Pedestrian head injury evaluation method based on high-precision accident reconstruction

Abstract

The application relates to the technical field of traffic accident injury evaluation, and discloses a pedestrian head comprehensive injury evaluation method based on high-precision accident reconstruction, wherein the method comprises the following steps: acquiring accident video information and head injury information of a pedestrian deep accident case; estimating a collision speed based on frame-by-frame analysis and a direct linear transformation method; reconstructing a pedestrian collision kinematic response through multi-body dynamics simulation; loading head collision boundary condition data to a human body finite element head model to obtain a head injury parameter set; screening a candidate injury criterion based on an F1 score; determining an injury threshold value by using a logistic regression method and determining a weight parameter based on injury type statistics; calculating a head comprehensive weighted injury criterion value to obtain a comprehensive injury evaluation result.

Description

Technical Field

[0001] This invention relates to the field of traffic accident injury assessment technology, specifically to a method for comprehensive assessment of pedestrian head injuries based on high-precision accident reconstruction. Background Technology

[0002] In car-pedestrian collisions, a pedestrian's head may collide with the hood, windshield, or ground, among other things. The collision patterns are varied, leading to diverse types of head injuries. The pedestrian's head has a complex anatomical structure and can simultaneously withstand linear and rotational loads during a collision, resulting in complex and severe brain injuries caused by multiple injury types, including skull fractures, diffuse brain injury, and subdural hematoma.

[0003] The field of head injury assessment has developed a variety of injury criteria, such as head injury criteria. Peak angular acceleration is used to assess the risk of skull fracture caused by linear loading, while peak angular acceleration is used to assess the risk of diffuse brain injury caused by rotational loading.

[0004] However, existing injury criteria mainly evaluate single types of head injuries. Each criterion can only reflect the injury risk corresponding to a specific load form. It is difficult to assess the risk of complex severe brain injury caused by the combined effects of linear and rotational loads. It lacks the ability to comprehensively evaluate the concurrent occurrence of multiple injury types, resulting in significant limitations of existing methods in the comprehensive evaluation of pedestrian head injuries. Summary of the Invention

[0005] This invention provides a method for evaluating comprehensive head injuries of pedestrians based on high-precision accident reconstruction, which solves the technical problems of difficulty in accurately assessing the risk of head injuries of pedestrians and the lack of a unified comprehensive injury evaluation standard in related technologies.

[0006] This invention discloses a method for evaluating comprehensive pedestrian head injuries based on high-precision accident reconstruction. The method includes: acquiring accident video information and head injury information from deep pedestrian accident cases to generate an accident case dataset; analyzing and processing the accident video information to estimate the pre-collision velocities of the vehicle and pedestrian, obtaining collision velocity data; inputting the collision velocity data and initial collision condition parameters into multibody dynamics simulation software to reconstruct the pedestrian collision kinematic response, and extracting head collision boundary condition data from the kinematic reconstruction results; loading the head collision boundary condition data into a human finite element head model, performing finite element simulation calculations to obtain a set of head injury parameters; calculating the predictive performance score of each injury criterion based on the head injury parameter set and head injury information from all cases, selecting injury criteria with high predictive performance scores corresponding to different injury types, and generating a set of candidate injury criteria; determining the injury threshold and weight parameters for each injury criterion in the candidate injury criterion set; normalizing the damage parameter values ​​of each candidate injury criterion by dividing them by the corresponding injury threshold, and then weighting and summing them using the weight parameters to obtain a weighted comprehensive head injury criterion value, which serves as the comprehensive head injury evaluation result.

[0007] Furthermore, the accident video information is analyzed and processed to estimate the pre-collision velocities of the vehicles and pedestrians involved in the accident, obtaining collision velocity data. This includes frame-by-frame analysis of the accident video, extracting the position coordinates of the vehicles and pedestrians in consecutive frames before the collision, and calculating a preliminary estimate of the pre-collision velocity based on the ratio of inter-frame displacement to frame rate. A direct linear transformation method is then used to solve for the transformation parameters based on the mapping relationship between the pixel coordinates of the control points at the accident scene and the actual three-dimensional spatial coordinates, converting the pixel coordinates in consecutive frames into actual spatial coordinates, calculating the actual displacement between adjacent frames, and obtaining a precise estimate of the pre-collision velocity. The preliminary estimate and the precise estimate are cross-validated. When the deviation between the two is within a preset threshold range, the average of the two is taken as the collision velocity data; when the deviation exceeds the preset threshold, the precise estimate is used as the collision velocity data.

[0008] Furthermore, in the direct linear transformation method, at least six control points with known three-dimensional spatial coordinates are selected at the accident site. The transformation formula is that the pixel coordinates of each control point are equal to the quotient obtained by dividing the three-dimensional spatial coordinates of the control point by another linear combination of the three-dimensional spatial coordinates of the same set of transformation parameters after linear combination of the three-dimensional spatial coordinates of the control point through eleven transformation parameters.

[0009] Furthermore, the collision velocity data and initial collision condition parameters are input into multibody dynamics simulation software to reconstruct the pedestrian collision kinematic response. Head collision boundary condition data are extracted from the kinematic reconstruction results, including the extraction of initial collision condition parameters based on accident video information. These initial collision condition parameters include the collision angle, collision position, and the point of head-vehicle collision. The collision velocity data and initial collision condition parameters are then input into the multibody dynamics simulation software to adjust the friction coefficients between the vehicle and the pedestrian, between the pedestrian and the ground, and between the vehicle and the ground, so that the simulation results are consistent with the actual response in the accident video information, generating kinematic reconstruction results. The head centroid three-axis velocity and three-axis angular velocity at the moment of head-vehicle collision, as well as the head centroid three-axis velocity and three-axis angular velocity at the moment of head-ground collision, are extracted from the kinematic reconstruction results as head collision boundary condition data.

[0010] Furthermore, the verification of the kinematic reconstruction results includes comparing the simulation output with the accident video information item by item. The comparison items include the collision points of the vehicle and the pedestrian, the pedestrian's trajectory during the collision, rotation angle, landing posture, landing point, and the final stopping position of the vehicle and the pedestrian. The absolute error between the simulation-obtained final stopping position of the vehicle and the pedestrian and the actual stopping position at the accident scene is calculated. When the absolute error is less than the preset accuracy threshold, the kinematic reconstruction results are considered valid.

[0011] Further, the predictive performance scores of each injury criterion are calculated, and injury criteria with the highest predictive performance scores corresponding to different injury types are selected to generate a candidate injury criterion set. This includes binarizing the head injury level of each case in the accident case dataset, labeling cases with injury levels greater than or equal to AIS level 4 as positive samples, and labeling cases with injury levels less than AIS level 4 as negative samples. For each injury criterion in the head injury parameter set, the precision and recall of the criterion are calculated based on the injury parameter values ​​and binarization results corresponding to each case. Based on the precision and recall, the accuracy and recall of each injury criterion are calculated. Fraction, The score is equal to twice the product of precision and recall, divided by the sum of precision and recall; according to The damage criteria are sorted from highest to lowest score, and then selected. The damage criteria with the highest scores, which correspond to different damage types, are used to generate a set of candidate damage criteria.

[0012] Furthermore, the damage threshold and weight parameters of each damage criterion in the candidate damage criterion set are determined. This includes using a logistic regression method for each damage criterion in the candidate damage criterion set, with the damage parameter value of the criterion for each case as input and the corresponding binary annotation result as the output target, to fit the curve and obtain the AIS 4+ damage risk curve corresponding to the criterion. The criterion value corresponding to the damage probability equal to the preset probability value is extracted from the damage risk curve and used as the damage threshold of the damage criterion. The number of cases of each damage type in the accident case dataset is counted, and the ratio of the number of cases of each damage type to the total number of head injury cases is calculated. The ratio is used as the weight parameter of the corresponding damage criterion.

[0013] Furthermore, the fitting formula for the logistic regression method is the sum of the exponential function value of the damage probability equal to 1 divided by 1 and the negative value of the linear combination of the regression coefficient and the damage criterion value; the preset probability value is 0.5.

[0014] Furthermore, the formula for calculating the comprehensive weighted injury criterion value of the head is as follows: in, This represents the comprehensive weighted injury criterion value for the head. For the first Damage parameter values ​​for each candidate damage criterion; For the first Damage thresholds for each candidate damage criterion; For the first Weight parameters for each candidate damage criterion; The number of candidate damage criteria. When When the value is greater than or equal to 1, the risk of comprehensive head injury is determined to reach the AIS 4+ severe brain injury level; when When the score is less than 1, the risk of comprehensive head injury is considered to be lower than the AIS 4+ severe brain injury level.

[0015] This invention discloses a pedestrian head injury comprehensive assessment system based on high-precision accident reconstruction, used to execute the aforementioned method. It includes a data acquisition module for acquiring accident video information and head injury information from pedestrian deep-accident cases, generating an accident case dataset; a collision velocity estimation module for analyzing and processing the accident video information, estimating the pre-collision velocities of the accident vehicle and pedestrian, and acquiring collision velocity data; a kinematic reconstruction module for inputting the collision velocity data and initial collision condition parameters into multibody dynamics simulation software to reconstruct the pedestrian collision kinematic response, and extracting head collision boundary condition data from the kinematic reconstruction results; and a head injury reconstruction module for loading the head collision boundary condition data into... The system employs a finite element model of the human head to perform finite element simulation calculations and obtain a set of head injury parameters. A criterion selection module calculates the predictive performance score of each injury criterion based on the head injury parameter set and information from all cases, selecting the injury criteria with the highest predictive performance scores for different injury types to generate a candidate injury criterion set. A parameter determination module determines the injury threshold and weight parameters for each injury criterion in the candidate injury criterion set. A comprehensive injury evaluation module normalizes the injury parameter values ​​of each candidate injury criterion by dividing them by the corresponding injury threshold, then performs a weighted summation using the weight parameters to obtain a comprehensive weighted head injury criterion value, which serves as the comprehensive head injury evaluation result.

[0016] This invention employs a dual velocity estimation method combining frame-by-frame analysis and direct linear transformation, and obtains head injury parameters through multi-stage reconstruction using multi-body dynamics simulation and finite element simulation, providing a reliable data foundation for comprehensive evaluation. Using scores as the selection criterion comprehensively considers the impact of precision and recall, avoiding biases that arise when selecting criteria based on a single evaluation indicator. Normalization is achieved by dividing the damage parameter values ​​of each candidate damage criterion by the corresponding damage threshold, eliminating comparability differences between criteria with different dimensions. Weight parameters for each criterion are determined based on statistical results of damage types from accident cases, ensuring that the comprehensive evaluation results reflect the frequency of occurrence of each damage type in real accidents. The final obtained head comprehensive weighted damage criterion value... By integrating multiple criteria corresponding to different injury types, this study solves the technical problem that existing single injury criteria cannot assess the risk of complex severe brain injury with multiple injury types occurring simultaneously, and achieves the technical effect of comprehensively evaluating the risk of AIS 4+ combined head injury in pedestrians. Attached Figure Description

[0017] Figure 1 This is a technical roadmap of the pedestrian head comprehensive weighted injury criterion evaluation method according to an embodiment of the present invention, which illustrates the overall technical process from accident case screening, high-precision accident reconstruction, injury criterion assessment to the establishment of comprehensive weighted injury criteria; Figure 2 This is a flowchart of a high-precision accident reconstruction according to an embodiment of the present invention, which includes three stages: vehicle and pedestrian speed estimation stage, pedestrian kinematics reconstruction stage, and pedestrian head injury reconstruction stage. Figure 3 This is an evaluation result diagram of the predictive effect of various injury criteria on head injury in the embodiments of the present invention, showing the comparison of the predictive performance of 16 head injury criteria based on the F1-score evaluation index. Detailed Implementation Example 1

[0018] This embodiment provides a method for evaluating comprehensive pedestrian head injuries based on high-precision accident reconstruction, used to assess the risk of severe brain injury (AIS 4+) caused by multiple injury types simultaneously. The method for evaluating comprehensive pedestrian head injuries based on high-precision accident reconstruction is executed by computer, with the simulation calculations running on a computer system equipped with multibody dynamics simulation software and finite element analysis software. AIS is a simplified injury grading standard, and AIS 4+ indicates a severe injury with a grade greater than or equal to 4.

[0019] A method for comprehensive assessment of pedestrian head injuries based on high-precision accident reconstruction according to an embodiment of the present invention includes the following steps: Step 1: Obtain in-depth accident case data This study acquires accident video and head injury information from pedestrian deep-impact accident cases to generate an accident case dataset. The accident video information consists of image records of the entire accident process. The head injury information includes the injury type and severity.

[0020] It should be noted that each case in the aforementioned accident case dataset meets the following conditions: the accident video is clear and complete, allowing identification of the movement trajectories of the vehicle and pedestrians before and after the collision; the head injury report details the injury type and injury level according to AIS standards. Head injury types include one or more of the following: skull fracture, subdural hematoma, subarachnoid hemorrhage, cerebral contusion, and diffuse brain injury. Diffuse brain injury includes concussion and diffuse axonal injury.

[0021] Step 2: Estimate the collision speed based on accident video information. The accident video information is analyzed and processed to estimate the pre-collision speed of the vehicles and pedestrians involved in the accident and obtain collision speed data.

[0022] Step 2 may specifically include: Step 201: Analyze the accident video frame by frame, extract the position coordinates of the accident vehicle and pedestrian in consecutive frames before the collision, and calculate a preliminary estimate of the pre-collision velocity based on the ratio of inter-frame displacement to frame rate.

[0023] Step 202: The direct linear transformation method is used to solve the mapping relationship between the two-dimensional pixel coordinates in the accident video and the actual three-dimensional spatial coordinates, obtaining an accurate estimate of the pre-collision velocity. The input of the direct linear transformation method is the pixel coordinates of the control points in the video frame and their corresponding actual three-dimensional spatial coordinates; the output is the transformation parameters. At least six control points with known three-dimensional spatial coordinates are selected at the accident site. The transformation formula is as follows: in, and These are the pixel coordinates of the control point within the video frame; , , These are the actual three-dimensional spatial coordinates of the control points, in meters; to Here are eleven transformation parameters to be solved, and are dimensionless ratio coefficients to ensure dimensional consistency between the two sides of the formula. Based on the solved transformation parameters, the pixel coordinates of the accident vehicle and pedestrian in consecutive frames are converted into actual spatial coordinates, the actual displacement between adjacent frames is calculated, and thus an accurate estimate of the pre-collision velocity is obtained.

[0024] In this embodiment of the application, to improve the accuracy of collision speed estimation, the preliminary estimate obtained in step 201 and the precise estimate obtained in step 202 are cross-validated. When the deviation between the estimation results of the two methods is within a preset threshold range, the average of the two is taken as the final collision speed data; when the deviation exceeds the preset threshold, the precise estimate of the direct linear transformation method is used as the final collision speed data.

[0025] Step 3, Reconstructing the pedestrian collision kinematic response For cases where collision velocity data has been obtained, step 3 can be further performed. Step 3 may specifically include: Step 301: Extract initial collision condition parameters based on accident video information. These parameters include the collision angle, collision location, and the point of impact between the head and the vehicle.

[0026] Step 302: Input the collision velocity data and initial collision condition parameters into the multibody dynamics simulation software, adjust the friction coefficients between the vehicle and pedestrian, between the pedestrian and the ground, and between the vehicle and the ground to make the simulation results consistent with the actual response in the accident video, and generate kinematic reconstruction results. The kinematic reconstruction results include the entire motion trajectory of the pedestrian and vehicle collision process, landing posture, impact point, and the final stopping position of the vehicle and pedestrian.

[0027] Step 303: Extract head collision boundary condition data from the kinematic reconstruction results. The head collision boundary condition data includes the three-axis velocities and three-axis angular velocities of the head's center of mass at the moment of head-vehicle collision, and the three-axis velocities and three-axis angular velocities of the head's center of mass at the moment of head-ground collision.

[0028] It should be noted that the aforementioned multibody dynamics simulation software is a coupled simulation environment of PC-Crash and MADYMO. PC-Crash receives collision velocity data and initial collision condition parameters as input and outputs the simulation results of vehicle collision dynamics. MADYMO receives the vehicle dynamics results output by PC-Crash as input and outputs the simulation results of pedestrian multibody kinematic response. The two are coupled to achieve co-simulation of vehicle collision dynamics and pedestrian multibody kinematics.

[0029] It should be noted that the kinematic reconstruction results are verified by comparing the simulation output with the accident video recording item by item. The comparison items include: the collision point between the vehicle and the pedestrian, the pedestrian's trajectory during the collision, rotation angle, landing posture, impact point, and the final stopping positions of both the vehicle and the pedestrian. The kinematic reconstruction results are considered valid only when all the comparison results match the accident video recording.

[0030] In this embodiment, to further quantify the accuracy of kinematic reconstruction, the absolute error between the simulated final stopping positions of the vehicle and pedestrians and the actual stopping positions at the accident scene is calculated. When the absolute error is less than a preset accuracy threshold, the kinematic reconstruction accuracy is considered to meet the requirements.

[0031] Step 4, Reconstructing the head injury The head collision boundary condition data is loaded into the human finite element head model, and finite element simulation calculations are performed to obtain the head injury parameter set.

[0032] It should be noted that the above-mentioned human finite element head model is the head sub-model of the THUMS human finite element model. The human finite element head model receives head collision boundary condition data as the boundary condition input applied to the head's center of mass, and outputs the mechanical response parameters of each anatomical structure of the head. The human finite element head model includes finite element mesh representations of the head's anatomical structures such as the skull, dura mater, arachnoid mater, brain parenchyma, and brainstem.

[0033] It should be noted that the above-mentioned set of head injury parameters includes two categories: injury parameters based on head kinematics and injury parameters based on brain tissue response. Injury parameters based on head kinematics include: Values, peak linear acceleration, peak angular acceleration, and peak angular velocity. Among them, These are the head injury criteria values ​​calculated based on the synthesized linear acceleration of the head's center of mass within a 15-millisecond time window. Injury parameters based on brain tissue response include: maximum principal strain of brain tissue, peak intracranial impact pressure, and cumulative strain damage measure. The head injury parameter set covers a total of sixteen injury indicators.

[0034] Step 5: Select damage criteria with better prediction performance. Based on the head injury parameter set and corresponding head injury information of all cases in the accident case dataset, the predictive performance of each injury criterion for head AIS 4+ injury is evaluated, and the injury criteria with better predictive performance are selected to obtain a candidate injury criterion set.

[0035] Step 5 may specifically include: Step 501: Binarize and label the head injury level of each case in the accident case dataset. Cases with an injury level greater than or equal to AIS 4 are labeled as positive samples, and cases with an injury level less than AIS 4 are labeled as negative samples.

[0036] Step 502: For each of the sixteen damage criteria, based on the damage parameter values ​​and binarized annotation results corresponding to each case, calculate the accuracy, precision, recall, and specificity of the criterion.

[0037] Step 503: Based on the precision and recall of each damage criterion, calculate the accuracy and recall of each criterion. Fraction. The formula for calculating fractions is as follows: in, For accuracy, that is, the proportion of samples predicted to be positive that are actually positive; Recall rate is the proportion of all positive samples that are correctly predicted to be positive.

[0038] Step 504, according to The sixteen damage criteria were ranked from highest to lowest score, and then selected... The damage criteria with the highest scores, which correspond to different damage types, are used to generate a set of candidate damage criteria.

[0039] It should be noted that the following is adopted: The score is used as a screening criterion because The score comprehensively measures precision and recall, and can weigh the differences in sensitivity of different evaluation indicators to positive and negative category data, avoiding the bias that may occur when relying on a single indicator criterion.

[0040] In this embodiment of the application, the candidate damage criterion set includes four damage criteria, namely peak angular acceleration, Peak intracranial impact pressure and maximum principal strain in brain tissue. Peak angular acceleration corresponds to the assessment of diffuse brain injury. The peak intracranial impact pressure corresponds to the assessment of brain contusion, and the maximum principal strain of brain tissue corresponds to the assessment of subdural hematoma and subarachnoid hemorrhage.

[0041] Step 6: Determine the damage threshold and weight parameters for each candidate damage criterion. Based on the accident case dataset and the candidate injury criterion set, the injury threshold and weight parameters of each candidate injury criterion are determined, and the parameter data of the head comprehensive weighted injury criterion are obtained.

[0042] Step 6 may specifically include: Step 601: Determine the damage threshold. For each damage criterion in the candidate damage criterion set, a logistic regression method is used. The damage parameter value of each case for that criterion is used as input, and the corresponding binary annotation result is used as the output target to fit the curve, obtaining the AIS 4+ damage risk curve corresponding to that criterion. The fitting formula for the logistic regression method is as follows: in, For the damage criterion value The probability of AIS 4+ injury occurring; and The regression coefficients are obtained by fitting logistic regression. This represents an exponential function. Extracted from the AIS 4+ damage risk curve. The corresponding criterion value is used as the damage threshold for this damage criterion. ,in This is the number of the candidate damage criterion.

[0043] Step 602: Determine the weighting parameters. Collect all cases with detailed head injury reports from the accident case dataset, and calculate the number of cases for each of the four injury types: skull fracture, subdural hematoma and subarachnoid hemorrhage, cerebral contusion, and diffuse brain injury. Calculate the ratio of the number of cases for each injury type to the total number of head injury cases, and use this ratio as the weighting parameter for the corresponding injury criterion. .

[0044] It should be noted that the correspondence between the above injury types and candidate injury criteria is as follows: diffuse brain injury corresponds to peak angular acceleration, skull fracture corresponds to... Cerebral contusion corresponds to the peak intracranial impact pressure, while subdural hematoma and subarachnoid hemorrhage correspond to the maximum principal strain in brain tissue. The basis for these correspondences is the biomechanical injury mechanism of each injury type: diffuse brain injury is mainly caused by rotational loading, with peak angular acceleration being an indicator of the intensity of the rotational load; skull fractures are mainly caused by linear loading. The index is used to characterize the intensity of linear load; brain contusion is related to local intracranial impact pressure, and the peak value of intracranial impact pressure is used to characterize the intensity of local pressure; subdural hematoma and subarachnoid hemorrhage are related to bridging vein rupture caused by brain tissue deformation, and the maximum principal strain of brain tissue is used to characterize the degree of brain tissue deformation.

[0045] Furthermore, the damage parameter values ​​of the four candidate damage criteria They differ in physical dimensions: the unit of peak angular acceleration is... , The value is dimensionless; the unit for peak intracranial impact pressure is _____. The maximum principal strain of brain tissue is dimensionless. To eliminate the influence of dimensional differences on subsequent weighted summation calculations, in step 7, the values ​​of each criterion damage parameter are... Divide by the corresponding damage threshold Normalization is performed to make the ratios of each criterion equal. Both are dimensionless quantities, thus making them comparable.

[0046] Step 7: Calculate the comprehensive weighted injury assessment results for the head. Based on the damage parameter values, damage thresholds, and weight parameters of each criterion in the candidate damage criterion set, the comprehensive weighted damage criterion value for the head is calculated to obtain the comprehensive head damage evaluation result. The calculation formula for the comprehensive weighted damage criterion for the head is as follows: in, The comprehensive weighted injury criterion value for the head is used to characterize the comprehensive risk of severe brain injury (AIS 4+). For the first Damage parameter values ​​for each candidate damage criterion; For the first Damage thresholds for each candidate damage criterion; For the first Weight parameters for each candidate damage criterion; The number of the candidate damage criterion; The number of candidate damage criteria.

[0047] It should be noted that in the above formula This represents the normalized results of the damage parameter values ​​for each criterion. By dividing the damage parameter values ​​for each criterion by the corresponding threshold when the AIS 4+ damage risk is 50%, the damage criteria with different dimensions are made comparable. To ensure that the weighted average of head injuries is based on the proportion of each injury type in accident statistics, the comprehensive weighted injury criterion can reflect the frequency of each injury type in real accidents.

[0048] Furthermore, when At that time, the risk of comprehensive head injury in the collision was determined to be AIS4+, a level of severe brain injury; when At that time, the risk of comprehensive head injury was determined to be lower than the AIS 4+ severe brain injury level. The above-mentioned thresholds are based on the injury thresholds of each criterion. All correspond to the AIS 4+ damage risk criterion value of 50%, when each normalized ratio When the weighted sum equals 1, it indicates that the overall damage risk has just reached the 50% AIS 4+ damage probability boundary.

[0049] In this embodiment, the specific damage threshold and weight parameters obtained in steps 601 and 602 are substituted into the above general formula to obtain the specific head comprehensive weighted damage criterion calculation equation as follows: in, Peak angular acceleration, in units of The damage threshold is The weight parameters are ; The 15-millisecond head injury criterion value is used, and the injury threshold is... The weight parameters are ; Peak intracranial impact pressure, in units of The damage threshold is The weight parameters are ; The maximum principal strain in brain tissue is [value], and the damage threshold is [value]. The weight parameters are The sum of the four weight parameters is .

[0050] In this embodiment of the application, for verification The predictive performance of the risk of comprehensive head injury to pedestrians was evaluated using the area under the receiver operating characteristic curve (AUROC). The predictive performance of each case was compared and evaluated with that of each individual damage criterion. The values ​​and binarized annotation results are input into the AUROC calculation process to obtain... The AUROC value is calculated. Simultaneously, the AUROC value for each individual damage criterion is calculated and compared. The AUROC values ​​differ from those of each individual criterion. Building upon step 6, independent real-world accident cases not involved in the determination of injury thresholds and weight parameters are used, and the head injury parameter values ​​from these independent real-world accident cases are substituted into the above equations for calculation. The values ​​were compared with actual head injury information from independent real-world accident cases to verify them. Predictive effectiveness for diverse brain injuries.

[0051] On the other hand, this invention also proposes a comprehensive pedestrian head injury assessment system based on high-precision accident reconstruction, comprising: The data acquisition module is used to acquire accident video information and head injury information of pedestrian deep accident cases and generate accident case datasets.

[0052] The collision speed estimation module is used to perform frame-by-frame analysis and direct linear transformation on accident video information to estimate the pre-collision speeds of the accident vehicles and pedestrians and obtain collision speed data.

[0053] The kinematics reconstruction module is used to input collision velocity data and initial collision condition parameters into multibody dynamics simulation software to reconstruct the pedestrian collision kinematic response and obtain head collision boundary condition data.

[0054] The head injury reconstruction module is used to load head collision boundary condition data into the human finite element head model, perform finite element simulation calculations, and obtain the head injury parameter set.

[0055] The criterion selection module is used to calculate the criterion for each injury based on the set of head injury parameters and head injury information from all cases. The scores are used to filter out the damage criteria with better prediction performance and obtain a set of candidate damage criteria.

[0056] The parameter determination module is used to determine the damage threshold of each candidate damage criterion using logistic regression, determine the weight parameters of each candidate damage criterion based on the statistical results of damage types in accident cases, and obtain the parameter data of the head comprehensive weighted damage criterion.

[0057] The comprehensive damage assessment module is used to calculate the comprehensive weighted head damage criterion formula based on the damage parameter values, damage thresholds, and weight parameters of each criterion in the candidate damage criterion set. The value is used to obtain the comprehensive head injury assessment results.

[0058] By way of example, the system proposed in this invention may also include various features and combinations thereof in the method embodiments, which will not be elaborated here.

[0059] The pedestrian head injury comprehensive assessment method proposed in this embodiment obtains accurate head injury parameters through dual velocity estimation using frame-by-frame video analysis and direct linear transformation, as well as multi-stage accident reconstruction coupled with multibody dynamics and finite element methods, providing a reliable data foundation for subsequent comprehensive assessment. In the criterion selection stage, the method employs... The score, as a comprehensive evaluation indicator, balances the impact of precision and recall, avoiding the criterion selection bias caused by a single evaluation indicator. In the comprehensive evaluation stage, criteria based on head kinematics and criteria based on brain tissue response are combined through normalization and weighted summation, resulting in... It can simultaneously reflect the risk of different types of damage to the head caused by both linear and rotational loads. Compared with a single damage criterion, This study integrates four criteria corresponding to different injury types, enabling the assessment of the risk of complex severe brain injury caused by multiple injury types simultaneously. It overcomes the limitation of existing single criteria that can only evaluate a single injury form, and provides a more comprehensive reference method for head injury assessment in pedestrian protection virtual assessment.

[0060] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for comprehensive assessment of pedestrian head injuries based on high-precision accident reconstruction, characterized in that, Includes the following steps: Obtain accident video information and head injury information of pedestrian deep accident cases to generate an accident case dataset; The accident video information is analyzed and processed to estimate the pre-collision speed of the accident vehicle and pedestrian, and to obtain collision speed data; The collision velocity data and initial collision condition parameters are input into multibody dynamics simulation software to reconstruct the pedestrian collision kinematic response, and the head collision boundary condition data are extracted from the kinematic reconstruction results. The head collision boundary condition data is loaded into the human finite element head model, and finite element simulation calculation is performed to obtain the head injury parameter set. Based on the head injury parameter set and head injury information of all cases, the prediction performance score of each injury criterion is calculated, and the injury criteria corresponding to different injury types and with the highest prediction performance scores are selected to generate a candidate injury criterion set. Determine the damage threshold and weight parameters for each damage criterion in the candidate damage criterion set; The damage parameter values ​​of each candidate damage criterion are normalized by dividing them by the corresponding damage threshold, and then weighted and summed using the weight parameters to obtain the comprehensive weighted damage criterion value for the head, which is used as the comprehensive head damage evaluation result.

2. The method for comprehensive evaluation of pedestrian head injuries based on high-precision accident reconstruction according to claim 1, characterized in that, The process of analyzing and processing the accident video information to estimate the pre-collision velocities of the vehicles and pedestrians, and to obtain collision speed data, includes: The accident video is analyzed frame by frame to extract the position coordinates of the accident vehicle and pedestrian in consecutive frames before the collision, and a preliminary estimate of the pre-collision velocity is calculated based on the ratio of inter-frame displacement to frame rate. The direct linear transformation method is adopted to solve the transformation parameters based on the mapping relationship between the pixel coordinates of the control points at the accident site and the actual three-dimensional spatial coordinates. The pixel coordinates in the continuous frames are converted into actual spatial coordinates, the actual displacement between adjacent frames is calculated, and the accurate estimated value of the pre-collision velocity is obtained. The preliminary estimate and the precise estimate are cross-validated. When the deviation between the two is within a preset threshold range, the average of the two is taken as the collision speed data; when the deviation exceeds the preset threshold, the precise estimate is taken as the collision speed data.

3. The method for comprehensive evaluation of pedestrian head injuries based on high-precision accident reconstruction according to claim 2, characterized in that, In the direct linear transformation method, at least six control points with known three-dimensional spatial coordinates are selected at the accident site. The transformation formula is: the pixel coordinates of each control point are equal to the quotient obtained by dividing the three-dimensional spatial coordinates of the control point by another linear combination of the three-dimensional spatial coordinates using the same set of transformation parameters after linear combination of the three-dimensional spatial coordinates using eleven transformation parameters.

4. The method for comprehensive evaluation of pedestrian head injuries based on high-precision accident reconstruction according to claim 1, characterized in that, The process involves inputting the collision velocity data and initial collision condition parameters into multibody dynamics simulation software to reconstruct the pedestrian collision kinematic response, and extracting head collision boundary condition data from the kinematic reconstruction results, including: Based on the accident video information, the initial collision condition parameters are extracted, including the collision angle, collision position, and the point of head-to-vehicle collision. The collision velocity data and the initial collision condition parameters are input into the multibody dynamics simulation software. The friction coefficients between the vehicle and the pedestrian, between the pedestrian and the ground, and between the vehicle and the ground are adjusted to make the simulation results consistent with the actual response in the accident video information, and the kinematic reconstruction results are generated. The head's triaxial velocity and triaxial angular velocity at the moment of head-vehicle collision, and the head's triaxial velocity and triaxial angular velocity at the moment of head-ground collision are extracted from the kinematic reconstruction results and used as the head collision boundary condition data.

5. The method for comprehensive evaluation of pedestrian head injuries based on high-precision accident reconstruction according to claim 4, characterized in that, The verification of the kinematic reconstruction results includes: The simulation output is compared with the accident video information item by item. The comparison items include the collision point between the vehicle and the pedestrian, the pedestrian's motion trajectory during the collision, rotation angle, landing posture, landing point, and the final stopping position of the vehicle and the pedestrian. The absolute error between the final stopping position of the vehicle and pedestrian obtained from the simulation and the actual stopping position at the accident scene is calculated. When the absolute error is less than a preset accuracy threshold, the kinematic reconstruction result is considered valid.

6. The method for comprehensive evaluation of pedestrian head injuries based on high-precision accident reconstruction according to claim 1, characterized in that, The process involves calculating the predictive performance score of each damage criterion, filtering out the damage criteria with the highest predictive performance scores corresponding to different damage types, and generating a candidate damage criterion set, including: The head injury level of each case in the accident case dataset is binarized and labeled. Cases with an injury level greater than or equal to AIS 4 are labeled as positive samples, and cases with an injury level less than AIS 4 are labeled as negative samples. For each injury criterion in the head injury parameter set, the precision and recall of the criterion are calculated based on the injury parameter values ​​corresponding to each case and the binarized annotation results. Each damage criterion is calculated based on the aforementioned precision and recall. Fractions, the The score is equal to twice the product of precision and recall, divided by the sum of precision and recall. According to the above The damage criteria are sorted from highest to lowest score, and then selected. The candidate damage criterion set is generated by ranking the damage criteria with the highest scores and corresponding to different damage types.

7. The method for comprehensive evaluation of pedestrian head injuries based on high-precision accident reconstruction according to claim 1, characterized in that, The step of determining the damage threshold and weight parameters of each damage criterion in the candidate damage criterion set includes: For each damage criterion in the candidate damage criterion set, a logistic regression method is used to fit the damage parameter value of the criterion for each case as input and the corresponding binary annotation result as output target to obtain the AIS4+ damage risk curve corresponding to the criterion. The criterion value corresponding to the damage probability equal to the preset probability value is extracted from the damage risk curve and used as the damage threshold of the damage criterion. The number of cases of each injury type in the accident case dataset is counted, and the ratio of the number of cases of each injury type to the total number of head injury cases is calculated. The ratio is used as the weight parameter of the corresponding injury criterion.

8. The method for comprehensive evaluation of pedestrian head injuries based on high-precision accident reconstruction according to claim 7, characterized in that, The fitting formula for the logistic regression method is: the damage probability equals the sum of the exponential function value of 1 divided by 1 and the negative value of the linear combination of the regression coefficient and the damage criterion value; the preset probability value is 0.

5.

9. The method for comprehensive evaluation of pedestrian head injuries based on high-precision accident reconstruction according to claim 1, characterized in that, The formula for calculating the comprehensive weighted injury criterion value of the head is as follows: ； in, This represents the comprehensive weighted injury criterion value for the head. For the first Damage parameter values ​​for each candidate damage criterion; For the first Damage thresholds for each candidate damage criterion; For the first Weight parameters for each candidate damage criterion; The number of candidate damage criteria; When the When the value is greater than or equal to 1, the risk of comprehensive head injury is determined to reach the AIS 4+ severe brain injury level; when the... When the score is less than 1, the risk of comprehensive head injury is considered to be lower than the AIS 4+ severe brain injury level.

10. A pedestrian head injury assessment system based on high-precision accident reconstruction, used to execute the pedestrian head injury assessment method based on high-precision accident reconstruction as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to acquire accident video information and head injury information of pedestrian deep accident cases and generate accident case datasets; The collision speed estimation module is used to analyze and process the accident video information, estimate the pre-collision speed of the accident vehicle and pedestrian, and obtain collision speed data. The kinematics reconstruction module is used to input the collision velocity data and initial collision condition parameters into the multibody dynamics simulation software, reconstruct the pedestrian collision kinematic response, and extract the head collision boundary condition data from the kinematics reconstruction results. The head injury reconstruction module is used to load the head collision boundary condition data into the human finite element head model, perform finite element simulation calculations, and obtain the head injury parameter set. The criteria screening module is used to calculate the prediction performance score of each injury criterion based on the head injury parameter set and head injury information of all cases, and to screen the injury criteria that correspond to different injury types and have the highest prediction performance scores to generate a candidate injury criterion set. The parameter determination module is used to determine the damage threshold and weight parameters of each damage criterion in the candidate damage criterion set. The comprehensive damage assessment module is used to normalize the damage parameter values ​​of each candidate damage criterion by dividing them by the corresponding damage threshold, and then summing them by weighting the weight parameters to obtain the comprehensive weighted damage criterion value of the head, which is used as the comprehensive damage assessment result of the head.

Images