Pedestrian traffic accident damage reconstruction method and system based on video image

Abstract

The invention provides a pedestrian traffic accident damage reconstruction method and system based on a video image, and relates to the technical field of traffic safety, and the method comprises the following steps: screening pedestrian traffic accidents, and obtaining video image data; obtaining a pedestrian attitude parameter from the video image data; constructing a vehicle model and a pedestrian model of the traffic accident; simulating and reconstructing a collision process of the vehicle and the pedestrian, and calculating kinematics parameters and damage parameters of the pedestrian; and evaluating the credibility of the damage reconstruction result based on the sequential comprehensive quantitative evaluation, and constructing a damage reconstruction database. Effective pedestrian traffic accident video data are screened, pedestrian posture parameters are obtained, models of vehicles and pedestrians are constructed, the collision process of the vehicles and the pedestrians is simulated, reconstructed and restored, the personnel injury mechanism is clarified, data support is provided for accurate judgment of accident liability, analysis and iteration of vehicle safety performance, traffic safety management optimization and the like, and the method is suitable for popularization and application. The automation of traffic accident deep damage reconstruction is realized, and the precision and efficiency of damage reconstruction are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of traffic safety, and in particular, the present application relates to a pedestrian traffic accident injury reconstruction method and system based on video images. BACKGROUND

[0002] At present, China's automobile ownership ranks first in the world, and the difficulty of traffic safety management is increasing accordingly. The number of pedestrian traffic accident casualties accounts for a higher and higher proportion of all traffic accident casualties. This proportion may fluctuate in different years and regions, but it remains at a high level overall.

[0003] The deep injury reconstruction technology based on real traffic accidents is the key to clarify the injury mechanism and injury factors of the injured in traffic accidents. Through deep simulation reconstruction of a large number of traffic accidents, the law is found, which is of great significance to reduce pedestrian injuries.

[0004] Chinese patent application patent CN119323169A discloses a collision accident rapid reconstruction platform and a reconstruction method, which includes an accident information collection module for collecting real traffic accident information to be reconstructed; a multi-rigid-body modeling module for modeling the first collision vehicle, the second collision vehicle and the traffic participant in the accident in MADYMO software; a simulation analysis module for simulating and calculating the simulation matrix generated by the optimization algorithm module; a parameter conversion module for converting the parameters extracted from the traffic accident video into modeling parameters identifiable in the simulation model; an optimization algorithm module for determining the multi-objective optimization algorithm, the design variable, the constraint condition and the reconstruction target. The above-mentioned application significantly improves the collision accident reconstruction efficiency.

[0005] However, the above-mentioned application patent still has problems. On the one hand, the above-mentioned patent is only applicable to kinematic reconstruction and cannot obtain detailed injury biomechanics information; secondly, the quality of traffic accident information is uneven and the granularity is not uniform, and the availability of the accident is not screened and distinguished; thirdly, the pedestrian posture parameters and pedestrian motion trajectory are not extracted based on the video information of the key collision moment of the accident; fourthly, there is no credibility judgment of the reconstruction result. At present, there is no uniform standard for injury reconstruction based on real traffic accidents in China, and the judgment of the effectiveness of the reconstruction method and the reconstruction result mostly depends on the subjective judgment of technical personnel, resulting in low persuasiveness and credibility of the injury result. And the traditional traffic accident injury reconstruction usually needs to obtain the kinematic information of the key parts through kinematic reconstruction first, and then perform finite element reconstruction, which is time-consuming and has information transmission loss.

[0006] Therefore, in order to solve the above problems, it is necessary for us to design a pedestrian traffic accident injury reconstruction method based on video images. SUMMARY

[0007] The purpose of this invention is to provide a method for pedestrian traffic accident injury reconstruction based on video images. This method filters effective pedestrian traffic accident video data, obtains pedestrian posture parameters, constructs vehicle and pedestrian models, simulates and reconstructs the collision process between vehicles and pedestrians, clarifies the mechanism of personal injury, and provides scientific data support for accurate determination of accident liability, resolution of insurance claim disputes, iterative analysis of vehicle safety performance, and optimization of traffic safety management. It automates deep injury reconstruction of traffic accidents, improving the accuracy and efficiency of injury reconstruction.

[0008] To achieve the above objectives, the present invention employs the following technical solution: A method for pedestrian traffic accident injury reconstruction based on video images, the method includes the following steps: S1: Filter pedestrian traffic accidents from the traffic accident database and obtain video image data of pedestrian traffic accidents; S2: Obtain pedestrian posture parameters from video image data of pedestrian traffic accidents; S3: Construct vehicle and pedestrian models for traffic accidents; S4: Simulate and reconstruct the collision process between a vehicle and a pedestrian, and calculate the kinematic parameters and injury parameters of the pedestrian; S5: Based on time-series comprehensive quantitative evaluation, evaluate the credibility of damage reconstruction results and construct a damage reconstruction database.

[0009] As a preferred embodiment of the present invention, step S1 is specifically performed as follows: S11: Filter pedestrian traffic accidents from the traffic accident database; S12: Determine whether the completeness of pedestrian traffic accident data is higher than the first preset threshold; if yes, proceed to step S13; otherwise, do not perform the operation. S13: Acquire video image data of pedestrian traffic accidents.

[0010] As a preferred embodiment of the present invention, when performing step S11, pedestrian traffic accidents are screened from the traffic accident database, and the screening criteria include: A pedestrian comes into contact with or collides with a vehicle; The pedestrian sustained at least one injury; There is video footage of the collision.

[0011] As a preferred embodiment of the present invention, step S12 is specifically performed as follows: Read the structured information form of pedestrian traffic accident data to obtain accident data including scene information, vehicle information, personnel information, and collision information; The system uses a dynamic threshold method based on the missing rate of key fields to determine whether the completeness of the accident data is higher than the first preset threshold; if so, step S13 is executed; otherwise, no operation is performed.

[0012] As a preferred embodiment of the present application, when step S2 is performed, the pedestrian posture parameters are obtained from the video image data of the pedestrian traffic accident, specifically: If the video image data is obtained by a single camera, the video image data is fused by using the SMPL-X human body model and the NeRF neural radiation field fusion algorithm, and the three-dimensional joint angles of the human body are calculated by time sequence optical flow constraint, to obtain the pedestrian posture parameters. If the video image data is obtained by multiple cameras, the video image data of each camera is extracted, and the human body in different angle videos is synchronously detected and tracked to ensure that the two-dimensional key point information of the human body under different angles is obtained at the same time, to obtain the pedestrian posture parameters.

[0013] As a preferred embodiment of the present application, the video image data of the pedestrian traffic accident is preprocessed before step S2 is performed.

[0014] As a preferred embodiment of the present application, when step S3 is performed, the vehicle model and the pedestrian model of the traffic accident are constructed, specifically: S31: Constructing a vehicle model, adjusting and constructing a vehicle model according to the scanning point cloud or the proxy model and the vehicle information in the pedestrian traffic accident data; S32: Constructing a pedestrian model, constructing a pedestrian initial model according to the pedestrian information in the pedestrian traffic accident data; S33: According to the pedestrian posture parameters, determining the target angles of each joint and the target positions of each body part of the pedestrian initial model, and applying a gradually changing load to the pedestrian initial model in the finite element software to adjust the finite element pedestrian initial model to a state consistent with the pedestrian posture parameters, to obtain the pedestrian model.

[0015] As a preferred embodiment of the present application, when step S4 is performed, the collision process between the vehicle and the pedestrian is simulated and reconstructed, and the kinematic parameters and the damage parameters of the pedestrian are calculated, specifically: After adjusting the posture and position of the vehicle model and the pedestrian model, a multi-objective optimization calling program is used to call LS-DYNA for calculation and solving, and in the solving process, the software simulates the collision process between the vehicle and the pedestrian in the accident, and calculates the kinematic parameters and the damage parameters of the pedestrian.

[0016] As a preferred embodiment of the present application, when step S4 is performed, The kinematic parameters of the pedestrian include head angular acceleration, HIC 15 , chest compression, Rmax, chest viscosity index VC, femoral bending moment, femoral upper / lower cross-sectional force, iliac lateral compression force, knee ligament stretch, and tibial bending moment; Damage parameters include skull strain distribution, intracranial pressure distribution, cortical bone strain of each rib, visceral stress, pressure, femoral strain, pelvic strain, tibia strain, and fibular strain. When simulating an accident, the vehicle speed, vehicle deceleration, pedestrian speed, pedestrian joint parameters, and friction coefficient between the parties are used as design variables; the contact positions of different parts of the pedestrian with the vehicle are set as constraints, and the final landing position of the pedestrian and the relative position of the pedestrian and the vehicle are used as objective functions.

[0017] As a preferred embodiment of the present invention, when performing step S5, the credibility of the damage reconstruction result is evaluated based on the temporal comprehensive quantitative assessment. Specifically, the simulation results at the critical moment of the collision are compared with the pedestrian core joint parameters in the accident video image to determine whether there is a difference. The basic credibility is decayed exponentially to form a credibility rating. It is determined whether the credibility is higher than the second preset threshold. If so, the reconstruction result is credible and a damage reconstruction database is constructed. Otherwise, the reconstruction result is basically credible and the damage reconstruction database is constructed. Among them, the comparison of pedestrian core parameters includes: T before collision -1 The pedestrian's head and neck joint θ at five time points: contact time T0, separation time T1, pedestrian's mid-air flip time T2, and pedestrian's landing time T3. H-N pelvic joint θ Plvis , knee joint θ Knee angle.

[0018] Another aspect of the present invention is to provide a pedestrian traffic accident injury system based on video images, comprising: Data filtering module; Attitude parameter acquisition module; Model building module; Simulation reconstruction module; Database building module; The data filtering module filters pedestrian traffic accidents from the traffic accident database and obtains video image data of pedestrian traffic accidents; the posture parameter acquisition module obtains pedestrian posture parameters from the video image data of pedestrian traffic accidents; the simulation reconstruction module simulates and reconstructs the collision process between vehicles and pedestrians, and calculates the kinematic parameters and injury parameters of pedestrians; the database construction module evaluates the credibility of the injury reconstruction results based on time-series comprehensive quantitative evaluation and constructs an injury reconstruction database.

[0019] Thirdly, an apparatus is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a pedestrian traffic accident damage reconstruction method based on video images provided by any of the implementations of the first aspect above.

[0020] In a fourth aspect, the embodiments of the present application provide a computer storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, the program instructions can implement the method for pedestrian traffic accident injury reconstruction based on a video image according to any one of the implementation manners of the first aspect.

[0021] The method and system for pedestrian traffic accident injury reconstruction based on a video image have the following advantages: effective pedestrian traffic accident video data is screened, pedestrian posture parameters are obtained, a model of a vehicle and a pedestrian is constructed, a collision process between the vehicle and the pedestrian is simulated and reconstructed, a personnel injury mechanism is clarified, scientific data support is provided for accurate determination of accident liability, resolution of insurance claim disputes, iteration of vehicle safety performance analysis, optimization of traffic safety management, and the like, and traffic accident deep injury reconstruction automation is achieved, and the accuracy and efficiency of injury reconstruction are improved. BRIEF DESCRIPTION OF DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required to be used in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort.

[0023] Figure 1 FIG. 1 is a flowchart of the method for pedestrian traffic accident injury reconstruction based on a video image. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application.

[0025] In the following description, the terms "first" and "second" are only for the purpose of description, and cannot be understood as indicating or implying relative importance. The following description provides multiple embodiments of the present application, and different embodiments can be replaced or combined, so the present application can also be considered to include all possible combinations of the same and / or different embodiments described. Therefore, if one embodiment includes features A, B and C, and another embodiment includes features B and D, the present application should also be considered to include one or more embodiments of all other possible combinations of A, B, C and D, although the embodiment may not be explicitly described in the following content.

[0026] The following description provides examples, and is not intended to limit the scope, applicability or examples set forth in the claims. Alterations and changes in the function and arrangements of elements can be made without departing from the scope of the application. Various examples can omit, substitute, or add various procedures or components as appropriate. For instance, the methods described can be performed in an order different than described, and various steps can be added, omitted, or combined. Also, features described with respect to some examples can be combined in other examples.

[0027] Embodiment one: as Figure 1 shown, Figure 1 is a flowchart of a pedestrian traffic accident injury reconstruction method based on video images, which is only one embodiment of the present application, a pedestrian traffic accident injury reconstruction method based on video images, the method comprising the following steps: S1: screening pedestrian traffic accidents from the traffic accident database, and obtaining video image data of the pedestrian traffic accidents; In the traffic accident database, the data screening and image recognition technology are comprehensively used to automatically screen the pedestrian traffic accidents meeting the conditions as the data set for the accident deep injury reconstruction. The data of the accident case number, scene information, vehicle information, personnel information, collision information and the like meeting the screening conditions in the traffic accident data system are uploaded to the accident deep injury reconstruction system, and the uploaded information is marked as original accident information.

[0028] When step S1 is performed, specifically: S11: screening pedestrian traffic accidents from the traffic accident database; When step S11 is performed, pedestrian traffic accidents are screened from the traffic accident database. There are many kinds of traffic accidents, such as vehicle and object collision accidents, vehicle and vehicle collision accidents, and vehicle and pedestrian collision accidents. It is necessary to comprehensively use data screening and video analysis technology to screen the accident cases meeting the conditions, i.e. pedestrian collision traffic accidents, from the database of all traffic accidents, and the screening conditions include: 1) pedestrian and vehicle contact collision; 2) at least one injury of the pedestrian, preferably with detailed injury information recorded; 3) collision accident occurrence video, preferably video images of the whole process of the accident occurrence; Only the traffic accident data meeting the above three conditions can be extracted; S12: judging whether the completeness of the pedestrian traffic accident data is higher than a first preset threshold value; if yes, step S13 is performed; otherwise, no operation is performed. Here is the judgment of the structured information of the pedestrian accident data, read the structured information form of the pedestrian traffic accident data, obtain the accident data, the accident data contains field information, vehicle information, personnel information and collision information; Then judge whether the completeness of the accident data is higher than the first preset threshold value by using the dynamic threshold method based on the key field missing rate; The main is to judge the completeness of the following information: Key field information definition: field information: number of participants, participant type, vehicle type, road adhesion coefficient, accident scene sketch; Vehicle information: vehicle model, vehicle length, vehicle width, vehicle height, hood front edge height BLEH, hood angle BA, hood length BL, bumper ground clearance height BBH, bumper upper end ground clearance height BUH, bumper lower end ground clearance height BLH, bumper depth BUD, windshield angle WA, vehicle deformation position, vehicle deformation size, etc.; Personnel information: gender, age, height, weight, injury site, and AIS of each part of the pedestrian; Collision information: vehicle speed, whether to brake, whether to accelerate, whether to turn, vehicle deceleration / acceleration, pedestrian walking speed, throw distance, person-vehicle contact area, WAD, first collision point coordinates, etc.; The judgment condition for the completeness of the pedestrian traffic accident data is: 1. Let Ri be the key field missing rate of i-dimensional (i=1, 2, 3, 4; corresponding to field information, vehicle information, personnel information, and collision information 4 dimensions). The missing rate calculation formula is: Ri=R Q / R z , where R Q is the number of missing key fields in this dimension; R z is the total number of key fields in this dimension; 0≤Ri≤1, 0 represents no missing, and 1 represents all missing; 2. Set the grading threshold and rules: complete level threshold ai, when Ri=ai, it is judged as complete level; When ai 3. Global judgment rule: there is at most 1 dimension in failure level, and at most 2 dimensions in completable level, it is judged as overall failure, the accident cannot be used for reconstruction, otherwise the accident can be used for reconstruction.

[0029] It should be noted that the information dimension of the local failure level and the completable level can be supplemented according to the traffic accident original information in the traffic accident database, and the information can be re-evaluated.

[0030] In this way, only the traffic accidents that can be used for reconstruction can extract their video image data.

[0031] S13: Obtain the video image data of the pedestrian traffic accident that satisfies the three conditions of the accident type and the integrity is higher than the preset threshold.

[0032] It should be noted that when obtaining the video image data of the pedestrian traffic accident, the video needs to be analyzed.

[0033] If there is an accident process video, the YOLO target detection tool and the trajectory analysis algorithm are used to automatically extract the collision key moment picture information, specifically: 1) Video preprocessing: The video is disassembled by frame, and each frame image is normalized; 2) Target detection and tracking: The target detection model YOLO is used to detect pedestrians and vehicles in the image, and the DeepSORT algorithm is used to assign a unique ID to each detected target, continuously tracking its position and motion trajectory; 3) According to whether the overlap rate of pedestrians and vehicles changes continuously, the height of the pedestrian motion trajectory, the speed of the vehicle and other key parameters, judge the five collision key moments of pedestrians and vehicles, respectively defined as: collision before T -1 , contact moment T0, separation moment T1, pedestrian air turning moment T2, pedestrian landing moment T 3； 4) Key moment extraction and verification: Extract the corresponding frame, and use the CNN classification model to perform secondary verification on the extracted frame to ensure the logical correctness of the key moment.

[0034] Finally, upload the accident case number, accident structured information, and accident collision key moment picture of the accident case in the traffic accident database that meets the screening condition to the accident depth damage reconstruction system, generate the same accident number in the accident depth damage reconstruction system, and mark the uploaded information as the original information of the accident.

[0035] Steps S2 to S5 are performed by the accident depth damage reconstruction system.

[0036] S2: Obtain the pedestrian posture parameters from the video image data of the pedestrian traffic accident; Unlike vehicle collision simulation reconstruction, pedestrians have more postures and positions, and different postures and positions result in different degrees of injury to the pedestrian. Therefore, in order to more effectively simulate the accident collision scene, it is necessary to obtain the posture parameters in the pedestrian traffic accident.

[0037] After the accident depth reconstruction system obtains the video image data of the pedestrian traffic accident as the original accident information, before performing step S2, the video image data of the pedestrian traffic accident needs to be preprocessed, including removing noise and adjusting the video resolution to a unified standard, so as to facilitate subsequent parameter recognition processing. The YOLO target detection and trajectory analysis algorithm is used to automatically extract the picture information at the key moment of collision and detect the pedestrian target.

[0038] The specific preprocessing method includes: ① Video preprocessing: the video is disassembled by frame, and each frame image is normalized; ② Target detection and tracking: the YOLO target detection model is used to detect pedestrians and vehicles in the image, and the DeepSORT algorithm is used to assign a unique ID to each detected target, and continuously track its position and motion trajectory.

[0039] ③ Key moment logical judgment: according to whether the overlap rate of pedestrians and vehicles changes continuously, the height of the motion trajectory of pedestrians, the speed of vehicles and other key parameters, five key moments of collision between pedestrians and vehicles are judged, which are respectively defined as: T -1 , contact moment T0, separation moment T1, pedestrian air turning moment T2, and pedestrian landing moment T3.

[0040] ④ Key moment extraction and verification: the corresponding frame is extracted, and a CNN classification model is used to verify the extracted frame again to ensure the logical correctness of the key moment.

[0041] In the acquisition of pedestrian posture parameters, there are two cases of video image data obtained by single camera shooting and video image data obtained by multiple camera shooting, and the acquisition methods of pedestrian posture parameters in the two cases are different, as follows: The first case: If the video image data is obtained by single camera shooting, the video image data is fused by using the SMPL-X human body model and the NeRF neural radiation field fusion algorithm, and the three-dimensional joint angle of the human body is solved by time sequence optical flow constraint to obtain the pedestrian posture parameters, which are as follows: 1) Improve the SMPL-X model, introduce more detailed features and optimized parameter settings. Construct a neural radiation field (NeRF), use monocular camera shooting video image, use multi-layer perception (MLP) to render the human body in the scene, learn the geometric shape and appearance characteristics of the human body, and establish the three-dimensional information of the human body. Fuse the improved SMPL-X model with NeRF, use the three-dimensional human body geometric information generated by NeRF as the initial shape and position reference of the SMPL-X model, and realize the initialization of the model; 2) Utilize the YOLO-X target detection algorithm to detect pedestrian targets in monocular camera video images. Use a human key point detection network such as HRNet to extract two-dimensional key points of each part of the pedestrian's body, such as head, joint points, etc. At the same time, extract image features around the key points, such as color, texture, etc., to provide data support for subsequent three-dimensional reconstruction; 3) Based on the constructed NeRF model, train the images under different angles of view of the monocular video to optimize the network parameters of NeRF, so that it can accurately reconstruct the three-dimensional geometric shape of the human body. In the reconstruction process, use the two-dimensional key points and their surrounding image features as constraint conditions to guide NeRF to generate a three-dimensional model that conforms to the actual human body shape, improving the accuracy of reconstruction.

[0042] 4) Based on the obtained video sequence, calculate the optical flow field between adjacent frames to obtain the motion information of each part of the pedestrian's body in the time dimension. Combine the optical flow information with the improved SMPL-X model, and use the optical flow constraint as the optimization condition to adjust the joint angle parameters of the SMPL-X model by solving the optimization problem, so that the motion of the model matches the actual motion of the pedestrian in the video, thereby calculating the three-dimensional joint angles. In the optimization process, the Adam optimization algorithm is used to iteratively update the model parameters until the preset convergence condition is met; the calculated three-dimensional joint angles of the human body are used as the pedestrian posture parameters.

[0043] The second case: If the video image data is obtained by two or more multi-angle cameras, the video information of each camera image is extracted, and the human body in different angle videos is detected and tracked synchronously to ensure that the two-dimensional key point information of the human body under different angles is obtained at the same time. Get the pedestrian posture parameters; Specifically: 1) Stereo matching stage: Use a stereo matching algorithm based on deep learning, such as PSMNet, to process each frame of image of different angle videos. By extracting feature points in the image, calculating the disparity of feature points between different angle images, and obtaining the depth information of each point in the two-dimensional image. Disparity calculation is based on the epipolar constraint principle, searching for matching points on the corresponding epipolar line of different angle images to determine the distance of the object in space; 2) Voxel fusion stage: Convert the obtained depth information into voxel representation in three-dimensional space. Set an appropriate voxel resolution to balance the calculation efficiency and accuracy. For each voxel, according to its depth information and visibility under different angles, determine its final attribute value (such as color, density, etc.) through weighted averaging and other fusion strategies. With the advancement of the video, the voxel information is constantly updated, and a point cloud model containing the motion trajectory of the pedestrian is gradually constructed; 3) Posture parameter acquisition stage: using point cloud processing algorithm, such as curvature-based feature extraction algorithm, to extract feature points of each part of the human body from the constructed pedestrian motion trajectory point cloud. By analyzing the spatial position relationship of these feature points, combined with the human body kinematics model, the angle parameters of each joint of the human body, such as the angle of head-neck joint θ H-N , pelvis joint θ Plvis , knee joint θ Knee , etc. are calculated as pedestrian posture parameters.

[0044] S3: Constructing vehicle model and pedestrian model of traffic accident; It should be noted that the vehicle model can be constructed directly from the processed video image data of the traffic accident obtained in step S1, and the pedestrian model needs to be processed in multiple steps from the video image data of the traffic accident obtained in step S1 and the pedestrian posture parameters of step S2.

[0045] When performing step S3, the vehicle model and pedestrian model of the traffic accident are constructed, specifically: S31: Constructing vehicle model, according to the scanning point cloud or proxy model, and the vehicle information in the pedestrian traffic accident data, adjusting the parameters to construct the vehicle model; Here, the key parameters of the accident vehicle and the FE vehicle model are compared, and the height of the hood front edge BLEH, the hood angle BA, the hood length BA, the bumper ground clearance GC, and the windshield angle WA are set as the basis for determining whether the accident vehicle and the FE vehicle match. If the difference between the five parameters of the accident vehicle and the FE vehicle is within the preset error range, for example, within 10%, it is considered that the FE vehicle model is available; S32: Constructing pedestrian model, constructing pedestrian initial model according to pedestrian information in pedestrian traffic accident data; According to the gender and body size of the pedestrian in the accident, the THUMS or AC-HUMS human body model is called to match, the pedestrian posture parameters (including the position, angle parameters of each body part in three-dimensional space) identified are input into the.k file of LS-DYNA software, and the cards of the position, posture and size parameters of the pedestrian model are adjusted to quickly position the position and basic posture of the pedestrian model (dummy), this process is only used for adjustment without grid distortion; S33: According to the pedestrian posture parameters, determine the target angle of each joint of the pedestrian initial model and the target position of each body part, apply gradually changing load to the pedestrian initial model in the finite element software, adjust the finite element pedestrian initial model to the state consistent with the pedestrian posture parameters, and get the pedestrian model.

[0046] To prevent the grid cell distortion and penetration phenomenon in the posture adjustment process of the pedestrian model (dummy), the slow loading method is used for posture adjustment of the dummy. First, according to the input posture parameters, the target angles of each joint of the dummy and the target positions of each body part are determined. Then, in the finite element software, a script program is written to apply a gradually changing load to the dummy model. For example, for the adjustment of the joint angle, a torque load is applied at the joint, and the torque gradually increases over time, so that the joint gradually rotates to the target angle. For the adjustment of the body part position, a displacement load is applied at the corresponding part, and the displacement gradually increases over time, so that the body part is moved to the target position. Then, during the loading process, the mass parameters of the grid cells are monitored in real time, such as the aspect ratio and distortion of the cells. When it is detected that some cells are distorted or may be penetrated, the loading speed and load distribution are automatically adjusted to ensure that the deformation of the grid cells is within a reasonable range. Through this slow loading method, the finite element dummy is gradually adjusted to a state consistent with the identified pedestrian posture.

[0047] S4: simulate the collision process between the vehicle and the pedestrian, and calculate the kinematic parameters and damage parameters of the pedestrian; After the pedestrian and vehicle finite element models are constructed by the accident deep damage reconstruction system, the collision process between the pedestrian and the vehicle in the traffic accident needs to be simulated and reconstructed, so as to calculate the damage degree of the accident.

[0048] After the posture and position of the vehicle model and the pedestrian model are adjusted, a multi-objective optimization calling program is used to call LS-DYNA for calculation and solving. During the solving process, the software simulates the collision process between the vehicle and the pedestrian in the accident, and calculates the kinematic parameters and damage parameters of the pedestrian.

[0049] When step S4 is executed, the collision process between the vehicle and the pedestrian is simulated, and the kinematic parameters and damage parameters of the pedestrian are calculated, specifically as follows: After the posture and position of the vehicle model and the pedestrian model are adjusted, a MATLAB or Python program or an iSIGHT optimization software is used to automatically call LS-DYNA for calculation and solving. During the solving process, the software simulates the collision process between the vehicle and the pedestrian in the accident, and calculates the kinematic parameters and damage parameters of the pedestrian.

[0050] The kinematic parameters of the pedestrian include head angular acceleration, HIC 15 , chest compression, R max , chest viscosity index VC, femoral bending moment, femoral upper / lower cross-sectional force, iliac lateral compression force, knee ligament stretch, and tibial bending moment; The damage parameters include skull strain distribution, intracranial pressure distribution, rib cortical bone strain, internal organ stress, pressure, femur strain, pelvic strain, tibia strain, and fibula strain; In the simulation of the accident, the vehicle speed, the vehicle deceleration, the pedestrian speed, the pedestrian joint parameters, and the friction coefficient between the participants are used as the design variables. The contact position between the different parts of the pedestrian and the vehicle is set as the constraint condition. The final landing position of the pedestrian and the relative position between the pedestrian and the vehicle are used as the objective function.

[0051] After the initial parameter input and reconstruction are completed, the program or software automatically reads the position information and damage parameters of the pedestrian from the solution. The contact position between the pedestrian and the vehicle is set as the constraint condition. The final landing position of the pedestrian and the relative position between the pedestrian and the vehicle are used as the objective function.

[0052] When the initial parameters are input, the genetic algorithm is also used to automatically optimize the input parameters. The specific steps are as follows: 1. Population initialization: A set of initial input parameter combinations is generated. The vehicle speed, the vehicle deceleration, the pedestrian speed, the pedestrian posture, and the collision position are set according to the parameters extracted and calculated in the video. The friction coefficient is set according to the default value. The population size and the number of iterations are set. 2. Fitness calculation: The parameter combination of each individual is input into LS-DYNA for solving. The mean square error (MSE) of the simulation trajectory at the key moment of the collision is calculated. If the contact position is not consistent, a penalty term is applied. 3. Genetic operators: Selection operation: select excellent individuals according to the fitness value, eliminate individuals with low fitness, and retain high-quality genes. Cross operation: cross the selected individuals to generate new individuals, increasing the diversity of the population. Mutation operation: randomly mutate the parameters of the individuals to avoid falling into local optimal solutions. After selection, cross, and mutation operations, a new population is generated, and the solving and fitness calculation are performed again. 4. Optimization cycle: iterate until the convergence condition is met (the fitness value does not change significantly for several consecutive generations) or the maximum number of iterations is reached, and the calculation is stopped. The optimal input parameter combination is obtained, and the precise accident reconstruction including kinematics and human body damage is realized. 5. Post-processing and verification: perform high-precision simulation on the optimal individual, check the contact position, landing site, and physical rationality.

[0053] After calculating the kinematic parameters and damage parameters of the pedestrian, a damage reconstruction database needs to be constructed to store the structured (boundary parameter setting, pedestrian posture parameter, motion parameter) and unstructured information (model, animation, acceleration curve, stress map, etc.) of the damage reconstruction.

[0054] S5: Based on the time sequence, the credibility of the damage reconstruction result is quantitatively evaluated, and a damage reconstruction database is constructed. When step S5 is performed, the credibility of the damage reconstruction result is evaluated based on the time sequence comprehensive quantification, specifically: whether there is a difference between the simulation result of the key moment of the pedestrian and vehicle collision and the core joint parameters of the pedestrian in the accident video image, the basic credibility is exponentially attenuated, the credibility rating is formed, and whether the credibility is higher than the second preset threshold is judged, if yes, the reconstruction result is highly credible, and the damage reconstruction database is constructed; otherwise, the reconstruction result is basically credible, and is used as auxiliary data and comparison data to construct the damage reconstruction database. Wherein, the pedestrian core parameter comparison includes: the head and neck joint angle θ -1 , the pelvic joint angle θ Plvis , and the knee joint angle θ Knee of the pedestrian at the contact moment T0, the separation moment T1, the pedestrian air turning moment T2, and the pedestrian landing moment T3. H-N Plvis Knee .

[0055] The credibility evaluation is specifically as follows: 1. Evaluation object The key moment of the collision: the head and neck joint angle θ -1 , the pelvic joint angle θ Plvis , and the knee joint angle θ Knee of the pedestrian at the contact moment T0, the separation moment T1, the pedestrian air turning moment T2, and the pedestrian landing moment T3.

[0056] Each moment needs to verify 3 joint angles: the head and neck joint angle θ H-N , the pelvic joint angle θ Plvis , and the knee joint angle θ Knee . 2. Attenuation rule For each joint parameter that does not match in the simulation picture and the video picture, the total credibility is attenuated by 2%.

[0057] Credibility = 100% x 0.98 n , wherein n = the total number of joint parameters that do not match; 3. Determination threshold When the credibility is greater than 90%, it is determined to be highly credible (i.e. the second preset threshold is 90%); otherwise, it is marked as basically credible.

[0058] In summary, the accident deep damage reconstruction system of steps S2 to S5 uses intelligent simulation operation to determine whether the consistency of the key parameters of the vehicle finite element model meets the requirements, according to the determination result, selects the vehicle finite element model, and automatically reads the damage reconstruction result, improves the accuracy and efficiency of the damage reconstruction.

[0059] The application discloses a pedestrian traffic accident injury reconstruction method based on video images.

[0060] Embodiment two, only one embodiment of the application, the application also provides a pedestrian traffic accident injury reconstruction system based on video images using the method of embodiment one, comprising: A data screening module; A posture parameter acquisition module; A model construction module; A simulation reconstruction module; A database construction module; The data screening module screens pedestrian traffic accidents from a traffic accident database and obtains video image data of the pedestrian traffic accidents; the posture parameter acquisition module obtains pedestrian posture parameters from the video image data of the pedestrian traffic accidents; the model construction module constructs vehicle models and pedestrian models of the traffic accidents; the simulation reconstruction module simulates a collision process between the vehicles and the pedestrians, calculates kinematic parameters and injury parameters of the pedestrians; and the database construction module evaluates the credibility of injury reconstruction results based on time sequence and comprehensive quantization, constructs an injury reconstruction database, and stores structured (boundary parameter setting, pedestrian posture parameters, motion parameters) and unstructured information (models, animations, acceleration curves, stress maps, etc.) of the injury reconstruction.

[0061] It should be noted that the pedestrian traffic accident injury reconstruction system based on video images provided in the application works to realize the pedestrian traffic accident injury reconstruction method based on video images provided in one of the implementation manners of embodiment one.

[0062] Embodiment three, the application also provides a device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor realizes the pedestrian traffic accident injury reconstruction method based on video images provided in one of the implementation manners of embodiment one when executing the computer program.

[0063] Embodiment four, the application provides a computer storage medium, the computer storage medium stores a computer program, the computer program comprises program instructions, and the program instructions can realize the pedestrian traffic accident injury reconstruction method based on video images provided in one of the implementation manners of embodiment one when executed by a processor.

[0064] It should be noted that, for the foregoing method embodiments, the sequences of the described actions can be modified, and certain actions can be performed simultaneously or in different sequences. In addition, the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0065] In the above embodiments, the description of each embodiment focuses on different aspects, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0066] In the several embodiments provided by the present application, it should be understood that the disclosed apparatus can be implemented in other manners. For example, the division of the described units is only a logical function division, and there can be another division manner in actual implementation. For example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, or the among different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be in electrical, mechanical or other forms.

[0067] The units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0068] In addition, each functional unit in the embodiments of the present application can be integrated in a processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware or software functional units.

[0069] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable memory. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a memory and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned memory includes: a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0070] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be instructed by a program to be completed by relevant hardware, and the program can be stored in a computer readable memory, which can include a flash disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0071] The above is only exemplary embodiments of the present disclosure, and cannot limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure are still within the scope of the present disclosure. Those skilled in the art will easily think of embodiments of the present disclosure after considering the specification and practicing the disclosure herein. The present application is intended to cover any variations, uses or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and examples are only considered as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for pedestrian traffic accident injury reconstruction based on video images, characterized in that, Includes the following steps: S1: Filter pedestrian traffic accidents from the traffic accident database and obtain video image data of pedestrian traffic accidents; S2: Obtain pedestrian posture parameters from video image data of pedestrian traffic accidents; S3: Construct vehicle and pedestrian models for traffic accidents; S4: Simulate and reconstruct the collision process between a vehicle and a pedestrian, and calculate the kinematic parameters and injury parameters of the pedestrian; S5: Based on time-series comprehensive quantitative evaluation, evaluate the credibility of damage reconstruction results and construct a damage reconstruction database.

2. The method for pedestrian traffic accident damage reconstruction based on video images according to claim 1, characterized in that: When performing step S1, the specific steps are as follows: S11: Filter pedestrian traffic accidents from the traffic accident database; S12: Determine whether the completeness of pedestrian traffic accident data is higher than the first preset threshold; if yes, proceed to step S13; otherwise, do not perform the operation. S13: Acquire video image data of pedestrian traffic accidents.

3. The method for pedestrian traffic accident damage reconstruction based on video images according to claim 2, characterized in that: When performing step S11, pedestrian traffic accidents are filtered from the traffic accident database. The filtering criteria include: A pedestrian comes into contact with or collides with a vehicle; The pedestrian sustained at least one injury; There is video footage of the collision.

4. The method for pedestrian traffic accident damage reconstruction based on video images according to claim 2, characterized in that: When performing step S12, the specific steps are as follows: Read the structured information form of pedestrian traffic accident data to obtain accident data including scene information, vehicle information, personnel information, and collision information; A dynamic thresholding method based on the missing rate of key fields is used to determine whether the completeness of accident data is higher than a first preset threshold. If yes, then proceed to step S13; otherwise, do not proceed.

5. The method for pedestrian traffic accident damage reconstruction based on video images according to claim 1, characterized in that: When performing step S2, pedestrian posture parameters are obtained from the video image data of pedestrian traffic accidents, specifically: If the video image data is captured by a single camera, the video image data is fused with the SMPL-X human body model and the NeRF neural radiation field fusion algorithm, and the three-dimensional joint angles of the human body are calculated through temporal optical flow constraints to obtain the pedestrian posture parameters. If the video image data is acquired by multiple cameras, the video information of each camera image is extracted, and the human body in the video from different angles is detected and tracked simultaneously to ensure that the two-dimensional key point information of the human body from different perspectives is acquired at the same time, so as to obtain the pedestrian posture parameters.

6. The method for pedestrian traffic accident damage reconstruction based on video images according to claim 1, characterized in that: When performing step S3, construct the vehicle model and pedestrian model for the traffic accident, specifically as follows: S31: Construct a vehicle model by adjusting parameters based on the vehicle information in the scanned point cloud or proxy model and pedestrian traffic accident data. S32: Construct a pedestrian model. Based on pedestrian information in the pedestrian traffic accident data, construct an initial pedestrian model. S33: Based on the pedestrian posture parameters, determine the target angles of each joint and the target positions of each body part of the initial pedestrian model. Apply gradually changing loads to the initial pedestrian model in the finite element software to adjust the finite element initial pedestrian model to a state consistent with the pedestrian posture parameters, thus obtaining the pedestrian model.

7. The method for pedestrian traffic accident damage reconstruction based on video images according to claim 1, characterized in that: When performing step S4, the collision process between the vehicle and the pedestrian is simulated and reconstructed, and the kinematic parameters and injury parameters of the pedestrian are calculated, specifically: After adjusting the posture and position of the vehicle and pedestrian models, the multi-objective optimization program calls LS-DYNA to perform calculations and solutions. During the solution process, the software simulates the collision process between the vehicle and the pedestrian in the accident and calculates the kinematic parameters and injury parameters of the pedestrian.

8. The method for pedestrian traffic accident damage reconstruction based on video images according to claim 7, characterized in that: When performing step S4, Pedestrian kinematic parameters include head angular acceleration, HIC 15 Chest compression volume, R max Chest viscosity index (VC), femoral bending moment, femoral upper / lower section force, iliac bone lateral compression force, knee joint ligament tension, and tibial bending moment; Damage parameters include skull strain distribution, intracranial pressure distribution, cortical bone strain of each rib, visceral stress, pressure, femoral strain, pelvic strain, tibia strain, and fibular strain. When simulating accidents, vehicle speed, vehicle deceleration, pedestrian speed, pedestrian joint parameters, and friction coefficient between the parties involved are used as design variables; The constraints are set at the contact positions between different parts of the pedestrian and the vehicle, and the objective functions are the final landing position of the pedestrian and the relative position between the pedestrian and the vehicle.

9. A method for pedestrian traffic accident damage reconstruction based on video images according to claim 1, characterized in that: When executing step S5, the credibility of the damage reconstruction result is evaluated based on the temporal comprehensive quantitative assessment. Specifically, the simulation results at the critical moment of the collision are compared with the pedestrian core joint parameters in the accident video image to determine if there are any differences. The basic credibility is decayed exponentially to form a credibility rating. When the credibility is higher than the second preset threshold, the reconstruction result is considered credible, and a damage reconstruction database is constructed. Among them, the comparison of pedestrian core parameters includes: T before collision -1 The pedestrian's head and neck joint θ at five time points: contact time T0, separation time T1, pedestrian's mid-air flip time T2, and pedestrian's landing time T3. H-N pelvic joint θ Plvis , knee joint θ Knee angle.

10. A pedestrian traffic accident injury system based on video images, characterized in that, include: Data filtering module; Attitude parameter acquisition module; Model building module; Simulation reconstruction module; Database building module; The data filtering module filters pedestrian traffic accidents from the traffic accident database and obtains video image data of pedestrian traffic accidents; the posture parameter acquisition module obtains pedestrian posture parameters from the video image data of pedestrian traffic accidents. The model building module constructs vehicle and pedestrian models for traffic accidents; the simulation and reconstruction module simulates and reconstructs the collision process between vehicles and pedestrians, and calculates the kinematic parameters and injury parameters of pedestrians. The database construction module evaluates the reliability of damage reconstruction results based on time-series comprehensive quantitative assessment and constructs a damage reconstruction database.

Images