Pedestrian-ground collision injury protection method based on pedestrian rotation angle prediction
Through the intelligent vehicle perception and prediction module, predicting the rotation angle of pedestrians and controlling vehicle braking, the problem of difficulty in accurately predicting medium and low-speed collision damage in the prior art is solved, and the effect of reducing damage in human-ground collisions is achieved.
Patent Information
- Application Number
- CN202411303717.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-18
- Publication Date
- 2024-12-27
AI Technical Summary
In human-vehicle collisions, the prior art is difficult to accurately predict ground-related damage to pedestrians in medium- and low-speed collisions, resulting in vehicle braking in some cases increasing human-ground collision damage.
Data is collected through the intelligent vehicle perception module, the pedestrian rotation angle is predicted, the rotation angle interval is determined using the prediction model, and the vehicle braking is controlled according to the prediction results to reduce damage. If it is a safe angle, the pedestrian rotation is adjusted if it is a dangerous angle. Angle to reduce damage.
It effectively avoids the increased damage risk of vehicle braking when there is less human-ground collision damage, and reduces human-ground collision damage by predicting the rotation angle and controlling the braking, thereby improving the protection effect.
Smart Images
Figure CN119189923A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of automobile traffic safety, and in particular to a pedestrian-ground collision damage protection method based on pedestrian rotation angle prediction. Background Art
[0002] According to existing research, in a collision between a pedestrian and a vehicle, injuries to pedestrians usually come from the first collision between the pedestrian and the vehicle and the second collision between the pedestrian and the ground. A large number of studies have shown that pedestrian injuries caused by vehicles are often more serious, but some studies have also found that ground-related injuries in low- and medium-speed collisions cannot be ignored.
[0003] There are three main methods for protecting people from collisions with vehicles. The first is the design of the front end of the vehicle, such as pedestrian-friendly front and intelligent engine hood. Although this type of design takes into account the collision damage between people and vehicles at medium and low speeds, its main purpose is still to reduce the collision damage between people and vehicles. The second is to use external pedestrian protection devices to hinder or even prevent the occurrence of collisions with people and vehicles, such as equipping the front of the vehicle with mechanical capture devices, external airbag devices and adhesives. The core idea is to hinder or prevent people from falling to the ground, but its use in actual accidents has not been verified by simulation. The third is to control the braking of the vehicle so that the vehicle affects the landing angle of the human body and thus reduces the collision damage between people and vehicles. So far, the simulation results show that the collision damage between people and vehicles can be reduced by nearly 90% under ideal conditions. Currently, many patents have adopted the third method to protect people from damage caused by collision with the ground, for example: CN116901894A a protective device and method for protecting people from damage caused by collision with the ground, CN112918469A a pedestrian automatic emergency braking system with active and passive safety functions, CN112918421A a vehicle motion control method combined with airbags to protect people from damage caused by collision with the ground, etc., which shows that this solution has good application prospects.
[0004] Nevertheless, the vehicle control braking method also has its drawbacks: in some cases where full braking can reduce the damage caused by human-ground collision, the use of controlled vehicle braking will increase the damage caused by human-ground collision. This point is also mentioned in articles such as Zou Tiefang et al.'s vehicle braking control method for reducing human-ground collision damage and a review of human-ground collision damage protection in human-vehicle collision accidents. Therefore, if it is possible to accurately predict the damage caused by human-ground collision to pedestrians, and use vehicle control braking when the damage caused by human-ground collision is serious, the risk of using this method can be reduced. However, most existing studies have shown that it is extremely difficult to predict ground injuries, and Han Yong's study on the analysis of pedestrian kinematic responses in automobile collision accidents based on video information also shows that human-ground collision damage is related to the rotation angle of pedestrians when they contact the ground. For this reason, the present invention will propose a human-ground collision damage protection method based on the prediction of pedestrian rotation angle, first predicting the pedestrian rotation angle, then linking the relationship between the pedestrian rotation angle and human-ground collision damage, and finally deciding whether to control vehicle braking at this angle based on the degree of damage. Summary of the invention
[0005] The present invention designs a pedestrian-ground collision damage protection method based on pedestrian rotation angle prediction, which is used to protect people from damage caused by secondary collision with vehicles.
[0006] In order to achieve the above and related purposes, the present invention provides a method for predicting and controlling the rotation angle of a pedestrian after a collision between a pedestrian and a vehicle. The specific method comprises the following steps:
[0007] S1: Collecting information about the collision between a car and a pedestrian. When the smart car finds that the collision between a car and a pedestrian is unavoidable, the smart car obtains the vehicle speed, the pedestrian's hip height, the engine hood height, the time of the first contact between the car and the pedestrian, the time of the first contact between the head and the vehicle, the head acceleration when the lead vehicle contacts the head, and the head acceleration when the lead vehicle separates from the head through the perception module, and uploads this data to the vehicle prediction module.
[0008] S2: Predict the pedestrian's rotation angle. The vehicle prediction module first makes a preliminary prediction of the pedestrian's rotation angle range using the pedestrian rotation angle prediction rule based on the data from the perception module, and then further predicts the pedestrian's rotation angle after the collision using the pedestrian rotation angle prediction model. Finally, the prediction result is input into the vehicle decision module.
[0009] S3: Make a decision based on the prediction results. After receiving the information from the car prediction module, the decision module controls the car to make different responses according to different prediction results. If the prediction result is a safe rotation angle, the vehicle is fully braked until it stops. If the prediction result is a dangerous rotation angle, the vehicle is controlled to brake to allow the vehicle to continue moving forward and change the pedestrian's rotation angle.
[0010] Preferably, the functions of each module of the intelligent vehicle pedestrian protection system according to the present invention are as follows:
[0011] Before and when a collision occurs, the perception module collects the vehicle speed v, pedestrian hip height h1, engine hood height h2, first contact time of human-vehicle collision, first contact time of human head and vehicle, head acceleration when the lead vehicle contacts, and head acceleration when the lead vehicle separates through the front camera and radar of the car, and then uploads all the information to the car prediction module; after receiving the data, the prediction module first makes a preliminary prediction of the rotation angle range of the collision according to the prediction rules, and then further confirms the pedestrian rotation angle through the pedestrian rotation angle prediction model, and uploads the final prediction result to the decision module; the decision module makes decisions based on the prediction results fed back by the prediction module to control the car to take different measures in different situations. If it is a dangerous pedestrian rotation angle, the vehicle is braked to make the vehicle continue to move forward, thereby changing the pedestrian rotation angle to protect the pedestrian; if it is a safe pedestrian rotation angle, the vehicle is completely braked until the vehicle stops.
[0012] Preferably, the car perception module includes a camera, a car radar, etc. The camera is installed at the connection between the car hood and the windshield, and is used to collect vehicle speed v, pedestrian hip height h1, hood height h2, time of human-vehicle collision, head acceleration when the head vehicle contacts, and head acceleration when the head vehicle separates. The radar is installed at the front end of the vehicle to collect the real-time distance between the pedestrian and the car; in order to ensure the accuracy of the prediction results, the sampling step of the camera is not less than 0.001 seconds, and the camera is a conventional camera. The coordinate system is required to determine the ratio h1 / h2 of the pedestrian hip height h1 and the vehicle hood height h2 before the human-vehicle collision. The zero point is on the ground and the vertical is positive. The car prediction module includes prediction rules and prediction models. The prediction rules come from the prediction model obtained through research and collation in the paper of Zou Tiefang et al.; the prediction model first collects accident data, and then classifies and predicts the factors affecting the pedestrian rotation angle, and then uses the prediction results as the input features of the second stage to finally obtain the prediction model. The car decision module is connected to the car braking system and issues different commands to the braking system in response to different prediction results.
[0013] Preferably, the prediction module preliminarily predicts the pedestrian's rotation angle using the following prediction rules:
[0014] Rule 1: When h1 / h2<1.2 or v<20km / h, the pedestrian's motion trajectory after the collision is judged to be a forward projection trajectory, and the pedestrian's rotation angle is in the interval A[0°, 90°]. In this case, it is considered that the pedestrian's rotation angle must be a safe rotation angle, and the pedestrian rotation angle prediction model does not need to perform a secondary prediction;
[0015] Rule 2: When h1 / h2>1.2 and v>20km / h, the pedestrian's trajectory after the collision is judged to be a diffraction trajectory, and the pedestrian's rotation angle may be any angle in [0°, 360°]. In this case, it is impossible to determine whether the pedestrian's landing angle is safe, and further prediction model needs to be used for prediction;
[0016] Preferably, the construction process of the pedestrian rotation angle prediction model has the following four steps:
[0017] Step 1: First, collect 2,000 sets of pedestrian-vehicle collision accident data from the Internet. The accident case screening criteria are that only one Sedan / SUV is involved in the pedestrian-vehicle collision accident, the contact between the person and the vehicle occurs at the front of the vehicle, and the accident video must be able to observe the entire accident process to obtain pre-collision parameters such as vehicle speed, relative position of the person and the vehicle, and the pedestrian's rotation angle when the person and the ground are in contact. Then select 200 cases from these data to test the seven parameters of vehicle speed, pedestrian hip height, engine hood height, first contact time of pedestrian-vehicle collision, first contact time of the head and vehicle, head acceleration when the head vehicle contacts, and head acceleration when the head vehicle separates. They have a significant effect on the pedestrian's rotation angle, and then collect relevant data from the remaining 1,800 cases.
[0018] Step 2: Use K-Nearest Neighbors (K-NN), Support Vector Machine (SVM), Decision Tree (DT) and Random Forest (RF) as the first stage classification method to classify and predict the data after feature extraction, and then use K-fold cross validation to improve the reliability and learning ability of the prediction model. Using multiple classification methods at the same time can effectively improve the classification efficiency of data.
[0019] Step 3: Use the K-fold cross-validated data as the input features of the neural network to further learn and adjust to generate the final prediction results, thereby constructing a prediction model based on the multi-model ensemble method with seven parameters as input parameters, including vehicle speed, pedestrian hip height, engine hood height, first contact time of human-vehicle collision, first contact time of human head and vehicle, head acceleration when the lead vehicle contacts, and head acceleration when the lead vehicle separates, and the pedestrian rotation angle as the output result. The neural network method is adopted to find the weight relationship (approximate) between the input and output, and then use such a weight relationship for simulation.
[0020] Step 4: During the annual vehicle inspection, add no less than 500 sets of the latest human-vehicle collision accident data to update the prediction model and replace the existing model inside the car. In order to better deal with human-vehicle collisions in different situations, the prediction model needs to be continuously updated.
[0021] Preferably, after the pedestrian is hit, the pedestrian is rotated around the Y axis according to the pedestrian's normal standing angle of 0°, and the pedestrian is divided according to the final angle of the pedestrian. After referring to Shi Liangliang's research on the injury mechanism and risk of pedestrian head landing in car collision, the present invention is divided into four intervals: [0°, 90°], [90°, 180°], [180°, 270°], [270°, 360°], and the four rotation angle intervals are A, B, C, and D. When the pedestrian lands horizontally, the interval is divided according to whether the pedestrian's head is the first landing part. If the pedestrian's head is the first landing part, the landing posture should be matched to interval B or C. If the pedestrian's head is not the first landing part, the landing posture should be matched to interval A or D.
[0022] Analysis based on the landing site: In the A and D intervals, the pedestrian's lower limbs or pelvis collide with the ground first, followed by the torso, upper limbs and head. In these two rotation angle intervals, the pedestrian's head is not the first landing site, and the lower limbs, torso and other parts touch the ground first, which plays a certain buffering role in the head-to-ground collision. In the B and C intervals, the pedestrian's head is the first landing site, and there is no buffering protection from any other body parts before the head hits the ground, resulting in serious head injuries in the collision between the person and the ground.
[0023] Analysis based on the linear velocity of the pedestrian's head rotation: In the A and B intervals, the linear velocity of the head rotation is perpendicular to the ground and downward, which will cause the pedestrian's head to rotate toward the ground, increasing the injury to the pedestrian. In the C and D intervals, the linear velocity of the head rotation is perpendicular to the ground and upward, which will cause the head to rotate away from the ground.
[0024] If only the influence of the first landing part on the collision injury is considered, there is a greater risk of head injury in the direct collision between the head and the ground in intervals B and C because there is no protection from any other body parts. If only the influence of the rotation mechanism is considered, the collision between the head and the ground in intervals A and B is more severe. However, considering the motion response of pedestrians in real accidents, the influence of the first landing part and the rotation mechanism on the collision injury between people and the ground should be paid attention to at the same time. Comprehensively considering the two factors of the first landing part and the rotation mechanism when pedestrians land, the characteristics of the head-ground collision injury risk in each pedestrian landing interval are clearly shown. It can be concluded that in interval B, the pedestrian's head is the first landing part, and the head rotation speed is downward, so it should be the most dangerous rotation angle interval; in interval C, the pedestrian's head is the first landing part, and the head rotation speed is upward, which is a dangerous rotation angle interval; in interval A, although the head rotation speed is downward, the pedestrian's lower limbs are the first landing part, which is a safe rotation angle interval; in interval D, the pedestrian's lower limbs are the first landing part, and the head rotation speed is upward, which is a safe rotation angle interval.
[0025] Preferably, the vehicle controlled braking method can only achieve its protective effect on pedestrians when the damage caused by the collision between the pedestrian and the ground is serious. In articles such as A vehicle braking control method for reducing damage caused by human-ground collision by Zou Tiefang et al. and A review of protection against damage caused by human-ground collision in human-vehicle collision accidents, it is proposed that in some cases where full braking can reduce the damage caused by the collision between the pedestrian and the ground, the use of the controlled vehicle braking method will increase the damage caused by the collision between the pedestrian and the ground.
[0026] Preferably, the time a for implementing vehicle control braking is in the range of [0.2, 1.6] s. This is because 0.2 s is the shortest coordination time of the braking system, and Zou Tiefang et al. have shown in a vehicle braking control method for reducing human-ground impact damage that 1.6 s after t1 is sufficient for the car to achieve high-quality braking control to reduce human-ground collision damage.
[0027] Preferably, for a specific vehicle, the method used to determine the time t2 at which the vehicle is fully braked again is as follows:
[0028] First, input the contour parameters of the front of different cars, including nine parameters: bumper lower height, bumper lower depth, bumper height, bumper upper height, hood front edge height, hood front edge depth, hood angle, hood length, and windshield angle. Then, establish its multi-rigid body model in the MADYMO simulation software, and divide the interval of [0.2, 1.6]s into 1000 equal parts. Each node represents a different t2 moment, and 1000 control braking methods with different control braking durations can be obtained. Finally, for each method, multiple simulations are carried out under the same vehicle speed, pedestrian height and pedestrian gait to obtain the method with the smallest WIC, which is also the method with the best comprehensive pedestrian protection effect, and the car control braking time a under the front model can be obtained.
[0029] Preferably, the 3 vehicle speeds, 4 pedestrian heights, and 2 pedestrian gaits in the virtual simulation system are: 3 vehicle speeds are 21, 31, and 41 km / h; 4 pedestrian heights are 5% percentile female, 5% percentile male, 50% percentile male, and 90% percentile male; 2 pedestrian gaits are 50% and 100%;
[0030] Preferably, WIC is a very scientific indicator proposed by Li et al. for comprehensive evaluation of human injuries, which includes injuries to the pedestrian's head, chest, pelvis, lower limbs and other parts. Obviously, the higher the WIC value, the greater the damage to the human body, and vice versa, the less damage to the human body. The WIC value W IC The calculation formula is as follows:
[0031]
[0032] where p i=p si ×p hi ×p gi ,∑p i =1; where: p si , p hi , p gi are the ratio of vehicle speed, pedestrian height and pedestrian gait, I Ci is the injury cost of the human body in the i-th simulation in a set of simulations.
[0033] Compared with the prior art, the technical solution of the present invention has the beneficial effect that: the method of the present invention can effectively avoid the situation where the damage caused by the collision between pedestrians and ground is small, and the increase of the damage caused by the collision between pedestrians and ground caused by the use of vehicle control braking. In the technical solution of the present invention, the rotation angle of the pedestrian in the collision is predicted by the vehicle prediction module to predict the degree of damage caused by the collision between pedestrians and ground, and the vehicle control braking is used to reduce the damage when the damage between pedestrians and ground is serious. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is a composition diagram of the pedestrian protection system of the present invention.
[0035] Figure 2 It is a flowchart of the present invention.
[0036] Figure 3 This is the prediction model building process of the present invention.
[0037] Figure 4 It is a diagram of the front end shape of a vehicle according to the present invention.
[0038] Among them, 1 is the height of the lower part of the bumper, 2 is the depth of the lower part of the bumper, 3 is the height of the bumper, 4 is the height of the upper part of the bumper, 5 is the height of the front edge of the engine hood, 6 is the depth of the front edge of the engine hood, 7 is the angle of the engine hood, 8 is the length of the engine hood, and 9 is the angle of the windshield. Specific implementation methods
[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] In order to reduce the secondary pedestrian-ground collision damage suffered by pedestrians in the event of an accident, a pedestrian-ground collision damage protection method is designed that predicts the pedestrian's rotation angle and then changes the pedestrian's dangerous rotation angle by controlling the brakes. The specific method steps include: first constructing a pedestrian rotation angle prediction model, then adding the prediction model to the vehicle prediction module and forming a complete pedestrian protection system with the perception module and decision module, and finally controlling the vehicle braking to change the pedestrian's rotation angle when the pedestrian-ground collision damage is serious.
[0042] When a collision between a person and a vehicle is about to occur, the process of the smart car protecting the person from damage to the ground is as follows:
[0043] When the vehicle perception module finds that a collision between a car and a pedestrian is unavoidable, it collects seven data, including vehicle speed, pedestrian hip height, engine hood height, first contact time of human-vehicle collision, first contact time of the human head and vehicle, head acceleration when the lead vehicle contacts, and head acceleration when the lead vehicle separates, and uploads all the data to the vehicle prediction module.
[0044] After the vehicle prediction module receives the data uploaded by the perception module, it first makes a preliminary prediction of the pedestrian's rotation angle after the collision based on the pedestrian rotation angle prediction rule. It is considered that the pedestrian trajectory of this collision is a diffraction trajectory and needs to further use the prediction model to predict the pedestrian's rotation angle. Then, all the data is imported into the pedestrian rotation angle prediction model to predict the pedestrian's rotation angle. The prediction result is that the pedestrian's rotation angle after this collision is in the B interval, which is a dangerous rotation angle. Finally, the prediction result is uploaded to the vehicle decision module.
[0045] When the vehicle decision module receives the information that the pedestrian's rotation angle is predicted to be a dangerous rotation angle, it immediately controls the vehicle to fully brake, and releases the car brakes at the moment t1 when the pedestrian's head contacts the vehicle to allow the vehicle to continue moving forward, thereby making secondary contact with the pedestrian and changing the pedestrian's rotation angle. Then, at the moment t2 when the pedestrian's rotation angle is considered to have reached a safe angle, the braking control is stopped and replaced with full braking until the vehicle is completely stationary.
[0046] For different types of cars, the stop control braking moment t2 is also different. For a common car, its t2 can be determined as follows: first, extract the contour parameters of the front of the car, including nine parameters: bumper lower height, bumper lower depth, bumper height, bumper upper height, hood front edge height, hood front edge depth, hood angle, hood length, and windshield angle, and build a corresponding front model. Then, let t2=t1+a, a is taken from the interval [0.2, 1.6]s, 1000 points are taken out, and a virtual simulation system that meets 3 vehicle speeds, 4 pedestrian heights and 2 pedestrian gaits is selected to conduct simulation experiments with the front model, and obtain 1000 different stop control braking moments t2 WIC values. When the WIC value is the smallest, the corresponding control braking time a is the optimal control braking time a0, and then the time t2 for canceling the control braking can be obtained from t2=t1+a0.
[0047] In another accident, due to the difference in collision data, the vehicle prediction module in this accident determined that the pedestrian's movement trajectory after the collision was forward projection, and the pedestrian would be in a safe rotation angle. After uploading the prediction results to the vehicle decision module, the decision module controls the vehicle to fully brake until the vehicle is completely stationary to prevent the pedestrian from being crushed and causing other injuries.
[0048] Obviously, the above embodiments of the present invention are merely examples for clearly explaining the present invention, and are not intended to limit the implementation methods of the present invention. Anyone familiar with the technology may modify or change the above embodiments without violating the scope of the present invention. Therefore, any modification, equivalent substitution, and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the claims of the present invention.
Claims
1. A pedestrian-ground collision damage protection method based on pedestrian rotation angle prediction, characterized in that: First, the information of pedestrians and vehicles before and during collision is collected, then the pedestrian's rotation angle is predicted, and finally the vehicle braking is controlled to change the pedestrian's rotation angle so that the vehicle that continues to move forward can achieve the effect of protecting pedestrians. The steps of this method to protect pedestrians from collision damage are as follows: S1: Collecting information about the collision between a car and a pedestrian. When the smart car finds that the collision between a car and a pedestrian is unavoidable, the smart car obtains the vehicle speed, the pedestrian's hip height, the engine hood height, the time of the first contact between the car and the pedestrian, the time of the first contact between the head and the vehicle, the head acceleration when the lead vehicle contacts the head, and the head acceleration when the lead vehicle separates from the head through the perception module, and uploads this data to the vehicle prediction module. S2: Predict the pedestrian's rotation angle. The vehicle prediction module first makes a preliminary judgment on the pedestrian's rotation angle range through the pedestrian rotation angle prediction rule based on the data of the perception module, and then further predicts the pedestrian's rotation angle through the pedestrian rotation angle prediction model. Finally, the prediction result is input into the vehicle decision module. S3: Make a decision based on the prediction results. After the decision module receives the prediction results from the pedestrian rotation angle prediction model, it controls the car to make different responses according to different prediction results. If the prediction result is a safe rotation angle, the vehicle is fully braked until it stops. If the prediction result is a dangerous rotation angle, the vehicle is controlled to brake to allow the vehicle to continue moving forward and change the pedestrian's rotation angle.
2. The pedestrian-ground collision damage protection method based on pedestrian rotation angle prediction according to claim 1 is characterized in that: The method constitutes an intelligent vehicle pedestrian protection system, which is a subsystem of the vehicle automatic emergency braking system, and includes three modules: perception, prediction and decision-making; the perception module collects human-vehicle collision information through conventional cameras, radars and other equipment, the prediction module predicts the pedestrian's rotation angle through prediction rules and prediction models, and the decision-making module makes decisions based on different prediction results and controls the vehicle braking to change the pedestrian's rotation angle to protect the pedestrian.
3. The pedestrian-ground collision damage protection method based on pedestrian rotation angle prediction according to claim 1 is characterized in that: The prediction module makes a preliminary judgment on the pedestrian's rotation angle based on the collected vehicle speed v, pedestrian hip height h1, and vehicle hood height h2 through prediction rules. The prediction rules in the prediction module are as follows: S31: When h1 / h2<1.2 or v<20km / h, the pedestrian's motion trajectory after the collision is judged to be a forward projection trajectory, and the pedestrian's rotation angle is always in the interval A[0°, 90°]. In this case, the pedestrian's rotation angle is considered to be a safe rotation angle, and the pedestrian's collision damage is small, so there is no need to use the pedestrian rotation angle prediction model for secondary prediction; S32: When h1 / h2>1.2 and v>20km / h, the pedestrian's motion trajectory after the collision is judged to be a diffraction trajectory, and the pedestrian's rotation angle may be any angle in [0°, 360°]. In this case, the pedestrian's landing angle cannot be determined and further prediction model is required.
4. The pedestrian-ground collision damage protection method based on pedestrian rotation angle prediction according to claim 1 is characterized in that: The pedestrian rotation angle prediction model in step S2 is obtained based on historical data, specifically: S41: Collect data on no less than 2,000 pedestrian-vehicle collision accidents, including seven parameters: vehicle speed, pedestrian hip height, engine hood height, first contact time of pedestrian-vehicle collision, first contact time between the head and the vehicle, head acceleration when the lead vehicle contacts the head, and head acceleration when the lead vehicle separates from the head; S42: Then use K-Nearest Neighbors (K-NN), Support Vector Machine (SVM), Decision Tree (DT) and Random Forest (RF) as the first stage classification method, perform classification prediction on the data after feature extraction and use the prediction results of the first stage model after K-fold cross validation as the input features of the second stage learning model; S43: Selecting a neural network as the learning model of the second stage, learning to predict the pedestrian's rotation angle through seven perception parameters, thereby constructing a prediction model based on a multi-model ensemble method with seven parameters as input parameters, namely, vehicle speed, pedestrian hip height, engine hood height, first contact time of human-vehicle collision, first contact time of human head and vehicle, head acceleration when the lead vehicle contacts, and head acceleration when the lead vehicle separates, and with the pedestrian's rotation angle as the output result; S44: During annual vehicle maintenance, add no less than 500 sets of the latest human-vehicle collision accident data to update the prediction model and replace the vehicle's internal prediction model.
5. The pedestrian-ground collision damage protection method based on pedestrian rotation angle prediction according to claim 1 is characterized in that: In step S3, the safe rotation angle is the rotation angle in the A interval [0°, 90°] and the D interval [270°, 360°]; the dangerous rotation angle is the rotation angle in the B interval [90°, 180°] and the C interval [180°, 270°].
6. The pedestrian-ground collision damage protection method based on pedestrian rotation angle prediction according to claim 1 is characterized in that: The method for controlling the vehicle braking in step S3 is: fully braking the vehicle before time t1 when the pedestrian's head first hits the vehicle, releasing the vehicle brake after time t1 until time t2 when the vehicle is fully braked again to a standstill; for a specific vehicle, t2 can be determined in the following manner: S61: extracting the contour parameters of the front of the vehicle and constructing a front model of the vehicle, wherein the specific parameters include nine parameters: bumper lower height, bumper lower depth, bumper height, bumper upper height, engine hood front edge height, engine hood front edge depth, engine hood angle, engine hood length, and windshield glass angle; S62: Let t2=t1+a, where a is taken from the interval [0.2, 1.6]s, and 1000 points are taken from it; a virtual simulation system including 3 vehicle speeds, 4 pedestrian heights and 2 pedestrian gaits is selected for simulation, and the weighted injury value WIC output by the virtual simulation system under 1000 different control braking durations a is calculated, and its value is calculated using the following formula: where p i =p si ×p hi ×p gi ,Σp i =1; where: p si , p hi , p gi are the ratios of vehicle speed, pedestrian height and pedestrian gait, respectively; I Ci is the injury cost of the human body in the ith simulation in a set of simulations; S63: According to the value obtained in step S62, when the WIC value is the smallest, the corresponding control braking time a is the optimal control braking time a0, and the time t2 for canceling the control braking can be obtained from t2=t1+a0.