncertainty degree through the trained HIC prediction model according to the collision point data and the point cloud data.

[0093] Specifically, the prediction system includes a model construction module, a data acquisition module, and an indicator prediction module. The model construction module can use the aforementioned construction method to construct a HIC prediction model, resulting in a trained HIC prediction model. The data acquisition module can extract the collision point data and the point cloud data surrounding the collision point from the results of a pedestrian head impact simulation test. The indicator prediction module can input the collision point data and point cloud data into the trained HIC prediction model. Using the HIC prediction model, the HIC prediction value corresponding to the simulation test and the corresponding degree of uncertainty of the HIC prediction value can be obtained.

[0094] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. A method for constructing a pedestrian head impact injury prediction model, characterized in that: include: Acquire multiple sets of simulated data from pedestrian head impact tests to construct a data set. The simulated data includes the HIC value and impact point data of each impact point, as well as the point cloud data of the vehicle body within a certain range around the impact point. A deep learning fusion model is constructed based on the point cloud feature extraction model and the uncertainty quantification model, and the deep learning fusion model is trained using the data set to obtain a HIC prediction model.

2. The method for constructing a pedestrian head impact injury prediction model according to claim 1, characterized in that: Acquire simulation data for pedestrian head impact simulation tests, including: Calculating the HIC value ratio between the collision point and the nearest collision points, and comparing the calculated HIC value ratio with a preset threshold; In response to the HIC value ratio not exceeding a preset threshold, a linear interpolation data enhancement algorithm is used to calculate the HIC value corresponding to each target point between the collision point and the other collision points.

3. The method for constructing a pedestrian head impact injury prediction model according to claim 1, characterized in that: Acquire simulation data for pedestrian head impact simulation tests, including: Taking the collision point as the center of the sphere, the point cloud data of the vehicle body structure around the collision point is extracted according to the set radius.

4. The method for constructing a pedestrian head impact injury prediction model according to claim 1, characterized in that: Both point cloud data and collision point data include the three-dimensional coordinate information of the finite element mesh nodes, as well as the corresponding component names, material properties and material thickness.

5. The method for constructing a pedestrian head impact injury prediction model according to claim 1, characterized in that: The point cloud feature extraction model uses a hierarchical feature extraction network to output HIC prediction values.

6. The method for constructing a pedestrian head impact injury prediction model according to claim 5, characterized in that: The uncertainty quantification model uses multiple shallow regression models for parallel prediction based on the intermediate layer features output by the hierarchical feature extraction network; The mean and standard deviation of the prediction results of all shallow regression models were calculated, and the standard deviation was compared with the preset threshold. The degree of uncertainty of the HIC prediction value was determined based on the comparison results.

7. A method for predicting pedestrian head collision injuries, characterized in that: include: A HIC prediction model is constructed using the pedestrian head impact injury prediction model construction method according to any one of claims 1 to 6; Obtain the collision point data of the pedestrian head collision simulation test, as well as the point cloud data within a certain range around the collision point; According to the collision point data and point cloud data, the HIC prediction value and its corresponding uncertainty degree are obtained through the trained HIC prediction model.

8. The pedestrian head impact injury prediction method according to claim 7, characterized in that: Also includes: The HIC prediction values ​​and their corresponding uncertainty levels obtained multiple times by the HIC prediction model are obtained, and a head shape score map is constructed according to all the HIC prediction values ​​and their corresponding uncertainty levels.

9. The pedestrian head impact injury prediction method according to claim 8, characterized in that: Constructing the head shape score map includes: Obtain the HIC predicted value and HIC true value corresponding to each collision point in the head shape collision area, and determine the two-dimensional coordinates of the collision point based on the point information of each collision point; According to the two-dimensional plane coordinates, the HIC predicted value, the HIC true value, and the error interval corresponding to each collision point are respectively filled into the zero matrix corresponding to the head shape collision area to form corresponding predicted value score spectrum matrix, true value score spectrum matrix, and error interval spectrum matrix respectively; Based on the predicted value score spectrum matrix, the true value score spectrum matrix and the error interval spectrum matrix respectively, the corresponding predicted value score spectrum, true value score spectrum and error interval spectrum are drawn.

10. A pedestrian head collision injury prediction system, characterized in that: include: A model building module configured to build a HIC prediction model using the pedestrian head impact injury prediction model building method according to any one of claims 1 to 6; a data acquisition module configured to acquire collision point data of a collision point in a pedestrian head collision simulation test, and point cloud data within a certain range around the collision point; The indicator prediction module is configured to obtain the HIC prediction value and its corresponding uncertainty degree through the trained HIC prediction model according to the collision point data and the point cloud data.

Images