An in-vehicle occupant collision injury warning method, system and storage medium

Abstract

The application discloses a kind of in-vehicle personnel collision injury early warning method, it includes the following steps: step S10, when meeting trigger condition, depth camera is collected depth image data;Step S11, according to the depth image data collected, the current sign parameter and posture data of in-vehicle personnel are identified using a trained depth learning neural network;Step S12, the sign parameter and posture data identified are input into human body injury prediction model, obtain the data of injury prediction of different parts of human body;Step S13, the data of injury prediction of different parts of human body is sent to car machine and is displayed.The application also discloses corresponding system and storage medium.Implementation of the present application can obtain the data of injury prediction of different parts of human body in real time, can remind automobile user to drive according to standard sitting posture, improve driving safety and use experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of automobile active safety technology, and in particular to a method and system for collision injury warning of a person in a vehicle and a storage medium. BACKGROUND

[0002] In the driving process of a vehicle, an unsafe driving posture can cause greater harm to a person in the vehicle when a collision occurs. However, there is currently no system on the vehicle for real-time evaluation of collision safety for a person in the vehicle. The current automobile occupant restraint system is designed for an average height male in a standard sitting posture. However, in reality, the height, weight, age, and gender of a person in the vehicle are not the same, and the driving posture of different persons in the vehicle also has great differences during driving. In particular, compared with an average height male, women, the elderly, and obese people are all vulnerable groups. Therefore, how to take effective measures to reduce the injury probability of vulnerable occupants and out-of-position occupants in a collision accident and provide the best safety protection for them has become a hot issue in the field of automobile safety. SUMMARY

[0003] The technical problem to be solved by the present application is to provide a method and system for collision injury warning of a person in a vehicle and a storage medium, which can obtain injury prediction data of different parts of the human body in real time and provide reminders, thereby improving driving safety and user experience.

[0004] To solve the above technical problems, as one aspect of the present application, a method for collision injury warning of a person in a vehicle is provided, which includes the following steps:

[0005] Step S10, when it is determined that a trigger condition is reached, collecting depth image data by a depth camera, the depth image data including at least current image data and 3D posture data of a person in the vehicle;

[0006] Step S11, identifying current sign parameters and posture data of a person in the vehicle by using a trained deep learning neural network according to the collected depth image data; wherein the sign parameters include height, weight, age, and gender data of a person in the vehicle, and the posture data includes at least head position, shoulder position, arm position, and chest position data;

[0007] Step S12, inputting the identified sign parameters and posture data into a human body injury prediction model to obtain injury prediction data of different parts of the human body;

[0008] Step S13, sending the injury prediction data of different parts of the human body to a vehicle machine for display to remind a person in the vehicle to drive in a standard sitting posture.

[0009] The trigger condition is that the current vehicle speed is greater than a predetermined threshold.

[0010] Further comprising the step of obtaining training data, specifically:

[0011] Collecting the depth image data of the people in the vehicle through the depth camera installed on the vehicle, continuously collecting data of fixed time length at a predetermined frequency, and recording the height, weight, age and gender information of the people in the vehicle;

[0012] Performing data balancing processing on the obtained continuous data, and obtaining the retained image data of the people in the vehicle after excluding part of the data of normal sitting posture;

[0013] Labeling the human body key points on the retained image data of the people in the vehicle to form a training data set; wherein the key points include multiple points of head position, chest position and limb position; each key point parameter is represented as (u, v, d), wherein u and v represent the horizontal and vertical coordinate positions of the key point on the image, and d represents the distance of the position from the TOF camera.

[0014] Further comprising the steps of pre-establishing a deep learning neural network and training it using the training data set, including:

[0015] Establishing a deep learning neural network, which includes a basic network for feature extraction on the input side, the basic network is connected with a sign recognition branch network and a posture recognition branch network, and the sign recognition branch network and the posture recognition branch network are both connected with 3 layers of fully connected neural network to the output end; the input of the deep learning neural network is the collected depth image data of the people in the vehicle, and the data form is (R, G, B, D), and the output is respectively multiple sign parameters and multiple posture parameters; the basic network, the sign recognition branch network and the posture recognition branch network all adopt convolutional neural network; the sign parameters include height, weight, age, gender, and the posture parameters at least include multiple key point parameters of head position, chest position and limb position;

[0016] Training the deep learning neural network using the obtained training data set, and obtaining the trained deep learning neural network.

[0017] Further comprising the step of pre-establishing a human body injury prediction model, specifically including:

[0018] Adjusting the input value of the standard human body model, forming a transformed human body parameter model by inputting the feature parameters and posture parameters of different human bodies;

[0019] Importing the transformed human body parameter model into the whole vehicle computer simulation model platform for simulation processing, and obtaining the human body injury value under the condition of frontal collision at a predetermined speed;

[0020] The injury values under different sign parameters and different human postures are selected, and a human injury prediction model is established by a second-order response surface model method, and the second-order response surface model is shown in formula (3):

[0021]

[0022] Wherein: f(x) is a human injury prediction value; a0, a i , a ij are fitting parameters of the second-order response surface model; x i , x j are different sign parameters and posture parameters.

[0023] The step S12 further comprises:

[0024] In step S120, the posture data of the current occupant in the vehicle obtained by the trained deep learning neural network is converted into a vehicle coordinate system.

[0025] In step S121, the current sign parameters of the occupant in the vehicle obtained by the trained deep learning neural network and the human posture parameters after coordinate conversion are input into the pre-generated human injury model to predict the injury values of the head, chest and limbs of the occupant, and determine the corresponding grade of the injury values, including normal, mild and severe grades.

[0026] The step S13 further comprises:

[0027] The injury values of the head, chest and limbs of the occupant predicted by the human injury prediction model are sent to the car machine for display, and the corresponding grade is prompted; the prompt includes sound prompt, text prompt or color prompt on the corresponding part of the displayed human body.

[0028] The injury prediction data of different parts of the human body is sent to the car machine for display, reminding the occupant in the vehicle to drive in a standard posture.

[0029] The step S120 further comprises:

[0030] The posture parameters of the occupant in the vehicle predicted by the step S11 are the horizontal and vertical coordinate positions on the image and the distance from the TOF camera, represented as (u, v, d), which are converted into the camera coordinate system by formula (1):

[0031]

[0032] wherein: x', y', z' are coordinate values of the space point in the camera coordinate system; u is the x coordinate of the space point in the image; v is the y coordinate of the space point in the image; d is the distance of the space point to the TOF camera; u0 is the x coordinate of the camera center point in the image; v0 is the y coordinate of the camera center point in the image; f is the camera focal length; dx is the width of a unit pixel in the x direction; and dy is the width of a unit pixel in the y direction;

[0033] The camera coordinate system is converted to the vehicle coordinate system by the following space coordinate conversion formula (2):

[0034]

[0035] wherein: x, y, z are coordinate values of the space point in the vehicle coordinate system; x', y', z' are coordinate values of the space point in the camera coordinate system; x0, y0, z0 are coordinate values of the camera center point in the vehicle coordinate system; and α, β, γ are angles of rotation of the camera coordinate system around the x, y, z axes when converted to the vehicle coordinate system.

[0036] Correspondingly, another aspect of the present application also provides an in-vehicle occupant collision injury early warning system, which at least comprises:

[0037] a depth image data acquisition unit configured to acquire depth image data by a depth camera when it is determined that a trigger condition is reached, wherein the depth image data at least comprises current image data and 3D posture data of an in-vehicle occupant;

[0038] a recognition unit configured to recognize current sign parameters and posture data of the in-vehicle occupant by using a trained deep learning neural network according to the acquired depth image data, wherein the sign parameters comprise height, weight, age and gender data of the in-vehicle occupant, and the posture data at least comprises data of head position, shoulder position, arm position and chest position;

[0039] a prediction unit configured to input the recognized sign parameters and posture data into a human body injury prediction model to obtain injury prediction data of different parts of the human body;

[0040] a prediction result display unit configured to send the injury prediction data of different parts of the human body to a vehicle machine for display to remind the in-vehicle occupant to drive in a standard sitting posture.

[0041] Correspondingly, still another aspect of the present application also provides a computer readable storage medium having a series of computer executable instructions stored thereon, when the series of computer executable instructions are executed by one or more computing devices, the one or more computing devices implement the foregoing method.

[0042] The embodiment of the present application has the following beneficial effects:

[0043] The application provides a method and system for early warning of collision injury of a person in a vehicle and a storage medium. Depth camera is used to collect depth image data of the person in the vehicle in real time, a deep learning model is used to identify the vital sign parameters and body posture of the person in the vehicle, the identified posture data is converted to a vehicle coordinate system, and finally the posture data in the vehicle coordinate system and the vital sign parameters of the person in the vehicle are input into a human body injury prediction model, the model predicts the injury values of the head, chest and limbs of the occupant respectively, and displays the corresponding grade of the injury values on the vehicle through a CAN bus, wherein the grade can be represented by color. Thus, the user of the vehicle can be reminded to drive in a standard sitting posture in time, and the injury degree of the occupant in the vehicle in a collision is reduced.

[0044] In the embodiment of the application, the human body parameter model is generated under different vital sign parameters and body postures, the human body injury under a standard collision is predicted through a simulation model, the human body injury prediction model under different vital sign parameters and body postures is established, and the application is more extensive.

[0045] In the embodiment of the application, the injury of the driver and the passenger is predicted and evaluated by using a computer simulation method, and the vehicle instrument interacts with the person in the vehicle, thereby improving the experience and safety of the user. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only some embodiments of the application, and for those skilled in the art, other drawings can be obtained according to these drawings without creative labor.

[0047] Figure 1 The application provides a main flow schematic diagram of an embodiment of a collision injury early warning method for a person in a vehicle.

[0048] Figure 2 The application relates to a structure and principle schematic diagram of a deep learning neural network.

[0049] Figure 3 The application relates to a principle schematic diagram of a human body injury prediction model.

[0050] Figure 4 The application relates to a schematic diagram of an interface of a vehicle instrument display.

[0051] Figure 5 The application provides a structure schematic diagram of an embodiment of a collision injury early warning system for a person in a vehicle. DETAILED DESCRIPTION

[0052] In order to make the objects, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings.

[0053] As shown in Figure 1 Fig. 1, a main flow diagram of an embodiment of a method for predicting collision injury of a person in a vehicle is shown. In combination with Figures 2 to 4 Fig. 2, in the embodiment, the method comprises at least the following steps:

[0054] Step S10, when it is determined that the triggering condition is reached, collecting depth image data by a depth camera, the depth image data at least comprising current image data and 3D posture data of a person in the vehicle; the person in the vehicle can be a driver or a front seat passenger.

[0055] In a specific example, the triggering condition is that the current vehicle speed is greater than a predetermined threshold (e.g. 30 km / h), and it is determined that when the vehicle speed exceeds 30 km / h, the real-time collection of the depth camera is started.

[0056] Step S11, according to the collected depth image data, a trained deep learning neural network is used to identify the current physical parameters and posture data of the person in the vehicle; wherein the physical parameters include height, weight, age and gender data of the person in the vehicle, and the posture data at least includes head position, shoulder position, arm position and chest position data; it can be understood that the posture data can also include hip position and thigh position data.

[0057] Step S12, inputting the identified physical parameters and posture data into a human body injury prediction model to obtain injury prediction data of different parts of the human body;

[0058] Step S13, sending the injury prediction data of different parts of the human body to a vehicle machine for display, reminding the person in the vehicle to drive according to the standard sitting posture.

[0059] It can be understood that in the embodiment of the present application, a deep learning neural network needs to be established and trained in advance to identify the current physical parameters and posture data of the person in the vehicle.

[0060] More specifically, the further comprising a step of obtaining training data, specifically:

[0061] The depth image data of the people in the vehicle is collected by a depth camera installed on the vehicle, such as arranged on the A-pillar on the driver's side, the data is continuously collected at a predetermined frequency for a fixed time length, and the height, weight, age and gender information of the people in the vehicle are recorded; in an example, the collection frequency is 10 Hz, and the data is continuously collected for more than 5 hours, and the height (CM), weight (kg), age (years) and gender (0 represents female, 1 represents male) of the people in the vehicle are recorded during the collection.

[0062] The obtained continuous data is subjected to data balancing processing, and after excluding part of the data of normal sitting posture, the retained image data of the people in the vehicle is obtained;

[0063] The human key points are labeled on the retained image data of the people in the vehicle to form a training data set; wherein the key points include multiple part points of the head position, chest position and limb position; each key point parameter is represented as (u, v, d), wherein u and v represent the horizontal and vertical coordinate positions of the key point on the image, and d represents the distance of the position from the TOF camera. More specifically, in an example, the key points are: head vertex, nose center point, left ear position, right ear position, left clavicle position, right clavicle position, left shoulder position, right shoulder position, neck center point, sternum position, left hip position, right hip position, left elbow position, right elbow position, left wrist position, right wrist position, left knee position and right knee position, a total of 18 key points;

[0064] Further comprising the steps of: pre-establishing a deep learning neural network and training the same by using the training data set, including:

[0065] A deep learning neural network is established, which includes a basic network for feature extraction on the input side, the basic network is connected with a sign recognition branch network and a posture recognition branch network, and the sign recognition branch network and the posture recognition branch network are both connected with 3 layers of fully connected neural networks to the output end; the input of the deep learning neural network is the collected depth image data of the people in the vehicle, and the data form is (R, G, B, D), and the output is respectively multiple sign parameters (such as 4) and multiple posture parameters (such as 18); the basic network, the sign recognition branch network and the posture recognition branch network all adopt a convolutional neural network; the sign parameters include height (range 1-2m), weight (range 30-150kg), age (range 18-70 years old), and gender (0 is female, 1 is male), and the posture parameters at least include multiple key point parameters (such as the aforementioned 18 key points) of the head position, chest position and limb position; as shown in the figure, the structure and principle of the deep learning neural network involved in the application are shown. Figure 2 ​

[0066] The deep learning neural network is trained using the obtained training dataset, and a trained deep learning neural network is obtained.

[0067] It is understood that the method further includes the step of pre-establishing a human injury prediction model, specifically including:

[0068] The input values ​​of the standard human body model are adjusted by inputting feature parameters and posture parameters of different human bodies to form a transformed human body parameter model. It can be understood that the standard human body model here can be an existing human body parameter model, such as an existing human body parameter model for the 50th percentile of average height for males. By modifying the parameters, the standard human body model can be modified to obtain human body parameter models for different human bodies.

[0069] The transformed human body parameter model is imported into the vehicle computer simulation model platform for simulation processing to obtain the human body injury value under frontal collision at a predetermined speed. The vehicle computer simulation model platform can be implemented using finite element simulation software such as MADYMO and LSDYNA. For different human body parameters, injury values ​​of different human bodies and different human body postures can be obtained.

[0070] Injury values ​​under different vital signs and different human postures were selected, and a human injury prediction model was established by using the second-order response surface model method. The second-order response surface model is shown in formula (3):

[0071]

[0072] Where f(x) is the predicted value of human injury; α0, α i ,α ij Parameters for fitting a second-order response surface model; x i ,x j For different vital signs and posture parameters; the second-order response surface is a multi-input single-output model fitted with a quadratic polynomial.

[0073] like Figure 3 The diagram illustrates the principle of obtaining a human injury prediction model according to the present invention. It is understood that the obtained human injury prediction model can be stored in the vehicle controller.

[0074] Step S12 further includes:

[0075] Step S120: The current posture data of the occupants in the vehicle obtained through the trained deep learning neural network is transformed into vehicle coordinates.

[0076] Step S121, input the current body sign parameter of the passenger in the vehicle obtained by the trained deep learning neural network and the body posture parameter after coordinate conversion into the pre-generated body injury model, predict the injury value of the head, chest and limbs of the occupant, and determine the corresponding grade of the injury value, including normal, mild and severe grades.

[0077] The step S13 further comprises:

[0078] The injury value of the head, chest and limbs of the occupant predicted by the body injury prediction model is sent to the car machine through the CAN bus for display, and the corresponding grade is prompted; the prompt includes sound prompt, text prompt or color prompt on the corresponding part in the displayed body. For example, in one example, green can be used to represent normal grade, yellow can be used to represent mild grade, and red can be used to represent severe grade.

[0079] The injury prediction data of different parts of the body is sent to the car machine for display, reminding the passengers in the vehicle to drive according to the standard sitting posture. Figure 4 As shown in the figure, a car machine display interface schematic diagram is shown, which shows the schematic diagram of the driver and the front passenger, and the injury value is prompted by color on the corresponding part of the body.

[0080] In a specific example, the step S120 further comprises:

[0081] The posture parameter of the passenger in the vehicle predicted by the step S11 is the horizontal and vertical coordinate position on the image and the distance from the TOF camera, represented as (u, v, d), which is converted to the camera coordinate system by formula (1):

[0082]

[0083] Wherein: x', y', z' are the coordinate values of the space point in the camera coordinate system, with the unit of meter; u is the x coordinate of the space point in the image, with the unit of pixel; v is the y coordinate of the space point in the image, with the unit of pixel; d is the distance from the space point to the TOF camera, with the unit of meter; u0 is the x coordinate of the camera center point in the image, with the unit of pixel; v0 is the y coordinate of the camera center point in the image, with the unit of pixel; f is the focal length of the camera, with the unit of millimeter; dx is the width of unit pixel in x direction, with the unit of millimeter / pixel; dy is the width of unit pixel in y direction, with the unit of millimeter / pixel;

[0084] The camera coordinate system is converted to the vehicle coordinate system by the following spatial coordinate conversion formula (2):

[0085]

[0086] Wherein: x, y, z are coordinate values of the space point in the vehicle coordinate system, in meters; x', y', z' are coordinate values of the space point in the camera coordinate system, in meters; x0, y0, z0 are coordinate values of the camera center point in the vehicle coordinate system, in meters; a, b, g are angles of rotation around the x, y, z axes when the camera coordinate system is converted to the vehicle coordinate system, in radians.

[0087] As shown in Figure 5 , a structural schematic diagram of an embodiment of a vehicle occupant collision injury early warning system provided by the application is shown. In this embodiment, the system 1 at least includes:

[0088] a depth image data acquisition unit 10, configured to acquire depth image data through a depth camera when it is determined that a trigger condition is reached, the depth image data at least including current image data and 3D posture data of the vehicle occupant;

[0089] an identification unit 11, configured to identify the current sign parameter and posture data of the vehicle occupant by using a trained deep learning neural network according to the acquired depth image data; wherein the sign parameter includes height, weight, age and gender data of the vehicle occupant, and the posture data at least includes data of head position, shoulder position, arm position and chest position;

[0090] a prediction unit 12, configured to input the identified sign parameter and posture data into a human body injury prediction model to obtain injury prediction data of different parts of the human body;

[0091] a prediction result display unit 13, configured to send the injury prediction data of different parts of the human body to a vehicle machine for display, so as to remind the vehicle occupant to drive in a standard sitting posture.

[0092] More details of the functions of the above units, pre-establishment of the deep learning neural network and training, and pre-formation of the human body injury prediction model can be referred to and combined with the foregoing description of the Figures 1 to 4 , and will not be described here.

[0093] Correspondingly, the application further provides a computer readable storage medium having a series of computer executable instructions stored thereon, when the series of computer executable instructions are executed by one or more computing devices, the one or more computing devices implement the foregoing method.

[0094] The embodiment of the application has the following beneficial effects:

[0095] The application provides a method and system for predicting collision injury of a person in a vehicle and a storage medium.

[0096] In the embodiment of the application, the human parameter model is generated under different body parameters and human postures, the human injury under a standard collision is predicted through a simulation model, the human injury prediction model under different body parameters and human postures is established, and the application is more extensive.

[0097] In the embodiment of the application, the computer simulation means is used to predict and evaluate the injury of the driver and passenger, and the vehicle instrument interacts with the person in the vehicle, thereby improving the experience and safety of the user.

[0098] Those skilled in the art should understand that the embodiments of the application can be provided as a method, device or computer program product.

[0099] The application is described with reference to flowcharts and / or block diagrams of the method, device (system) and computer program product according to the embodiments of the application. Figure 1 The device for implementing the function specified in one flow or multiple flows and / or one block or multiple blocks. Figure 1 The device for implementing the function specified in one flow or multiple flows and / or one block or multiple blocks.

[0100] The above merely provides the preferred embodiment of the present application, and cannot allude the protection scope of the present application, therefore, any equivalent changes made according to the claims of the present application shall be within the scope of the present application.

Claims

1. A method of collision injury warning of an occupant in a vehicle, characterized by, The method comprises the following steps: Step S10, when the trigger condition is met, collecting depth image data by a depth camera, the depth image data at least comprising current image data and 3D posture data of the person in the vehicle; Step S11, according to the collected depth image data, identifying the current sign parameter and posture data of the person in the vehicle by using a trained deep learning neural network; wherein the sign parameter comprises the height, weight, age and gender data of the person in the vehicle, and the posture data at least comprises the data of head position, shoulder position, arm position and chest position; Step S12, inputting the identified sign parameter and posture data into a human body injury prediction model to obtain the injury prediction data of different parts of the human body; Step S13, sending the injury prediction data of different parts of the human body to the vehicle machine for display to remind the person in the vehicle to drive in accordance with the standard sitting posture; The method further comprises the following steps of pre-establishing the human body injury prediction model, specifically comprising: adjusting the input value of the standard human body model, forming a transformed human body parameter model by inputting the feature parameters and posture parameters of different human bodies; importing the transformed human body parameter model into a whole vehicle computer simulation model platform for simulation processing to obtain the human body injury value under the condition of frontal collision at a predetermined speed; selecting the injury value under different sign parameters and different human body postures, and establishing a human body injury prediction model by a second-order response surface model method, the second-order response surface model being shown in formula (3): （3） Wherein: f(x) is the prediction value of human body injury; a0, a i , a ij are fitting parameters of the second-order response surface model; x i , x j are different sign parameters and posture parameters.

2. The method of claim 1, wherein, The trigger condition is that the current vehicle speed is greater than a predetermined threshold.

3. The method of claim 2, wherein, Further comprising the following steps of obtaining training data, specifically comprising: collecting the depth image data of the person in the vehicle by a depth camera installed on the vehicle, continuously collecting data of a fixed time length at a predetermined frequency, and recording the height, weight, age and gender information of the person in the vehicle; performing data balancing processing on the obtained continuous data, excluding part of the data of normal sitting posture, and obtaining the retained image data of the person in the vehicle; labeling the human body key points on the retained image data of the person in the vehicle to form a training data set; wherein the key points comprise a plurality of part points of head position, chest position and limb position; each key point parameter is represented as (u, v, d), wherein u and v represent the horizontal and vertical coordinate positions of the key point on the image, and d represents the distance of the position from the TOF camera.

4. The method of claim 3, wherein, Further comprising: pre-establishing a deep learning neural network and training it by using the training data set, comprising: A deep learning neural network is established, which comprises a basic network for feature extraction at the input side, the basic network is connected with a sign identification branch network and a posture identification branch network, and the sign identification branch network and the posture identification branch network are both connected with 3 layers of full connection neural network to the output end; the input of the deep learning neural network is the collected depth image data of the person in the vehicle, and the data form is (R, G, B, D), and the output is a plurality of sign parameters and a plurality of posture parameters; the basic network, the sign identification branch network and the posture identification branch network all adopt a convolutional neural network; the sign parameters include height, weight, age and gender, and the posture parameters at least include head position, chest position and a plurality of key point parameters of limb position; The deep learning neural network is trained by using the obtained training data set, and a trained deep learning neural network is obtained.

5. The method according to any one of claims 1 to 4, wherein the step S12 further comprises: S120, performing coordinate conversion on the current posture data of the person in the vehicle obtained by the trained deep learning neural network, and converting to the vehicle coordinate; S121, inputting the current sign parameters of the person in the vehicle obtained by the trained deep learning neural network and the human body posture parameters after the coordinate conversion into the pre-generated human body injury model, predicting the injury values of the head, chest and limbs of the occupant, and determining the corresponding grade of the injury values, wherein the grade includes normal, mild and severe grades.

6. The method of claim 5, wherein, The step S13 further comprises: sending the injury values of the head, chest and limbs of the occupant predicted by the human body injury prediction model to the vehicle machine for display, and prompting the corresponding grade; the prompt includes sound prompt, text prompt or color prompt on the corresponding part in the displayed human body; sending the injury prediction data of different parts of the human body to the vehicle machine for display, and reminding the person in the vehicle to drive in accordance with the standard sitting posture.

7. The method of claim 6, wherein, The step S120 further comprises: The posture parameters of the person in the vehicle predicted by the step S11 are the horizontal and vertical coordinate positions on the image and the distance from the TOF camera, which are represented as (u, v, d), and the coordinate system is converted to the camera coordinate system by formula (1): （1） Wherein: x', y', z' are the coordinate values of the space point in the camera coordinate system; u is the x coordinate of the space point in the image; v is the y coordinate of the space point in the image; d is the distance from the space point to the TOF camera; u0 is the x coordinate of the camera center point in the image; v0 is the y coordinate of the camera center point in the image; f is the camera focal length; dx is the width of unit pixel in x direction; dy is the width of unit pixel in y direction; The camera coordinate system is converted to the vehicle coordinate system by the following spatial coordinate conversion formula (2): （2） Wherein: x, y, z are coordinate values of the space point in the vehicle coordinate system; x', y', z' are coordinate values of the space point in the camera coordinate system; x0, y0, z0 are coordinate values of the camera center point in the vehicle coordinate system; α, β, γ are angles of rotation around the x, y, z axes when the camera coordinate system is converted to the vehicle coordinate system.

8. An in-vehicle occupant collision injury warning system characterized by comprising: At least comprising: A depth image data acquisition unit for acquiring depth image data through a depth camera when it is judged that the trigger condition is reached, wherein the depth image data at least includes current image data and 3D posture data of the person in the vehicle; An identification unit for identifying the current sign parameter and posture data of the person in the vehicle by using a trained deep learning neural network according to the acquired depth image data, wherein the sign parameter includes the height, weight, age and gender data of the person in the vehicle, and the posture data at least includes the data of the head position, shoulder position, arm position and chest position; A prediction unit for inputting the identified sign parameter and posture data into a human body injury prediction model to obtain the injury prediction data of different parts of the human body; A prediction result display unit for sending the injury prediction data of different parts of the human body to the vehicle machine for display to remind the person in the vehicle to drive in accordance with the standard sitting posture; Wherein, the human body injury prediction model is pre-established in the following manner: Adjust the input value of the standard human body model, form a transformed human body parameter model by inputting the characteristic parameters and posture parameters of different human bodies; Import the transformed human body parameter model into a whole vehicle computer simulation model platform for simulation processing to obtain the human body injury value under the front collision of the human body at a predetermined speed; Select the injury value under different sign parameters and different human body postures, and establish a human body injury prediction model by a second-order response surface model method, wherein the second-order response surface model is shown in formula (3): （3） Wherein: f(x) is the prediction value of human body injury; a0, a i , a ij are the fitting parameters of the second-order response surface model; x i , x j are different sign parameters and posture parameters.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a series of computer executable instructions, when the series of computer executable instructions are executed by one or more computing devices, the one or more computing devices implement the method as claimed in any one of claims 1-7.

Images