Intelligent connected vehicle safety control method and system based on active and passive safety information fusion

Abstract

The application discloses a smart connected vehicle safety control method and system based on active and passive safety information fusion, which comprises an environment detection module, a risk prediction module and a cooperative control module. The environment detection module comprises three parts of occupant state monitoring, vehicle system monitoring and traffic environment perception, the risk prediction module comprises two parts of vehicle collision risk prediction and occupant injury risk prediction, and the cooperative control module comprises control execution parts of an in-vehicle intelligent restraint system, a vehicle steering system and a braking system. The application fully utilizes the early warning capability of the active safety system on potential risks, and calculates the vehicle collision risk probability and the occupant injury probability in real time according to the established vehicle risk prediction model and the occupant injury risk prediction model. When encountering an unavoidable dangerous working condition, the system adjusts the state of the restraint system so that the restraint system reaches the best state of current occupant protection, and avoids the non-cooperation of the active safety system and the restraint system from causing additional risks suffered by the occupant.

Description

TECHNICAL FIELD

[0001] The application relates to a vehicle safety control method and system based on intelligent networked vehicle active and passive safety information fusion, and belongs to the technical field of vehicle safety control and occupant protection. BACKGROUND

[0002] Vehicle safety has always been one of the problems that researchers in the automotive field pay close attention to. Automotive safety technology can be divided into passive safety technology and active safety technology. The passive safety system of a vehicle refers to a series of devices and technologies for reducing or avoiding injury when an accident occurs. These systems do not involve the prevention of accidents, but protect vehicle occupants and pedestrians when an accident occurs. Active safety technology refers to the technology of sensing potential dangers in advance through cameras, radars and other devices, and intelligently controlling the braking and steering systems to automatically avoid collisions, such as automatic emergency braking systems (AEBS). Both technologies are safety lines for occupants, however, there is a certain fragmentation between active and passive safety technologies at present. Although the active safety system can avoid collision accidents or reduce the effective speed of the vehicle when a collision occurs, different collision avoidance control strategies can produce a large braking deceleration under certain typical traffic conditions, resulting in the phenomenon of occupant body forward displacement. This displacement reduces the protection efficiency of the traditional restraint system and increases the risk of secondary collision of the driver and passenger in the vehicle. Especially when the initial sitting posture of the occupant is in a displacement state, a large braking deceleration will further increase the secondary injury of the driver and passenger in the vehicle.

[0003] Therefore, it is increasingly important to study the integrated system of active and passive safety that works synergistically. On the one hand, the vehicle restraint system that integrates the sensing information of the active safety system can better adapt to the occupant state and traffic scene, and improve the protection efficiency of the occupant. On the other hand, the active safety system can intelligently adjust the time and intensity of the action to avoid negative effects on the restraint system. In addition, due to the current level of technology development and the improper behavior of other road users, it is not practical for intelligent vehicles to completely avoid collision accidents by relying on active safety technologies such as AEBS systems. Under this background, it is particularly important to study the protection efficiency of the occupant protection system after the action of the active safety system. SUMMARY

[0004] In view of the above background and the problems to be solved, the application aims to propose a vehicle safety control method and system based on intelligent networked vehicle active and passive safety information fusion, so as to make full use of the early warning capability of the active sensing system for risks, and to calculate the vehicle collision risk probability and occupant injury probability in real time. When encountering an unavoidable dangerous working condition, the system adjusts the state of the restraint system to achieve the best state of current occupant protection, and avoids the additional injury risk of the occupant caused by the non-synergistic action of the active safety system and the restraint system.

[0005] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0006] The vehicle safety control method and system based on intelligent networked automobile active and passive safety information fusion consider that the progress of future automatic driving technology and intelligent cabin technology will revolutionize the way passengers ride, and passengers will have a more comfortable and free riding way. This change also puts higher requirements on the active safety technology and passive safety system of the vehicle. The control system includes an environment detection module, a risk prediction module and a cooperative control module, which provides better protection for the passengers of the automatic driving vehicle through reasonable control of the vehicle. The environment detection module includes passenger state monitoring, vehicle system monitoring and traffic environment perception, the risk prediction module includes vehicle collision risk prediction and passenger injury risk prediction, and the cooperative control module includes the control execution part of the vehicle intelligent restraint system and the active safety system (such as vehicle automatic emergency steering system and automatic braking system).

[0007] The control method includes the following steps:

[0008] Step 1: The environment detection module obtains the vehicle information including the restraint system state, the passenger state information and the self driving state information (including position, speed and acceleration information) through the passenger state monitoring module and the vehicle system monitoring module;

[0009] Step 2: The environment detection module obtains the information of the vehicle driving environment through the traffic environment perception module, including the kinematics information and shape features of the vehicles, non-motor vehicles and pedestrians around the driving vehicle, and judges the additional danger coefficient of the target participant based on the pre-established danger source classification database;

[0010] Step 3: The risk prediction module judges the vehicle collision risk according to the environment perception information, calculates the minimum acceleration value required for collision avoidance and the vehicle speed when the vehicle drives to the target position after adopting this collision avoidance strategy; the passenger injury risk prediction module estimates the passenger injury risk probability according to the current minimum acceleration value required for collision avoidance, the vehicle speed at the target position, the additional danger information of the target participant, the restraint system state, the passenger state information and the self driving state information;

[0011] Step 4: The system coordination control module calculates the reasonable control parameters of the vehicle intelligent restraint system and the active safety system according to the above information.

[0012] Further, the occupant state monitoring module in step 1 includes a vehicle interior camera and pressure sensors on the seat cushion and backrest, which record and analyze occupant posture information and seat belt usage state information in real time; the vehicle system monitoring module includes a rotation angle sensor at the seat hinge, a seat forward and backward displacement sensor, a vehicle speed sensor at the power and transmission system, and a rotation angle sensor in the steering system, which obtains occupant riding state, vehicle speed, vehicle acceleration and front wheel rotation angle information in real time.

[0013] Wherein, the occupant posture information acquisition is realized based on multi-modal data fusion of image data and pressure sensor data. The occupant state monitoring module uses a neural network model to process real-time image data of the occupant to identify the human posture; the pressure sensor installed on the seat feedbacks the contact state information between the occupant and the seat in real time, and the mutual verification of both information increases the recognition success rate.

[0014] The acquisition of the kinematics information of the ego vehicle is to detect the driving speed of the vehicle in the vehicle driving scene, and to estimate the driving trajectory of the ego vehicle in a certain period of time in the future based on the current state information of the ego vehicle, and the representation parameters include the following categories: Wherein T self (t) represents the position of the ego vehicle after t time, v self , a self are the speed and acceleration information of the ego vehicle, θ self , is the current steering wheel rotation angle input and steering wheel rotation angle velocity information of the vehicle.

[0015] Further, the traffic environment perception module in step 2 includes an exterior camera and a radar, wherein the camera is installed on the front windshield to collect real-time images in front and on both sides during driving, to obtain the shape features of traffic participants and to identify object types; the radar is installed at the vehicle head position to detect the relative distance and relative speed and other kinematics information of the vehicle and the front obstacles.

[0016] The traffic participant information acquisition refers to detecting the motion trajectory, motion speed and obstacle type of the obstacles around the vehicle, and the representation parameters include: T veh (t) = {v veh , P veh , a veh , C veh}, wherein T veh (t) represents the position of other obstacles after t time, P veh represents the current position coordinate information of the obstacle, v self , a self are the speed and acceleration information of the obstacle, and C vehAdditional risk coefficient of different characteristic obstacles. The additional risk coefficient of the detected target obstacle is determined by a hazard source classification database. One method for establishing the hazard source classification database is to classify different characteristics of traffic participants or obstacles by their main characteristics such as size, shape, and color characteristics, and then train a CNN convolutional neural network, and use the trained CNN neural network model to process the information collected by the camera outside the vehicle to complete the identification and classification of the potential collision targets in front, and then determine the additional risk coefficient according to the collision response characteristics of each type of target.

[0017] Further, the prediction of the vehicle collision risk in step 3 is to determine the collision risk according to the intersection communication interval of the ego vehicle and the obstacle trajectory, t col is the minimum acceleration required for vehicle collision avoidance, and the risk existing in the vehicle at present is determined; the parameter R col characterizes the size of the collision risk faced, which can be obtained by the following equation:

[0018]

[0019] wherein a x_n , a y_n are the minimum longitudinal and lateral acceleration required to push the collision avoidance time point beyond the risk time threshold, R col is the vehicle collision risk parameter, F lane is the lane changing feasibility judgment, 0 represents that the lane cannot be changed, and 1 represents that the lane can be changed, a x_max , a y_max represent the maximum longitudinal deceleration and lateral acceleration allowed by the current road, K x_n , K y_n are the longitudinal deceleration risk coefficient and the lateral acceleration risk coefficient, respectively. When the estimated collision risk exceeds the risk threshold, the intervention of the vehicle braking system and the steering system is required.

[0020] The occupant injury risk prediction module, by establishing a member injury risk prediction database, and then based on machine learning, establishes an occupant injury risk prediction model. The occupant injury risk prediction model adopts a hybrid model (Hybrid MLP-SVM model) based on an improved multilayer perceptron (MLP) and a support vector machine (SVM) to perform a prediction task for a high-dimensional input problem. The hybrid model has a first level of MLP model, which is divided into an input layer, a hidden layer and an output layer. The pre-set occupant injury influencing factors are used as the input of the model. The hidden layer uses a ReLU activation function to improve the model's ability to handle nonlinear problems. The feature vector input of the output layer is transmitted to the second level of the SVM model. The extracted features are used as the input of the SVM. The SVM is trained for classification or regression tasks. A radial basis kernel function (RBF) is used to handle nonlinear separable cases. Finally, the prediction result of the occupant injury is obtained.

[0021] The process of making the occupant injury risk prediction database includes the following steps: first, a vehicle collision occupant injury value analysis model under the intervention of an active safety system is established, including a vehicle cabin model, an occupant restraint system model and a human body finite element model. A typical accident scene is reconstructed on a Prescan-Carsim-Simulink joint simulation platform. The vehicle motion waveform generated during the collision avoidance process is used as the boundary condition for solving the occupant restraint system model. The kinematic response of the vehicle during the active safety system intervention process and the occupant injury evaluation index value are analyzed. In order to ensure the rationality of the sample distribution in the database, the motion state of the vehicle at the time of collision, the action time and intensity of the active safety system, the use of the occupant safety belt, the occupant riding posture, the seat back angle and the seat displacement are used as variables. Based on the experimental design method, the simulation solution of the occupant injury value analysis model under different parameters is performed. The dynamic characteristics of the dummy model's head, neck and chest are obtained. A passenger response dataset with sufficient sample size is established. Among them, 80% of the sample data is used as training samples, and 20% of the sample data is used as test samples. Because there are many experimental influencing parameters, first, the results are subjected to Pearson correlation analysis, which is calculated as follows, and the factors that have a greater impact on the occupant injury are determined according to the results. The model structure is simplified and the operation efficiency is improved.

[0022]

[0023] wherein r XY represents the correlation coefficient, σ X , σ Y represents the standard deviation of X and Y, represents the mean value of X and Y, and n is the sample number. In the database analysis, X iRepresent the above influencing factors, such as vehicle collision time motion state, parameter, such as active safety system action time and action intensity, occupant seat belt use, occupant sitting posture, seat back angle, seat front and rear displacement, Y i The evaluation index value of the occupant injury. The present application takes two categories of indexes, the comprehensive injury evaluation index WIC and the AIS injury score, as the focus on the head, neck and chest of the occupant, and calculates the injury evaluation index value based on the output parameters of the parts in the test. WIC is used for the control optimization of the vehicle system, and the target is to minimize the WIC value through the reasonable control of the parameters of the vehicle braking system, steering system and restraint system, and the coefficients a, b and c are the statistical probabilities of the occurrence of injury of the parts in the simulation test; and the AIS injury score focuses on the risk of the severity of the injury of the key parts, and the risk level is determined by the probability of death caused by the injury of the parts, and the specific response value of the test model is connected with the AIS score by an empirical formula.

[0024] Further, the cooperative control module in step 4 selects the best driving control strategy in real time according to the information detected and calculated by each part. When the vehicle is in a typical risk scene such as a blind area scene or a complex intersection, the system cooperative control module formulates a suitable cooperative control strategy according to the vehicle collision risk index R col and the occupant injury risk index AIS and WIC calculated in real time by the risk prediction module. When the vehicle collision risk index exceeds the set threshold, it is judged whether the vehicle can avoid the accident under the current working condition according to the collision avoidance ability of the vehicle, and if the accident can be avoided, the system preferentially controls the vehicle to automatically complete the collision avoidance; if the collision accident is unavoidable, the system formulates the control rules of the vehicle braking and steering systems in combination with the restraint system information, the occupant posture information, the vehicle motion information and the collision target information, adjusts the state of the vehicle restraint system, and reduces the AIS occupant injury risk to a low level; when the AIS score level cannot be reduced, the reduction of the WIC score is taken as the optimization target.

[0025] The beneficial effects of the present application are as follows:

[0026] 1. The current occupant injury prediction model does not consider the influence of the active safety system on the restraint system, and there is a certain error in the response results obtained by the test according to the ideal restraint system state and the occupant sitting posture, and the present application fully considers the change factors of the matching relationship between the restraint system and the occupant, and the training scheme of the prediction model is more comprehensive.

[0027] 2. The present application proposes a passive safety fusion control strategy for the unavoidable accident scene, which can ensure the safety of the vehicle and improve the traffic efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 It is a schematic diagram of the overall principle of the system.

[0029] Figure 2 Fig. 1 shows the installation position of the camera and pressure sensor in the occupant posture detection scheme;

[0030] Figure 3 Fig. 2 shows the database of the occupant injury prediction model and the schematic diagram of the hybrid prediction algorithm. DETAILED DESCRIPTION

[0031] The application will be further described below with reference to the accompanying drawings.

[0032] A vehicle safety control method and system based on intelligent networked vehicle active and passive safety information fusion, as shown in Figure 1 The control system includes an environment detection module, a risk prediction module and a collaborative control module, which provides better protection for the passengers of the autonomous vehicle through reasonable control of the vehicle. The environment detection module includes occupant state monitoring, vehicle system monitoring and traffic environment perception, the risk prediction module includes vehicle collision risk prediction and occupant injury risk prediction, and the collaborative control module includes the control execution part of the in-vehicle intelligent restraint system, the vehicle steering system and the braking system.

[0033] To achieve the above purpose, the control method includes the following steps:

[0034] Step 1: Obtain the vehicle's own information through the occupant state monitoring module and the vehicle system monitoring module, including the state of the restraint system, the occupant state information and the own driving state information (including position, speed and acceleration information);

[0035] Step 2: Obtain the information of the vehicle driving environment through the traffic environment perception module, including the kinematic information and shape features of the traffic participants such as vehicles, non-motor vehicles and pedestrians around the driving vehicle, and judge the additional risk coefficient of the target participant based on the pre-established hazard source classification database;

[0036] Step 3: Determine the vehicle collision risk according to the environment perception information, calculate the minimum acceleration value required for collision avoidance and the vehicle speed when the vehicle reaches the target position after adopting this collision avoidance strategy; the occupant injury risk prediction module estimates the occupant injury risk probability caused by the current minimum acceleration value required for collision avoidance, the vehicle speed at the target position, the additional hazard information of the target participant, the state of the restraint system, the occupant state information and the own driving state information;

[0037] Step 4: Calculate the reasonable control parameters of the vehicle intelligent restraint system and the active safety system according to the above method.

[0038] In one embodiment, as Figure 2As shown, the occupant state monitoring module in step 1 includes vehicle interior cameras and pressure sensors on seat cushions and seatbacks, which record and analyze occupant posture information and seatbelt usage state information in real time; the vehicle system monitoring module includes rotation angle sensors at seat hinges, seat forward and backward displacement sensors, vehicle speed sensors at power and transmission systems, and rotation angle sensors in steering systems, which obtain occupant riding state, vehicle speed, vehicle acceleration, and front wheel rotation angle information in real time.

[0039] The occupant posture information acquisition is realized based on multi-modal data fusion of image data and pressure sensor data. The occupant state monitoring module uses the OpenPose model to process real-time image data of the occupant, which is an open-source human posture recognition model developed based on convolutional neural networks and supervised learning and using caffe as the framework. The model can generate human joint features in a plane and draw lines between joints to identify whether the occupant is in a normal sitting posture or an abnormal sitting posture. The occupant state monitoring module identifies the contact state of the occupant with the seat through the symmetrically installed pressure sensors on the seat. For example, the system can determine the occupant's inclination state according to the information of the left and right pressure sensors. The contact state can be further divided into occupant back dislocation, occupant deviation, and occupant diving. The occupant state monitoring module simultaneously receives data from the human posture recognition model and the pressure sensor, and the information from both parties is verified to increase the recognition success rate.

[0040] The acquisition of the kinematics information of the ego vehicle is to detect the driving speed of the vehicle in the vehicle driving scene, and to estimate the driving trajectory of the ego vehicle in a certain period of time in the future based on the current state information of the ego vehicle. The representative parameters include the following categories: Where T self (t) represents the position of the ego vehicle after t time, v self , a self are the speed and acceleration information of the ego vehicle, θ self , is the current steering wheel rotation angle input and steering wheel rotation angle velocity information of the vehicle.

[0041] In one embodiment, the traffic environment perception module in step 2 includes an outside camera and a radar. The camera is installed on the front windshield to collect real-time images in front and on both sides during driving, obtain shape features of traffic participants, and identify object types. The radar is installed at the front of the vehicle to detect kinematics information such as relative distance and relative speed between the vehicle and the front obstacle.

[0042] The traffic participant information acquisition refers to detecting the motion trajectory, motion speed, and obstacle type of the obstacles around the vehicle. The representative parameters include T veh (t) = {v veh , P veh , a veh , Cveh}, wherein T veh (t) represents the position of the obstacle after time t, P veh represents the current position coordinate information of the obstacle, v self , a self are the speed and acceleration information of the obstacle, respectively, C veh denotes the additional risk coefficient of different characteristic obstacles. The determination of the additional risk coefficient first establishes a hazard source identification database, i.e., relies on a CNN convolutional neural network model to complete the identification and classification of different characteristic traffic participant targets. The main identification features include the shape size, color features, etc. Considering the difference in the damage risk size of the structure of the front target to the rear vehicle, the preset labels of the traffic participants in the CNN model mainly include non-motor vehicles, passenger cars, regular rear commercial vehicles, irregular rear commercial vehicles, pedestrians, and fixed traffic facilities such as guardrails. Then, in the LS-dyna finite element simulation platform, finite element models of different types of traffic participants such as non-motor vehicles, passenger cars, pedestrians, etc. are established, and collision simulation tests of the ego vehicle and different obstacles are performed. In the tests, a finite element dummy model is placed at the seat of the ego vehicle model to measure the size of the injury by the dynamic response values of the key parts of the dummy model in the collision simulation tests. Then, the average response results of the ego occupant model are obtained through a large number of tests, and the occupant model response results are head combined acceleration HIC 15 , neck injury criterion N ij , and chest displacement C disp . Finally, the test results of the target obstacle type as a passenger car are taken as the reference 1, and the test results of other types of traffic participants are normalized to obtain the additional risk coefficients of different traffic participants.

[0043] In one embodiment, the prediction of the vehicle collision risk in step 3 is to determine the collision risk according to the intersection communication interval of the ego vehicle and the obstacle trajectory. The following method is used to determine the intersection, wherein the pedestrian protection factor is considered, and therefore the pedestrian-vehicle collision threshold is listed separately from other collision risks, T ped_r (t) represents the position information of the pedestrian after time t:

[0044] t i = {t | dis(T self (t), T ped_r (t)) < range_sr & dis(T self (t), T veh (t)) < range_sv}

[0045] t col = min(t i )

[0046] Where range_sv and range_sr are the impact thresholds for vehicle-to-vehicle and pedestrian-vehicle collisions, respectively, and are related to the vehicle's own body parameters and the size of other vehicles in the environment. i To predict the time points in the future when a vehicle may be at risk of collision, t col This is the most recent time point of collision risk.

[0047] Determine the current risk to a vehicle by the minimum acceleration required to avoid a collision:

[0048] a x_n =min{a x}

[0049] st{a x |t col >t col_r}

[0050] a y_n =min{a y}

[0051] st{a y |t col >t col_r}

[0052] Among them, t col_r The time threshold for collision risk is determined by the vehicle's own characteristics; a x_n a y_n These are the minimum longitudinal and lateral accelerations required to push the collision avoidance time point back to outside the risk time threshold, respectively.

[0053] Using parameter R col The parameter characterizing the magnitude of the collision risk can be obtained from the following equation:

[0054]

[0055] Among them, R col For vehicle collision risk parameters, F lane To determine the feasibility of changing lanes, 0 represents that changing lanes is not allowed, and 1 represents that changing lanes is allowed. x_max a y_max Represents the maximum longitudinal deceleration and lateral acceleration allowed by the current road surface, K. x_n K y_n These are the longitudinal deceleration risk coefficient and the lateral acceleration risk coefficient, respectively. The risk coefficients need to be given under the comprehensive consideration of traffic efficiency and driving safety. The provisional value for both is 1. When more emphasis is placed on driving safety, the coefficient values ​​can be appropriately increased.

[0056] When the estimated collision risk exceeds the risk threshold, the intervention of the vehicle braking system and steering system is required, wherein the risk threshold has a maximum value of 1, which is less than 1 considering the system error such as lateral maximum acceleration and longitudinal maximum deceleration identification deviation and system safety redundancy.

[0057] In one embodiment, the occupant injury risk prediction relies on deep learning to establish an occupant injury risk prediction model and its database, as shown in Figure 3

[0058] The occupant injury risk prediction model is based on a hybrid MLP-SVM model for prediction tasks of high-dimensional input problems, which combines the deep feature extraction capability of MLP and the excellent classification performance of SVM, and is suitable for processing complex data sets in high-dimensional feature space. The first level of the hybrid model is an MLP model, which is divided into an input layer, a hidden layer and an output layer. The pre-set occupant injury influencing factors are used as the input of the model. The hidden layer uses a ReLU activation function to improve the model's ability to handle nonlinear problems. The feature vector input by the output layer is transmitted to the second-level SVM model. The extracted features are used as the input of the SVM. The SVM is trained for classification or regression tasks. The radial basis kernel function (RBF) is used to handle non-linearly separable cases. Finally, the prediction result of the occupant injury is obtained.

[0059] The process of making the occupant injury risk prediction database includes the following steps: first, establish a vehicle collision occupant injury value analysis model under the intervention of the active safety system, including a vehicle cabin model, an occupant restraint system model and a human body finite element model. Reconstruct typical accident scenarios on a simulation platform. Use the vehicle motion waveform generated during the collision avoidance process as the boundary condition for solving the occupant injury value analysis model. Analyze the kinematic response of the vehicle during the intervention of the active safety system and the occupant injury evaluation index value. In order to ensure the rationality of the sample distribution in the database, the motion state of the vehicle at the time of collision, the action time and intensity of the active safety system, the use of the occupant safety belt, the occupant riding posture, the seatback angle, and the seat displacement are used as variables. Based on the experimental design method, the occupant injury value analysis model under different parameters is simulated and solved to obtain the dynamic characteristics of the dummy model's head, neck, chest and other parts. A sufficient number of occupant response data sets are established. Among them, 80% of the sample data are used as training samples, and 20% of the sample data are used as test samples. Due to the large number of experimental influencing parameters, first, perform a Pearson correlation analysis of the results, calculate as follows, and according to the results, determine the factors that have a greater impact on occupant injury, simplify the model structure, and improve the operation efficiency.

[0060] ​

[0061] wherein X i represents the above influencing factors, Y i represents the evaluation index value of occupant injury. Two types of indexes, comprehensive injury evaluation index WIC and AIS injury score, are taken, and the head, neck and chest of the occupant are focused on, and the injury evaluation index value is calculated based on the output parameters of the parts in the test. The response index calculation standards of the head, neck and chest of the occupant are head combined acceleration HIC 15 , neck injury reference N ij and chest displacement C disp , respectively.

[0062] In the embodiments of the present application, WIC is used for control optimization of a vehicle system, and the target is to minimize the WIC value by reasonable control of the parameters of the vehicle braking system, steering system and restraint system, and the index calculation method is as follows, and the coefficients a, b and c are the statistical probabilities of injury of the parts in the simulation test:

[0063] WIC = C veh (aHIC 15 +bN ij +cC disp )

[0064] s.t.a+b+c = 1

[0065] The AIS injury score focuses on the risk of injury severity of the key parts, and the risk level is determined by the probability of death caused by injury of the parts, and the specific response value of the test model is connected with the AIS score by an empirical formula, and the calculation method of the head, neck and chest of the occupant in the AIS each injury level probability is as follows:

[0066]

[0067] wherein P(AIS head ≥i), i = 1, 2, 3, 4, 5, represents the probability of head injury exceeding the i-th standard of the AIS injury evaluation standard, P(AIS neck ≥i), i = 2, 3, 4, 5, represents the probability of neck injury exceeding the i-th standard of the AIS injury evaluation standard, P(AIS chest ≥i), i = 2, 3, 4, represents the probability of chest injury exceeding the i-th standard of the AIS injury evaluation standard.

[0068] When the vehicle is in a typical risk scenario, the vehicle collision risk index R colAnd the occupant injury risk indicators AIS, WIC, when the vehicle collision risk indicators exceed the set threshold, determine whether the current working condition with the vehicle's ability to avoid collision can avoid the accident, can avoid the accident, then the cooperative control module preferentially controls the vehicle to automatically complete the collision avoidance; If the collision accident is unavoidable, then the cooperative control module combines the restraint system information, the occupant posture information, the vehicle motion information and the collision target information, formulates the control rules of the vehicle braking system and the steering system, adjusts the state of the vehicle intelligent restraint system, and reduces the AIS occupant injury risk to a low level; When the AIS score level cannot be reduced, the WIC score is reduced as the optimization target.

[0069] The above series of detailed descriptions are only specific descriptions of the feasible embodiments of the present application, and are not used to limit the protection scope of the present application. Any equivalent means or changes without departing from the technology of the present application shall be included in the protection scope of the present application.

Claims

1. An intelligent networked vehicle safety control system based on active and passive safety information fusion, characterized in that, The application relates to a vehicle intelligent safety system, which comprises an environment detection module, a risk prediction module and a cooperative control module. The environment detection module comprises a passenger state monitoring module, a vehicle system monitoring module and a traffic environment perception module. The risk prediction module comprises a vehicle collision risk prediction module and a passenger injury risk prediction module. The cooperative control module comprises a vehicle intelligent restraint system module, an automatic emergency steering system module and an automatic emergency braking system module. The environment detection module obtains vehicle self information, including restraint system state, passenger state information and self driving state information, including position, speed and acceleration information, through the passenger state monitoring module and the vehicle system monitoring module; and obtains vehicle driving environment information, including kinematics information and shape features of vehicles, non-motor vehicles and pedestrians around the driving vehicle, through the traffic environment perception module, and judges the additional risk coefficient of the target participant based on a pre-established hazard source classification database. The risk prediction module judges the vehicle collision risk through the vehicle collision risk prediction module based on the environment perception information, calculates the minimum acceleration value required for collision avoidance and the vehicle speed when the vehicle reaches the target position by adopting the collision avoidance strategy; and estimates the passenger injury risk probability caused by the minimum acceleration value required for collision avoidance, the vehicle speed at the target position, the additional hazard information of the target participant, the restraint system state, the passenger state information and the self driving state information through the passenger injury risk prediction module. The cooperative control module calculates reasonable control parameters of the vehicle intelligent restraint system and the active safety system. The vehicle collision risk prediction module determines the collision risk according to the intersection point communication interval of the trajectories of the ego vehicle and obstacles, and determines the intersection point by adopting the following method: The minimum acceleration required for vehicle collision avoidance is used to determine the risk currently existing in the vehicle. t i ={t|dis(T self (t),T ped_r (t))<range_sr&dis(T self (t),T veh (t))<range_sv} t col = min(t i ) Wherein, range sr, range sv are the impact threshold values in car-car collision and car-pedestrian collision, respectively, t i is the time node in which the vehicle is predicted to have a collision risk in the future, col is the nearest collision risk time node, ped_r (t) represents the category of pedestrians in the obstacle, which is listed separately for the consideration of pedestrian protection factors, self (t) represents the position of the ego vehicle after t, veh (t) represents the position of the obstacle after t. The passenger injury risk prediction module establishes a collision accident simulation database, and then establishes a passenger injury risk prediction model based on machine learning; the passenger injury value prediction model adopts a hybrid model based on an improved multilayer perceptron and a support vector machine, performs prediction on a high-dimensional input problem, the hybrid model has a first level of MLP model, is divided into an input layer, a hidden layer and an output layer, pre-set passenger injury influencing factors are used as inputs of the model, the hidden layer adopts a ReLU activation function to improve the ability of the model to process nonlinear problems, and a feature vector input by the output layer is transmitted to a second level of SVM model, extracted features are used as inputs of the SVM, the SVM is trained to perform classification or regression tasks, a radial basis kernel function (RBF) is used to process a nonlinear separable case, and finally a prediction result of passenger injury is obtained. a x_n = min{a x} s.t.{a x |t col >t col_r} a y_n = min{a y} s.t.{a y |t col >t col_r} where t col_r is the collision risk time threshold, determined by the vehicle's own characteristics; a x_n , a y_n are the minimum longitudinal and lateral accelerations required to push the collision avoidance time point beyond the risk time threshold, respectively. The parameter R is used col characterizes the magnitude of the collision risk, which can be obtained from the equation: wherein R col is a vehicle collision risk parameter, F lane is a lane change feasibility determination, 0 representing that lane changing is not possible and 1 representing that lane changing is possible, a x_max , a y_max represent the maximum longitudinal deceleration and lateral acceleration allowed by the current road, K x_n , K y_n are a longitudinal deceleration risk coefficient and a lateral acceleration risk coefficient, respectively, and C veh is an additional risk coefficient for different characteristic obstacles. The collision accident simulation database for training and testing of the passenger injury risk prediction model is established by the following process: ​ Firstly, the active safety system intervention under the vehicle collision occupant injury value analysis model is established, including the vehicle cabin model, the occupant restraint system model and the human body finite element model, the typical accident scene is reconstructed on the simulation platform, the vehicle motion waveform generated in the collision avoidance process is taken as the boundary condition of the occupant injury value analysis model, the kinematic response of the vehicle in the active safety system intervention process and the occupant injury evaluation index value are analyzed; In order to ensure the rationality of the sample distribution in the data set, the motion state of the vehicle at the time of collision, the action time and strength of the active safety system, the use of the occupant safety belt, the occupant riding posture, the seat back angle and the seat displacement are taken as variables, the simulation solution of the occupant injury value analysis model under different parameters is carried out, the dynamic characteristics of the dummy model head, neck and chest are obtained, and the occupant response data set with sufficient sample quantity is established; Among them, 80% of the sample data is used as the training sample, and 20% of the sample data is used as the test sample; The results are analyzed by Pearson correlation analysis, the calculation method is as follows, according to the results, the factors that have greater influence on the occupant injury are determined, the model structure is simplified, and the operation efficiency is improved; wherein r XY represents a correlation coefficient, σ X , σ Y represents a standard deviation of X, Y, represents a mean value of X, Y, n is a sample number, X i represents an influence factor, Y i represents an evaluation index value of occupant injury.

2. The intelligent vehicle safety control system based on active and passive safety information fusion according to claim 1, characterized in that, The occupant state monitoring module includes vehicle interior camera and pressure sensor on seat cushion and backrest, which records and analyzes occupant posture information and seat belt usage state information in real time; Among them, the acquisition of the occupant posture information is realized based on multi-modal data fusion of image data and pressure sensor data, the real-time image data of the occupant is processed by using neural network model to identify the human posture; The pressure sensor installed on the seat feedbacks the contact state between the occupant and the seat in real time, and the mutual verification of both sides increases the recognition success rate. 3.The intelligent networked vehicle safety control system based on active and passive safety information fusion according to claim 1, characterized in that, The vehicle system monitoring module includes the rotation angle sensor at the seat hinge, the seat forward and backward displacement sensor, the speed sensor at the power and transmission system, and the rotation angle sensor in the steering system, which can obtain the occupant riding state, vehicle speed, vehicle acceleration and front wheel rotation angle information in real time; The acquisition of the vehicle kinematics information is to detect the driving speed of the vehicle in a driving scene, and to estimate the driving trajectory of the vehicle in a future period of time based on the current state information of the vehicle, and the representation parameters include the following categories: wherein T self (t) represents the position of the vehicle after t time, v self , a self are the speed and acceleration information of the vehicle respectively, θ self , is the current steering wheel input and steering wheel angular velocity information of the vehicle.

4. The intelligent vehicle safety control system based on active and passive safety information fusion according to claim 1, wherein, The traffic environment perception module includes an outside camera and a radar, wherein the camera is installed on the front windshield to collect real-time images in front and on both sides during driving, and to obtain the shape features of traffic participants and identify object types; The radar is installed at the vehicle head position to detect the relative distance and relative speed between the vehicle and the front obstacle; The traffic participant information is obtained by detecting the motion trajectory, motion speed and obstacle type of the obstacles around the vehicle, and the characteristic parameters include: T veh (t) = {v veh , P veh , a veh , C veh}, wherein T veh (t) represents the position of the obstacle after t time, P veh represents the current position information of the obstacle, v veh , a veh are respectively the speed and acceleration information of the obstacle, and C veh indicates the additional risk coefficient of different characteristic obstacles. Different characteristic traffic participants use a neural network model to process the information collected by the camera outside the vehicle, complete the identification and classification of the potential collision targets in front, and the classification features include the size, color features. 5.The intelligent vehicle safety control system based on active and passive safety information fusion according to claim 1, wherein, The cooperative control module selects the optimal driving control strategy in real time according to the information detected and calculated by each module; when the vehicle is in a typical risk scenario, the cooperative control module formulates a suitable coordinated control strategy according to the vehicle collision risk index R col calculated by the risk prediction module in real time, in combination with the occupant injury risk index AIS and WIC, where WIC represents a comprehensive injury evaluation index and AIS represents an injury score index; when the vehicle collision risk index exceeds a set threshold, it is determined whether the vehicle can avoid the accident under the current working condition according to the collision avoidance capability of the vehicle; if the accident can be avoided, the system preferentially controls the vehicle to automatically complete collision avoidance. If the collision accident is inevitable, the system combines the restraint system information, the occupant posture information, the vehicle motion information and the collision target information to develop the control rules of the vehicle braking and steering system, adjusts the state of the vehicle restraint system, and reduces the AIS occupant injury risk to a low level; When it is impossible to reduce the AIS score level, the WIC score is reduced as the optimization target.

6. The intelligent vehicle safety control system based on active and passive safety information fusion according to claim 5, characterized in that, The WIC is used for control optimization of the vehicle system, the target is to minimize the WIC value by reasonably controlling the parameters of the vehicle braking system, the steering system and the restraint system, and the calculation method of the index is as follows, the coefficients a, b and c are the statistical probabilities of damage in the simulation test of the corresponding parts: WIC = C veh (aHIC 15 +bN ij +cC disp ) s.t.a+b+c=1 HIC = (a(t))2dt 15 N = (a(t))2dt ij C = (a(t))2dt disp D = (a(t))2dt AIS injury score focuses on the risk of injury severity of key parts, and its risk level is determined by the probability of death caused by injury of the corresponding part. The specific response value of the test model and the AIS score are linked by an empirical formula. The probability of the occupant's head, neck, and chest in each AIS injury level is calculated as follows: where P(AIS head ≥i), i = 1, 2, 3, 4, 5, represents the probability of head injury exceeding the i-th level of AIS injury evaluation standard, P(AIS neck ≥i), i = 2, 3, 4, 5, represents the probability of neck injury exceeding the i-th level of AIS injury evaluation standard, P(AIS chest ≥i), i = 2, 3, 4, represents the probability of chest injury exceeding the i-th level of AIS injury evaluation standard.

7. The control method of the intelligent connected vehicle safety control system based on active and passive safety information fusion according to any one of claims 1-6, characterized in that, The method comprises the following steps: Step 1: using the environment detection module to obtain vehicle information including constraint system state, occupant state information and self driving state information through the occupant state monitoring module and the vehicle system monitoring module, and the self driving state information includes position, speed and acceleration information; Step 2: using the environment detection module to obtain the information of the vehicle driving environment through the traffic environment perception module, including the kinematics information and shape features of the vehicles, non-motor vehicles and pedestrians around the driving vehicle, and judging the additional risk coefficient of the target participant based on the pre-established hazard source classification database; Step 3: using the risk prediction module to judge the vehicle collision risk according to the environment perception information, calculating the minimum acceleration value required for collision avoidance and the vehicle speed when the vehicle reaches the target position by adopting this collision avoidance strategy; The occupant injury risk prediction module estimates the occupant injury risk probability caused by the current minimum acceleration value required for collision avoidance, the vehicle speed at the target position, the additional hazard information of the target participant, the constraint system state, the occupant state information and the self driving state information; Step 4: using the cooperative control module to calculate the reasonable control parameters of the vehicle intelligent restraint system, the vehicle automatic emergency steering system and the vehicle automatic emergency braking system according to the above information; The occupant state monitoring module in step 1 includes vehicle interior cameras and pressure sensors on seat cushions and backrests, which record and analyze occupant posture information and seatbelt usage state information in real time; The vehicle system monitoring module includes angle sensors at seat hinges, seat displacement sensors, speed sensors at power and transmission systems, and angle sensors in steering systems, which obtain occupant seating state, vehicle speed, vehicle acceleration and front wheel angle information in real time; The occupant posture information acquisition is realized based on multi-modal data fusion of image data and pressure sensor data. The occupant state monitoring module uses a neural network model to process real-time image data of the occupant to identify the human posture. The pressure sensors installed on the seat feedback the contact state information between the occupant and the seat in real time, and the mutual verification of both sides increases the recognition success rate; The acquisition of the kinematics information of the ego vehicle is to detect the driving speed of the vehicle in a vehicle driving scene, and to estimate the driving trajectory of the ego vehicle in a future period of time based on the current state information of the ego vehicle, and the characterization parameters include the following categories: wherein T self (t) represents the position of the ego vehicle after t time, v self , a self are the speed and acceleration information of the ego vehicle respectively, θ self , is the current steering wheel input and steering wheel angular velocity information of the vehicle. The traffic environment perception module in step 2 includes an outside camera and a radar. The camera is installed on the front windshield to collect real-time images in front and on both sides during driving, obtain shape features of traffic participants and identify object types. The radar is installed at the vehicle head position to detect the relative distance and relative speed between the vehicle and the front obstacle. The traffic participant information acquisition refers to detecting the motion trajectory, motion speed and obstacle type of the obstacles around the vehicle, and the characteristic parameters include: T veh (t) = {v veh , P veh , a veh , C veh}, wherein T veh (t) represents the position of the vehicle after t time, v self , a self are respectively the vehicle speed and acceleration information, and C veh indicates the additional risk coefficient of different characteristic obstacles. Different characteristic traffic participants use a neural network model to process the information collected by the camera outside the vehicle, complete the identification and classification of the potential collision targets in front, and the main classification features include the shape size and color characteristics of the targets. The additional risk coefficient is determined according to the collision response characteristics of each type of target. The prediction of the vehicle collision risk in step 3 is determined according to the intersection communication interval of the ego vehicle and the obstacle trajectory, t col is the latest collision risk time node; the minimum acceleration required for the vehicle to avoid collision is determined as the risk currently existing for the vehicle; the parameter R col characterizes the size of the collision risk faced, which can be obtained by the following equation: wherein a x_n , a y_n are the minimum longitudinal and lateral acceleration required to push the collision avoidance time point beyond the risk time threshold, R col is the vehicle collision risk parameter, F lane is the lane change feasibility determination, 0 represents that the lane change is not possible, 1 represents that the lane change is possible, a x_max , a y_max represent the maximum longitudinal deceleration and lateral acceleration allowed by the current road, K x_n , K y_n are the longitudinal deceleration risk coefficient and lateral acceleration risk coefficient, respectively. The occupant injury risk prediction module, by establishing a crash accident simulation database, and then based on machine learning, establishes an occupant injury risk prediction model, the occupant injury risk prediction model adopts a hybrid model (Hybrid MLP-SVM model) based on an improved multilayer perceptron (MLP) and a support vector machine (SVM), for a prediction task of a high-dimensional input problem, the hybrid model has a first level of MLP model, which is divided into an input layer, a hidden layer and an output layer, pre-set occupant injury influencing factors are taken as inputs of the model, the hidden layer adopts a ReLU activation function to improve the ability of the model to handle nonlinear problems, and a feature vector input by the output layer is transmitted to a second level of SVM model, the extracted features are used as inputs of the SVM, the SVM is trained for classification or regression tasks, a radial basis kernel function (RBF) is used to handle nonlinear separable cases, and finally a prediction result of occupant injury is obtained; The process of establishing the crash accident simulation database includes the following steps: first, a vehicle crash occupant injury value analysis model under the intervention of an active safety system is established, including a vehicle cabin model, an occupant restraint system model and a human body finite element model, a typical accident scene is reconstructed on a simulation platform, vehicle motion waveforms generated during the collision avoidance process are taken as boundary conditions for solving the occupant injury value analysis model, and the kinematic response of the vehicle during the intervention process of the active safety system and the occupant injury evaluation index value are analyzed; in order to ensure the rationality of the sample distribution in the data set, the motion state of the vehicle at the time of the collision, the action time and strength of the active safety system, the occupant seat belt usage, the occupant riding posture, the seat back angle and the seat front and rear displacement are taken as variables, the simulation solving of the occupant injury value analysis model under different parameters is carried out based on the experimental design method, the dynamic characteristics of the dummy model head, neck and chest are obtained, and an occupant response data set with sufficient sample quantity is established; wherein 80% of the sample data is taken as training samples, and 20% of the sample data is taken as test samples; due to the large number of test influencing parameters, pearson correlation analysis is performed on the results, the calculation method is as follows, and the factors that have greater influence on the occupant injury are determined according to the calculation results; wherein X i represents the above-mentioned influence factors, Y i represents the evaluation index value of the occupant injury; The cooperative control module in step 4 is to select the best driving control strategy in real time according to the information detected and calculated by each system; when the vehicle is in a blind area scene with limited perception and a complex intersection traffic situation, the cooperative control module formulates a suitable coordinated control strategy according to the vehicle collision risk index R col and the occupant injury risk index AIS, WIC calculated by the risk prediction module in real time; when the vehicle collision risk index exceeds the set threshold, it is judged whether the vehicle can avoid the accident under the current working condition according to the collision avoidance ability of the vehicle; if the accident can be avoided, the system preferentially controls the vehicle to automatically complete the collision avoidance; if the collision accident is unavoidable, the system formulates the control rules of the vehicle braking and steering systems, adjusts the state of the vehicle restraint system, and reduces the AIS occupant injury risk to a low level; when the AIS score level cannot be reduced, the WIC score is reduced as the optimization target.

Images