Model training methods and methods for predicting the tendency of occupants to descend into zero gravity.

Abstract

This invention relates to a model training method and a method for predicting the descent trend of occupants in zero gravity. The model training method includes: determining sled collision trend parameters; constructing a training dataset based on the sled collision trend parameters; training a descent trend prediction model using the training dataset; wherein the descent trend prediction model is used to predict the descent trend of occupants based on the sled collision trend parameters; and obtaining the trained descent trend prediction model. Compared with existing technologies, this invention has advantages such as improved descent protection capabilities, optimized occupant protection, high prediction accuracy, and strong generalization ability.

Description

Technical Field

[0001] This invention relates to the field of vehicle safety technology, and in particular to a model training method and a method for predicting the tendency of occupants to descend into zero gravity. Background Technology

[0002] More and more vehicles are equipped with zero-gravity seats, but in the event of a collision, occupants in these seats face the risk of lurching, and the seatbelt can cause significant injury to the abdomen and neck. In a frontal collision, lurching is defined as the seatbelt slipping off the pelvis and directly intruding into the abdomen. When lurching occurs, the seatbelt's belly strap slips off the pelvis and directly intrudes into the abdomen, compressing internal organs, while the shoulder strap directly impacts the neck and head, causing injury, and the thighs impact the dashboard, resulting in lower limb injuries. In this situation, the seatbelt not only fails to protect the occupants but can also cause even more serious injuries.

[0003] Although manufacturers recommend using the zero-gravity function when not in motion, misuse by occupants during autonomous driving is unavoidable. Currently, some manufacturers use seat rail collapse mechanisms or back airbags to mitigate the risk of descent injuries during a collision with a zero-gravity seat, but these methods are costly, have limited effectiveness, and cause structural damage to the seat, increasing repair costs. If the descent trend and timing of the occupants could be predicted before it occurs, the safety system would have more reaction time, providing better protection. Real-time prediction of the occupants' descent trend in a zero-gravity posture could achieve superior active and passive safety protection.

[0004] Currently, the prediction accuracy of zero-gravity attitude divergence trend prediction schemes for drivers and passengers in related technologies is low, and they cannot achieve active and passive safety protection for drivers and passengers. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a model training method and a method for predicting the zero-gravity attitude diving trend of occupants, which improves diving protection capability, optimizes occupant protection, has high prediction accuracy and strong generalization ability.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] According to a first aspect of the present invention, a model training method is provided, the method comprising: determining sled collision trend parameters; constructing a training dataset based on the sled collision trend parameters; training a descent trend prediction model using the training dataset; wherein the descent trend prediction model is used to predict the descent trend of a driver or passenger based on the sled collision trend parameters; and obtaining the trained descent trend prediction model.

[0008] In one implementation of the first aspect, the scooter collision trend parameters include at least one of the following: occupant state parameters, zero-gravity seat state parameters, auxiliary mechanism state parameters, seat belt state parameters, and vehicle state parameters.

[0009] In one implementation of the first aspect, determining the sled collision trend parameters includes: determining sled collision trend parameters that include at least one of the following factors: parameter acquisition complexity, divergence trend correlation, and parameter differentiation.

[0010] In one implementation of the first aspect, the sled state collision trend parameters include:

[0011] The occupant status parameters include: the occupant's body dimensions;

[0012] The zero-gravity seat state parameters include: seat cushion tilt angle and backrest tilt angle;

[0013] The vehicle status parameters include: vehicle speed.

[0014] In one implementation of the first aspect, constructing the training dataset includes: obtaining a training dataset based on a sled collision model; wherein the sled collision model is used to obtain vehicle collision simulation results based on different input conditions; the training dataset includes the sled collision trend parameters and the diving trend results, wherein the diving trend results include the diving state and the diving time.

[0015] In one implementation of the first aspect, obtaining the training dataset based on the sled collision model includes: obtaining an initial training dataset based on the sled collision model; and performing augmentation processing on the initial training dataset to obtain an augmented training dataset.

[0016] In one implementation of the first aspect, the augmentation process of the initial training dataset to obtain an augmented training dataset includes: augmenting the initial training dataset using an interpolation method to obtain an augmented training dataset.

[0017] In one implementation of the first aspect, the diving trend prediction model is a model built based on a deep neural network (DNN).

[0018] According to a second aspect of the present invention, a method for predicting the descent trend of a driver or passenger in zero gravity is provided. The method includes: acquiring sled collision trend parameters; inputting the sled collision trend parameters into a descent trend prediction model to obtain a descent trend prediction result; wherein the descent trend prediction model is a trained descent trend prediction model obtained by using the method provided by the first aspect or any possible implementation of the first aspect.

[0019] In one implementation of the second aspect, obtaining the sled collision trend parameters includes: obtaining some or all of the sled collision trend parameters based on computer vision technology.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] I. Enhanced Descent Protection Capabilities: The model training method and the descent trend prediction method for zero-gravity occupants provided by this invention can greatly enhance the descent protection capabilities for occupants, especially those in zero-gravity states.

[0022] II. Optimized protection for drivers and passengers: The model training method and the driver and passenger zero-gravity attitude tendency prediction method provided by this invention can provide better protection for drivers and passengers during vehicle collisions and effectively reduce potential injuries.

[0023] III. High prediction accuracy and strong generalization ability: The model training method and the occupant zero-gravity attitude descent trend prediction method provided by this invention are based on computer vision and deep learning. The descent trend prediction model has excellent prediction accuracy and generalization ability and can adapt to different occupants and collision conditions. Attached Figure Description

[0024] Figure 1 A schematic flowchart of the model training method provided in an embodiment of the present invention;

[0025] Figure 2 A schematic diagram of the seat cushion tilt angle and backrest tilt angle provided in an embodiment of the present invention;

[0026] Figure 3 This is a schematic diagram illustrating a training method for a diving trend prediction model in a certain scenario provided in an embodiment of the present invention.

[0027] Figure 4 This is a schematic diagram of the architecture of a diving trend prediction model in a certain scenario provided by an embodiment of the present invention;

[0028] Figure 5 This is a schematic diagram illustrating the training process of a diving trend prediction model in a certain scenario provided in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] As used herein, “an embodiment” or “an embodiment” refers to a specific feature, structure or characteristic that may be included in at least one implementation of this application.

[0031] Figure 1 This is a flowchart illustrating the model training method provided in the embodiments of this application. This application provides the operational steps of the method described in the embodiments or flowcharts, but based on conventional or non-inventive labor, more or fewer operational steps may be included. The order of steps listed in the embodiments is merely one model training method and a method for predicting the tendency of a driver's attitude in zero gravity among many possible execution orders, and does not represent the only execution order. This method should be implementable by software and / or hardware. Please refer to... Figure 1 Model training methods can include:

[0032] Step S110: Determine the sled collision trend parameters;

[0033] Step S120: Construct a training dataset based on the sled collision trend parameters;

[0034] Step S130: Train the descent trend prediction model using the training dataset; wherein, the descent trend prediction model is used to predict the descent trend of the occupants based on the sled collision trend parameters;

[0035] Step S140: Obtain the trained diving trend prediction model

[0036] The following describes an implementation method for determining the sled collision trend parameters in step S110 above:

[0037] Optionally, the aforementioned sled collision trend parameters may include at least one of the following: occupant status parameters, zero-gravity seat status parameters, auxiliary mechanism status parameters, seat belt status parameters, and vehicle status parameters.

[0038] Understandably, the mechanisms and influencing factors of dives have been studied and applied in engineering for a long time. Dives are influenced by many factors, primarily the following:

[0039] (1) Occupant status parameters:

[0040] For example, the status parameters of the driver and passengers may include: the size and posture of the driver and passengers, the position of the feet, etc.

[0041] (2) Zero-gravity seat state parameters:

[0042] For example, zero-gravity seat state parameters may include: seat cushion tilt angle and backrest tilt angle;

[0043] A diagram showing the seat cushion angle and backrest angle is shown below. Figure 2 As shown, angle A represents the seat cushion tilt angle of the current human body posture, and angle B represents the backrest tilt angle of the current human body posture.

[0044] (3) Auxiliary mechanism state parameters:

[0045] Understandably, the design of seats and seat belt assist mechanisms helps to mitigate the risk of passengers falling down. Assist mechanisms include, for example, a descent lever, knee and back airbags, seat belt pretensioners, seat belt force limiters, and seat belt locking mechanisms.

[0046] For example, auxiliary mechanism status parameters may include: whether the auxiliary mechanism is installed or not, the installation location of the auxiliary mechanism, etc.

[0047] (4) Seatbelt status parameters:

[0048] Understandably, the descent is related to the placement of the upper and lower anchor points of the seatbelt.

[0049] For example, seat belt status parameters may include: the position of the lower anchor point of the seat belt, the position of the seat belt D-ring, and the gap between the seat belt and the person.

[0050] (5) Vehicle status parameters:

[0051] It is understandable that the collision speed is closely related to the descent; under the same attitude, the faster the collision speed, the higher the risk of descent.

[0052] For example, vehicle status parameters may include: vehicle speed.

[0053] Optionally, step S110 may include: determining a sled collision trend parameter that includes at least one of the following factors: occupant state parameters, zero-gravity seat state parameters, auxiliary mechanism state parameters, seat belt state parameters, and vehicle state parameters, based on at least one of the following considerations: parameter acquisition complexity, descent trend correlation, and parameter differentiation.

[0054] Optionally, in a certain scenario, considering at least one of the following factors: parameter acquisition complexity, correlation of descent trends, and parameter differentiation, the above-mentioned sled state collision trend parameters may include:

[0055] Occupant status parameters, including: occupant body dimensions;

[0056] Zero-gravity seat parameters include: seat cushion tilt angle and backrest tilt angle;

[0057] Vehicle status parameters, including: vehicle speed.

[0058] The scheme for training the descent trend prediction model using the four parameters mentioned above—human body dimensions of the driver and passengers, seat cushion tilt angle, backrest tilt angle, and vehicle speed—is as follows: Figure 3 As shown, the input to the descent trend prediction model consists of four parameters: human body size, seat cushion tilt angle, backrest tilt angle, and vehicle speed. The output is the descent state and descent time.

[0059] The following describes an optional implementation of step S120, which involves constructing the training dataset:

[0060] Optionally, step S120 above, which constructs the training dataset, includes: obtaining a training dataset based on the sled collision model; wherein, the sled collision model is used to obtain vehicle collision simulation results based on different input conditions; the training dataset includes sled collision trend parameters and diving trend results, and the diving trend results include diving state and diving time.

[0061] Understandably, the above-mentioned sled collision model can be built based on a vehicle simulation model.

[0062] Optionally, the above-mentioned acquisition of the training dataset based on the sled collision model includes: acquiring an initial training dataset based on the sled collision model; and augmenting the initial training dataset to obtain an augmented training dataset.

[0063] Optionally, the initial training dataset can be augmented to obtain an augmented training dataset, including: augmenting the initial training dataset using an interpolation method to obtain an augmented training dataset.

[0064] It is understandable that interpolation methods can augment the initial training dataset to obtain more comprehensive training data.

[0065] The training scheme for the diving trend prediction model in step S130 above is described below:

[0066] Optionally, the diving trend prediction model is a model built based on a deep neural network (DNN), and its structure is as follows: Figure 4As shown, the model can include an input layer, a hidden layer, and an output layer. The input layer takes into account sled collision trend parameters, such as human body size, vehicle speed, seat inclination angle, and backrest inclination angle. The output layer outputs the descent state (i.e., whether or not the vehicle is descent) and the descent time. The descent trend prediction model constructed using a deep neural network (DNN) exhibits good prediction accuracy and generalization ability.

[0067] Additionally, it is understandable that the parameters can be normalized and regularized before being input into the sled collision trend prediction model, so that the model can process the input data more quickly and efficiently.

[0068] In addition, it is understandable that the momentum gradient method can be used to train the diving trend prediction model, which is beneficial to improve the training efficiency and accuracy of the model and optimizes the problem that the model may get stuck in local optima.

[0069] Additionally, it is understandable that the above training dataset can be divided into a training set and a test set. The training set is used to train the diving trend prediction model, while the test set is used to test the diving trend prediction model until the diving trend prediction model reaches the expected accuracy or reaches the preset number of iterations.

[0070] Please see Figure 5 The following provides a complete training scheme for the above-mentioned diving trend prediction model in a certain scenario, which mainly includes:

[0071] Step 1: Build the sled collision model;

[0072] Step 2: Obtain the sled collision trend parameters;

[0073] Step 3: Perform interpolation to increase the data sample size and obtain the collision divergence trend dataset;

[0074] Step 4: Divide the collision diving trend dataset into a training set and a test set;

[0075] Step 5: Train the diving trend prediction model using the training set, which mainly includes:

[0076] (1) Normalize the data and process the input / output functions;

[0077] (2) Set the model training parameters;

[0078] (3) Train the diving trend prediction model (DNN model);

[0079] (4) Test the model using the test set;

[0080] (5) If the model accuracy reaches the expected accuracy, the trained model is obtained and the training process ends.

[0081] If the model accuracy does not reach the expected accuracy, adjust the training and model parameters, and return to step (3) above to retrain the model.

[0082] Based on the same inventive concept, embodiments of the present invention also provide a method for predicting the descent trend of occupants in zero gravity, the method comprising:

[0083] Step S210: Obtain the sled collision trend parameters;

[0084] Step S220: Input the sled collision trend parameters into the diving trend prediction model and obtain the diving trend prediction result; wherein, the diving trend prediction model is a trained diving trend prediction model obtained by any of the above model training methods.

[0085] Optionally, step S210 above may include: obtaining some or all of the sled collision trend parameters based on computer vision technology.

[0086] It is understandable that when an image acquisition device is installed inside the vehicle, some or all of the sled collision trend parameters can be obtained from the images acquired by the device. Taking a descent trend prediction model that uses human body size, seat cushion tilt angle, backrest tilt angle, and vehicle speed as inputs as an example, the image acquisition device installed inside the vehicle can at least acquire parameters such as human body size, seat cushion tilt angle, and backrest tilt angle, thereby achieving zero-gravity attitude descent trend prediction of occupants based on vision and deep neural learning.

[0087] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A model training method, characterized in that, The method includes: Determine the collision trend parameters of the sled; Based on the aforementioned trolley collision trend parameters, a training dataset is constructed; The descent trend prediction model is trained using the training dataset; wherein the descent trend prediction model is used to predict the descent trend of the occupants based on the sled collision trend parameters. Obtain the trained diving trend prediction model; The sled collision trend parameters include: The occupant status parameters include: the occupant's body dimensions; Zero-gravity seat parameters include: seat cushion tilt angle and backrest tilt angle; Vehicle status parameters, including: vehicle speed; The construction of the training dataset includes: A training dataset is obtained based on a sled collision model; wherein, the sled collision model is used to obtain vehicle collision simulation results based on different input conditions; the training dataset includes the sled collision trend parameters and the diving trend results, wherein the diving trend results include the diving state and the diving time.

2. The model training method according to claim 1, characterized in that, The training dataset obtained based on the sled collision model includes: Based on the sled collision model, obtain the initial training dataset; The initial training dataset is augmented to obtain an augmented training dataset.

3. The model training method according to claim 2, characterized in that, The augmentation process on the initial training dataset to obtain the augmented training dataset includes: The initial training dataset is augmented using an interpolation method to obtain an augmented training dataset.

4. The model training method according to any one of claims 1 to 3, characterized in that, The diving trend prediction model is a model built based on a deep neural network (DNN).

5. A method for predicting the tendency of occupants to descend into zero gravity, characterized in that, The method includes: Obtain the collision trend parameters of the sled; The collision trend parameters of the sled are input into the diving trend prediction model to obtain the diving trend prediction result; wherein, the diving trend prediction model is a trained diving trend prediction model obtained by the method described in any one of claims 1 to 4.

6. The method for predicting the descent trend of occupants in zero gravity according to claim 5, characterized in that, The acquisition of sled collision trend parameters includes: Based on computer vision technology, some or all of the sled collision trend parameters are obtained.

Images