Attitude prediction method, neural network model training method and equipment

Abstract

The invention discloses a posture prediction method, and the method comprises the steps: obtaining an input signal generated by a first object in a first time period, inputting the input signal into an action classifier which is trained by a database in advance, obtaining the action type of the first object in the first time period, and determining the posture of the first object in a second time period based on the database, as the database comprises the data pairs of the input signals, the action types and the postures in different time periods, the action classifier trained by the database can predict the postures of the first object in the future time period, so that the accuracy of compensation data is improved, and the display quality of vehicle head-up display imaging and vehicle lamp projection is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of vehicle technology, and specifically to a posture prediction method, a training method and a device for a neural network model. Background Art

[0002] With the development of science and technology, the automotive field has also ushered in a large number of innovations. For example, technologies such as head-up display (HUD) and headlight projection function have begun to be widely used.

[0003] Among them, in the intelligent headlight projection function, in order to ensure that the headlights illuminate a specific target area, it is necessary to adjust the headlight projection light type in real time in combination with the body posture. Failure to apply additional compensation will cause the illuminated area to be significantly deformed, affecting the headlight lighting experience. Similarly, in the HUD imaging process, in order to ensure that the projected marker object fits the target object, it is also necessary to adjust the imaging reminder area in combination with the body posture. Failure to apply compensation will cause the projected related markers and the target object to not fit, affecting the final display quality of the HUD.

[0004] However, there is a system delay in the imaging process of the headlights and HUD, and the body posture may produce slight changes in the system delay, resulting in the compensation result determined based on the body posture being inaccurate, thereby reducing the imaging and display quality. Summary of the invention

[0005] The embodiment of the present application provides a posture prediction method, a training method and a device for a neural network model, which are used to predict the posture of a vehicle in a future time period, thereby improving the accuracy of compensation data, and improving the display quality of vehicle head-up display imaging and headlight projection. The embodiment of the present application also provides a corresponding computing device, a vehicle, and a computer-readable storage medium.

[0006] The first aspect of the present application provides a posture prediction method, which includes: obtaining a first input signal generated by a first object in a first time period, and then inputting the first input signal into an action classifier pre-trained by a first database; wherein the output of the action classifier is a first action type of the first object in the first time period, and the first database includes N data pairs, each of the N data pairs includes a second input signal and a second action type of a second object in a third time period, and a second posture of the second object in a fourth time period, the third time period is continuous with the fourth time period, the fourth time period is a subsequent time period of the third time period, and N is a positive integer greater than 1. Finally, the first posture of the first object in the second time period is determined, the first posture corresponds to the first action type in the first database, the first time period is continuous with the second time period, and the second time period is a subsequent time period of the first time period.

[0007] In the present application, the first object is a deployed vehicle, and the deployed vehicle includes a vehicle signal collector, an inertial measurement unit, and other units or modules such as a data storage and computing unit. When the deployed vehicle is running, the first input signal generated by the deployed vehicle in the first time period is obtained through the vehicle signal collector and the inertial measurement unit. After the first input signal generated by the deployed vehicle in the first time period is input into the action classifier trained with the first database, the action classifier can predict the first action type of the deployed vehicle in the first time period, and then search the first database based on the first action type to find the posture corresponding to the first action type. The posture is the first posture of the deployed vehicle in the second time period, that is, the vehicle posture predicted by the action classifier.

[0008] In the present application, the first input signal, the first action type and the first posture are respectively all input signals, action types and postures obtained within the time period. For example, the first posture is an angle curve obtained by continuous changes in the three-axis angles of pitch angle, horizontal angle and roll angle within the time period.

[0009] Because the first input signal is a plurality of different signal sequences, there is noise interference in the signal sequences, the query complexity of matching between the sequences is high, and the accuracy of matching the corresponding posture is poor. Therefore, after predicting the predefined driving action type of the driver, the embodiment of the present application can match the posture by the action type, and the change pattern of the posture when a specific driving action occurs can better predict the future posture.

[0010] In the first aspect, an input signal generated by a first object in a first time period is obtained, and the input signal is input into a motion classifier that has been pre-trained with a database to obtain the motion type of the first object in the first time period, and the posture of the first object in the second time period can be determined based on the database. Since the database includes data pairs of input signals, motion types and postures in different time periods, the motion classifier trained with the database can predict the posture of the first object in future time periods, thereby improving the accuracy of the compensation data and improving the display quality of the vehicle head-up display imaging and headlight projection.

[0011] In a possible implementation of the first aspect, the method also includes: inputting the first posture and the first input signal into a posture predictor that has been pre-trained with a second database; wherein the output of the posture predictor is a third posture of the first object in a second time period, the second database includes M data pairs, each of the M data pairs includes a second input signal of the second object in a third time period, and a second posture and a fourth posture of the second object in a fourth time period, and the fourth posture of the second object in the fourth time period is determined based on the action classifier and the second input signal.

[0012] In this possible implementation, after the deployed vehicle obtains the first posture, since the first posture may not be accurate, the first posture and the first input signal may continue to be input into the posture predictor pre-trained by the second database, and the posture predictor may more accurately predict the second posture of the deployed vehicle in the second time period. Because posture prediction is a sequence prediction task, the posture predictor may further fuse the template postures of the first database and the second database and the first input signal collected in real time, the limited offline template postures and the current input signal to obtain a more accurate prediction result.

[0013] In a possible implementation manner of the first aspect, the first object and the second object are of the same model, and data collection accuracy of the second object is greater than that of the first object.

[0014] In this possible implementation, the vehicle collecting the database uses high-precision equipment to improve data accuracy, and the application vehicle uses low-precision equipment to reduce production costs without reducing prediction accuracy.

[0015] In a possible implementation of the first aspect, the first object is a vehicle; the first input signal includes at least one of inertial measurement information, vehicle speed, steering wheel information, positioning information and road surface information of the first object; the first action type includes at least one of low-speed start, high-speed start, braking, emergency braking, left turn, right turn, emergency left turn, emergency right turn and reversing of the first object; the first posture includes at least one of pitch angle information, horizontal angle information and roll angle information of the first object.

[0016] In this possible implementation, the posture prediction method is applied to a vehicle, and the first input signal, the first action type and the first posture of the vehicle can specifically adopt a combination of the above-mentioned information, thereby improving the feasibility of the solution.

[0017] In a possible implementation manner of the first aspect, a sum of the first time period and the second time period is less than or equal to 300 milliseconds.

[0018] In this possible implementation, since the system delay of the vehicle's HUD or headlight lighting system is usually around 100 milliseconds, the sum of the first time period and the second time period is less than or equal to 300 milliseconds, and the sum of the first time period and the second time period is greater than or equal to 200 milliseconds, which can match the system delay of the vehicle's HUD or headlight lighting system.

[0019] In a possible implementation manner of the first aspect, the action classifier adopts at least one of a support vector machine model, a random forest model, and an artificial neural network model.

[0020] In this possible implementation, the action classifier can adopt a variety of neural network model architectures, which improves the feasibility of the solution.

[0021] In a possible implementation manner of the first aspect, the posture predictor adopts at least one of a long short-term memory neural network model, an attention-based transformer network model, and a gradient descent-based sequence prediction network model.

[0022] In this possible implementation, the posture predictor can adopt a variety of neural network model architectures, which improves the feasibility of the solution.

[0023] A second aspect of the present application provides a training method for a neural network model, the method comprising: acquiring a first database, the first database comprising N data pairs, each of the N data pairs comprising a second input signal and a second action type of a second object in a third time period, and a second posture of the second object in a fourth time period, the third time period is continuous with the fourth time period, the fourth time period is a subsequent time period of the third time period, N is a positive integer greater than 1, and then training an action classifier based on the first database, the input of the action classifier being the second input signal, and the output of the action classifier being the second action type.

[0024] In the present application, the second object is a collection vehicle. When the collection vehicle is working, the second input signal of the collection vehicle in the third time period is obtained through the vehicle signal collector and the inertial measurement unit, and the second posture of the collection vehicle in the fourth time period is obtained through the vehicle posture collection device.

[0025] In the present application, when establishing the first database, it is not required that all data be acquired in continuous time periods. It is only required that the next time period after the time period for acquiring the second input signal and the second action type is the time period for acquiring the second posture.

[0026] In a possible implementation of the second aspect, the method further includes: acquiring a second database, the second database including M data pairs, each of the M data pairs including a second input signal of the second object in a third time period, and a second posture and a fourth posture of the second object in a fourth time period, the fourth posture of the second object in the fourth time period being determined based on the action classifier and the second input signal. A posture predictor is trained based on the second database, the input of the posture predictor is the second input signal and the fourth posture, and the output of the posture predictor is the second posture.

[0027] In this possible implementation, in the first database, since the second action type cannot cover all combinations of the second input signals, on the basis of training the action classifier, a posture predictor can be further trained to help prediction. In the first database, since the second action type cannot cover all combinations of the second input signals such as vehicle speed, steering wheel information, positioning information, road surface information and inertial measurement information, on the basis of training the action classifier, a posture predictor can be further trained to help prediction.

[0028] In a third aspect of the present application, a computing device is provided for executing the method in the first aspect or any possible implementation of the first aspect. Specifically, the computing device includes a module or unit for executing the method in the first aspect or any possible implementation of the first aspect, such as: an acquisition unit, a determination unit, a first processing unit, and a second processing unit.

[0029] In a fourth aspect of the present application, a computing device is provided for executing the method in the first aspect or any possible implementation of the first aspect. Specifically, the computing device includes a module or unit for executing the method in the first aspect or any possible implementation of the first aspect, such as: a first acquisition unit, a first training unit, a second acquisition unit, and a second training unit.

[0030] In a fifth aspect, the present application provides a computer device, which includes a processor, a memory, and a computer-readable storage medium storing a computer program; the processor is coupled to the computer-readable storage medium, and a computer-executable instruction running on the processor, when the computer-executable instruction is executed by the processor, the processor executes the method as described in the first aspect or any possible implementation of the first aspect. Optionally, the computer device may also include an input / output (I / O) interface, and the computer-readable storage medium storing the computer program may be a memory.

[0031] In a sixth aspect of the present application, a computer device is provided, the computer device comprising a processor, a memory, and a computer-readable storage medium storing a computer program; the processor is coupled to the computer-readable storage medium, and a computer-executable instruction is executed on the processor. When the computer-executable instruction is executed by the processor, the processor executes the method as described in the second aspect or any possible implementation of the second aspect. Optionally, the computer device may also include an input / output (I / O) interface, and the computer-readable storage medium storing the computer program may be a memory.

[0032] The seventh aspect of the present application provides a computer-readable storage medium storing one or more computer-executable instructions. When the computer-executable instructions are executed by a processor, the processor executes a method as described in the first aspect or any possible implementation of the first aspect.

[0033] In an eighth aspect of the present application, a computer-readable storage medium is provided that stores one or more computer-executable instructions. When the computer-executable instructions are executed by a processor, the processor executes a method as described in the second aspect or any possible implementation of the second aspect.

[0034] A ninth aspect of the present application provides a computer program product storing one or more computer-executable instructions. When the computer-executable instructions are executed by a processor, the processor executes a method as described in the first aspect or any possible implementation of the first aspect.

[0035] The tenth aspect of the present application provides a computer program product storing one or more computer-executable instructions. When the computer-executable instructions are executed by a processor, the processor executes a method as described in the second aspect or any possible implementation of the second aspect.

[0036] In the eleventh aspect of the present application, a chip system is provided, which includes at least one processor and an interface, the interface is used to receive data and / or signals, and at least one processor is used to support a computer device to implement the functions involved in the above-mentioned first aspect or any possible implementation of the first aspect. In one possible design, the chip system may also include a memory, which is used to store program instructions and data necessary for the computer device. The chip system can be composed of chips, or it can include chips and other discrete devices.

[0037] The twelfth aspect of the present application provides a chip system, which includes at least one processor and an interface, the interface is used to receive data and / or signals, and at least one processor is used to support a computer device to implement the functions involved in the above-mentioned second aspect or any possible implementation of the second aspect. In one possible design, the chip system may also include a memory, which is used to store program instructions and data necessary for the computer device. The chip system can be composed of chips, or it can include chips and other discrete devices.

[0038] The thirteenth aspect of the present application provides a vehicle, which is used to execute the method as described in the first aspect or any possible implementation method of the first aspect.

[0039] A fourteenth aspect of the present application provides a vehicle, which is used to execute a method as described in the second aspect or any possible implementation of the second aspect.

[0040] In an embodiment of the present application, an input signal generated by a first object in a first time period is obtained, and the input signal is input into an action classifier that has been pre-trained with a database to obtain the action type of the first object in the first time period, and the posture of the first object in the second time period can be determined based on the database. Since the database includes data pairs of input signals, action types and postures in different time periods, the action classifier trained with the database can predict the posture of the first object in future time periods, thereby improving the accuracy of the compensation data and improving the display quality of the vehicle head-up display imaging and headlight projection. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of a scene showing a change in vehicle posture;

[0042] Figure 2 A schematic diagram of the application architecture provided for an embodiment of the present application;

[0043] Figure 3 A schematic diagram of an embodiment of a training method for a neural network model provided in an embodiment of the present application;

[0044] Figure 4 A schematic diagram of the architecture of a training method for a neural network model provided in an embodiment of the present application;

[0045] Figure 5 A schematic diagram of an embodiment of a posture prediction method provided in an embodiment of the present application;

[0046] Figure 6 A schematic diagram of the architecture of the posture prediction method provided in the embodiment of the present application;

[0047] Figure 7 Another schematic diagram of the structure of the posture prediction method provided in the embodiment of the present application;

[0048] Figure 8 Another schematic diagram of the structure of the posture prediction method provided in the embodiment of the present application;

[0049] Fig. 9 A schematic diagram of an embodiment of a computing device provided in an embodiment of the present application;

[0050] Fig.10 A schematic diagram of another embodiment of a computing device provided in an embodiment of the present application;

[0051] Fig.11 A schematic diagram of another embodiment of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The following describes the embodiments of the present application in conjunction with the accompanying drawings. Obviously, the described embodiments are only embodiments of a part of the present application, rather than all embodiments. It is known to those skilled in the art that with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0053] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0054] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0055] In addition, in order to better illustrate the present application, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present application can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present application.

[0056] The following is an example of the application scenario involved in the embodiments of the present application.

[0057] With the development of science and technology, the automotive field has also ushered in a large number of innovations. For example, technologies such as head-up display (HUD) and headlight projection function have begun to be widely used.

[0058] Among them, in the intelligent headlight projection function, in order to ensure that the headlights illuminate a specific target area, it is necessary to adjust the headlight projection light type in real time in combination with the body posture. Failure to apply additional compensation will cause the illuminated area to be significantly deformed, affecting the headlight lighting experience. Similarly, in the HUD imaging process, in order to ensure that the projected marker object fits the target object, it is also necessary to adjust the imaging reminder area in combination with the body posture. Failure to apply compensation will cause the projected related markers and the target object to not fit, affecting the final display quality of the HUD.

[0059] The vehicle body posture includes the pitch angle, yaw angle and roll angle of the vehicle. It is usually defined as the rotation angle of each axis after constructing the xyz axis coordinate system with the center of the vehicle body and the vehicle's forward direction as the x-axis. Usually, the angles corresponding to the xyz axes are called roll, pitch and yaw angles, and the value of each angle is usually between [-180°, 180°].

[0060] For example, Figure 1 As shown, the target vehicle encounters a sharp turn during driving. At this time, the body posture 1 of the target vehicle when driving in the sharp turn will change compared to the body posture 2 when driving in a straight line or a small curve. At this time, the HUD imaging area of ​​the target vehicle will also shift, and the user cannot observe the accurate HUD image. Therefore, it is necessary to determine the compensation value of the HUD imaging according to the body posture and adjust the imaging area of ​​the HUD.

[0061] However, there is a system delay in the imaging process of the headlights and HUD, and the body posture may change slightly during the system delay, resulting in inaccurate compensation results based on the body posture, which reduces the imaging and display quality. Based on this, the embodiment of the present application provides a posture prediction method for predicting the posture of the vehicle in the future time period, thereby improving the accuracy of the compensation data and improving the display quality of the vehicle head-up display imaging and headlight projection.

[0062] like Figure 2 As shown, the posture prediction method provided in the embodiment of the present application is applied to the deployment vehicle, and the training method of the neural network model provided in the embodiment of the present application is applied to the collection vehicle. The collection vehicle is provided with a vehicle posture collection device, a vehicle signal collector and an inertial measurement unit, and the deployment vehicle is provided with a vehicle signal collector, an inertial measurement unit and a data storage and calculation unit, wherein the inertial measurement unit (IMU) is a device for measuring the three-axis posture angle (or angular velocity) and acceleration of an object. The collection vehicle establishes a database through the vehicle posture collection device, the vehicle signal collector and the inertial measurement unit, and trains a neural network model, and sends the database and the neural network model to the deployment vehicle for use. When predicting the body posture, the deployment vehicle uses the data obtained by the vehicle signal collector and the inertial measurement unit, combined with the neural network model and the database, and uses the data storage and calculation unit to complete the body posture prediction of the deployment vehicle. The deployment vehicle can use the predicted body posture to determine the imaging area of ​​the HUD or the headlights. The imaging area of ​​the HUD or headlights of the deployed vehicle is determined based on the predicted vehicle body posture. After a system delay, the imaging area of ​​the HUD or headlights of the deployed vehicle matches the actual vehicle body posture of the deployed vehicle at this time.

[0063] The vehicle in the present application is a vehicle in a broad sense, and the vehicle can be a means of transportation (such as a commercial vehicle, a passenger car, a motorcycle, a flying car, a train, etc.), an industrial vehicle (such as a forklift, a trailer, a tractor, etc.), an engineering vehicle (such as an excavator, a bulldozer, a crane, etc.), agricultural equipment (such as a lawn mower, a harvester, etc.), amusement equipment, a toy vehicle, etc. The embodiments of the present application do not specifically limit the type of vehicle.

[0064] The posture prediction method and the training method of the neural network model provided in the embodiments of the present application are described below in combination with the above application scenarios.

[0065] like Figure 3 As shown, an embodiment of the present application provides a training method for a neural network model, the method comprising:

[0066] 301. Obtain a first database.

[0067] In the present application embodiment, Figure 2 The training method of the neural network model is illustrated by taking the collection vehicle shown as an example. The collection vehicle includes other units or modules such as a vehicle posture collection device, a vehicle signal collector, and an inertial measurement unit.

[0068] The collection vehicle is the second object. When the collection vehicle is working, the second input signal of the collection vehicle in the third time period is obtained through the vehicle signal collector and the inertial measurement unit, and the second posture of the collection vehicle in the fourth time period is obtained through the vehicle posture collection device.

[0069] Specifically, the vehicle signal collector includes a vehicle system and a camera and other units. The vehicle system can obtain the vehicle speed, steering wheel information and positioning information (GPS) of the collected vehicle, the camera can obtain road information, and the inertial measurement unit can obtain the inertial measurement information of the collected vehicle. The inertial measurement information can also be understood as vehicle body posture information, but the accuracy is lower than the result collected by the vehicle posture collection device. Then the second input signal includes the vehicle speed, steering wheel information, positioning information, road surface information and inertial measurement information of the collected vehicle. The vehicle posture collection device is a high-precision vehicle posture collector, which can be a high-performance differential inertial navigation device or a visual posture collection device. It can obtain the second posture of the collected vehicle, and the second posture includes the pitch angle information, horizontal angle information and roll angle information of the collected vehicle. In addition, the collected vehicle also needs to determine its second action type, and the second action type is to collect various driving actions of the driver of the vehicle, such as low-speed start, high-speed start, braking, emergency braking, left turn, right turn, emergency left turn, emergency right turn and reversing.

[0070] A first database can be established based on data acquired by other units or modules such as a vehicle posture acquisition device, a vehicle signal collector, and an inertial measurement unit. The first database includes N data pairs, each of the N data pairs includes a second input signal and a second action type of a second object in a third time period, and a second posture of the second object in a fourth time period, the third time period is continuous with the fourth time period, the fourth time period is a subsequent time period of the third time period, and N is a positive integer greater than 1.

[0071] Table 1

[0072] Time period Second input signal Second action type Second stance T1 A1 B1 C1 T2 A2 B2 C2 T3 A3 B3 C3 T4 A4 B4 C4

[0073] Table 2

[0074] N data pairs Second input signal Second action type Second stance Data pair 1 A1 B1 C2 Data pair 2 A2 B2 C3 Data pair 3 A3 B3 C4

[0075] Exemplarily, as shown in Table 1, the second input signal, the second action type and the second posture of the collection vehicle are obtained in the four time periods of T1, T2, T3 and T4. T1 is a time period of 0-100 milliseconds (ms), T2 is a time period of 100-200 milliseconds, T3 is a time period of 200-300 milliseconds, and T4 is a time period of 300-400 milliseconds. When the third time period is T1, the fourth time period is T2, when the third time period is T2, the fourth time period is T3, and so on. After the collection vehicle obtains the data shown in Table 1, a first database as shown in Table 2 can be established in the collection vehicle or other computing devices.

[0076] It should be understood that when establishing the first database, it is not required that all data be acquired in continuous time periods. It is only required that the next time period after the time period for acquiring the second input signal and the second action type is the time period for acquiring the second posture.

[0077] In addition, the second input signal, the second action type and the second posture are respectively all input signals, action types and postures obtained within the time period. For example, the second posture is an angle curve obtained by continuous changes in the three-axis angles of pitch angle, horizontal angle and roll angle within the time period.

[0078] 302. Train an action classifier based on the first database.

[0079] After acquiring the first database, an action classifier can be trained based on the first database in a collection vehicle or other computing device. The essence of an action classifier is a neural network model, and the action classifier can specifically adopt architectures such as a support vector machine model, a random forest model, and an artificial neural network model. When training the action classifier, the input of the action classifier is the second input signal, and the output is the second action type. After training the second input signal and the second action type in N data pairs, the input signal is input into the action classifier, and the action classifier can output an action type.

[0080] Exemplarily, when A2 is input as an input signal to the action classifier, the action classifier will output B2, or when an input signal similar to A2 is input to the action classifier, the action classifier will also output B2.

[0081] 303. Obtain a second database.

[0082] 304. Train a posture predictor based on the second database.

[0083] In the first database, since the second action type cannot cover all combinations of the second input signals such as vehicle speed, steering wheel information, positioning information, road surface information and inertial measurement information, on the basis of training the action classifier, a posture predictor can be further trained to help prediction. On the basis of obtaining the first database and the action classifier, the second input signal needs to be input to the action classifier, and the action classifier will output the action type. Based on the action type, a search is performed in the first database to find the posture corresponding to the action type. The posture is the fourth posture, that is, the fourth posture is the posture determined based on the action classifier, and the second posture is the real posture.

[0084] Exemplary, based on the data obtained in Table 1, the fourth posture is determined based on the action classifier. As shown in Table 3, the second input signal of the acquisition vehicle, the action type determined based on the action classifier, the fourth posture and the second posture are all obtained in the six time periods of T1, T2, T3, T4, T5 and T6. T1 is a time period of 0-100 milliseconds (ms), T2 is a time period of 100-200 milliseconds, T3 is a time period of 200-300 milliseconds, T4 is a time period of 300-400 milliseconds, T5 is a time period of 400-500 milliseconds, and T6 is a time period of 500-600 milliseconds. When the third time period is T1, the fourth time period is T2, and when the third time period is T2, the fourth time period is T3, and so on. After the acquisition vehicle obtains the data shown in Table 3, a second database as shown in Table 4 can be established in the acquisition vehicle or other computing devices.

[0085] Table 3

[0086] Time period Second input signal Action type determined based on action classifier Second stance The fourth posture T1 A1 B1 C1 C1 T2 A2 B2 C2 C2 T3 A3 B3 C3 C3 T4 A4 B4 C4 C4 T5 A5 B5 C5 C5 T6 A6 B3 C1 C4

[0087] Among them, the second database includes M data pairs, each of the M data pairs includes a second input signal of the second object in a third time period, and a second posture and a fourth posture of the second object in a fourth time period, and the fourth posture of the second object in the fourth time period is determined based on the action classifier and the second input signal.

[0088] It should be understood that there is no correlation between M and N. M and N can have the same value or different values, and the embodiments of the present application do not limit this.

[0089] As shown in Table 4, in the second database, when the second input signal is A5, the actual posture of the vehicle in the next time period is C1. Since the second action type corresponding to A5 is not stored in the first database, and A5 is similar to A3, the output of the action classifier may be B3, and the corresponding predicted posture of the vehicle is C4, but C4 is not the actual posture of the vehicle in the next time period. At this time, after continuing to input A5 and C4 into the trained posture predictor, the output of the posture predictor is C1, so that the trained posture predictor can predict the vehicle posture more accurately.

[0090] Table 4

[0091] M data pairs Second input signal Second posture The fourth posture Data pair 1 A1 C2 C2 Data pair 2 A2 C3 C3 Data pair 3 A3 C4 C4 Data pair 4 A4 C5 C5 Data pair 5 A5 C1 C4

[0092] After acquiring the second database, a posture predictor can be trained based on the second database in a collection vehicle or other computing device. The essence of a posture predictor is a neural network model, and the posture predictor specifically adopts architectures such as a long short-term memory neural network model, an attention-based transformer network model, and a gradient descent-based sequence prediction network model. When training the posture predictor, the input of the posture predictor is the second input signal and the fourth posture, and the output of the posture predictor is the second posture. After training the second input signal, the second posture, and the fourth posture in M ​​data pairs, the input signal and the posture determined by the action classifier are input into the posture predictor, and the posture predictor can output a new predicted posture.

[0093] Exemplarily, if A5 is used as an input signal and C4 is used as a posture input to a posture predictor, the posture predictor will output C1; or if an input signal similar to A5 and a posture similar to C4 are input to the posture predictor, the posture predictor will also output C1.

[0094] like Figure 4As shown, the acquisition vehicle establishes a first database and a second database through a vehicle posture acquisition device, a vehicle signal collector and an inertial measurement unit, and trains an action classifier and a posture predictor based on the first database and the second database. After the first database, the second database, the action classifier and the posture predictor are deployed on the deployed vehicle, the future posture of the deployed vehicle can be predicted.

[0095] like Figure 5 As shown, the embodiment of the present application also provides a posture prediction method, the method comprising:

[0096] 501. Acquire a first input signal generated by a first object within a first time period.

[0097] 502. Input a first input signal to an action classifier that has been pre-trained with a first database.

[0098] 503. Determine a first posture of the first object in a second time period.

[0099] The first posture corresponds to the first action type in the first database, the first time period is continuous with the second time period, and the second time period is a subsequent time period of the first time period.

[0100] In the present application embodiment, Figure 2 The deployment vehicle shown in the figure performs the posture prediction method as an example for explanation. The deployment vehicle includes a vehicle signal collector, an inertial measurement unit, and other units or modules such as a data storage and computing unit.

[0101] After obtaining an action classifier and a posture predictor based on data obtained by the acquisition vehicle, the action classifier and the posture predictor are configured on a deployment vehicle, which is a first object. The action classifier adopts at least one of a support vector machine model, a random forest model, and an artificial neural network model. The posture predictor adopts at least one of a long short-term memory neural network model, an attention-based transformer network model, and a gradient descent-based sequence prediction network model.

[0102] When the deployed vehicle is running, a first input signal generated by the deployed vehicle in a first time period is obtained through the vehicle signal collector and the inertial measurement unit. The first input signal includes at least one of inertial measurement information of the first object, vehicle speed, steering wheel information, positioning information, and road surface information.

[0103] The first database includes N data pairs, each of the N data pairs includes a second input signal and a second action type of the second object in a third time period, and a second posture of the second object in a fourth time period, the third time period is continuous with the fourth time period, the fourth time period is a subsequent time period of the third time period, and N is a positive integer greater than 1.

[0104] like Figure 6 As shown, after the first input signal generated by the deployed vehicle in the first time period is input into the action classifier trained with the first database, the action classifier can predict the first action type of the deployed vehicle in the first time period, and then search the first database based on the first action type to find the posture corresponding to the first action type. The posture is the first posture of the deployed vehicle in the second time period, that is, the vehicle posture predicted by the action classifier.

[0105] Among them, the first action type includes at least one of low-speed start, high-speed start, braking, emergency braking, left turn, right turn, emergency left turn, emergency right turn and reversing of the first object, and the first posture includes at least one of the pitch angle information, horizontal angle information and roll angle information of the first object.

[0106] For example, referring to Table 2, when the second input signal is A2, the first action type output by the action classifier is B2, and the deployed vehicle can predict that the posture of the deployed vehicle is C3 in the second time period. The specific implementation of the action classifier can refer to the training method of the aforementioned neural network model, and the embodiments of this application will not be repeated here.

[0107] Because the first input signal is a plurality of different signal sequences, there is noise interference in the signal sequences, the query complexity of matching between the sequences is high, and the accuracy of matching the corresponding posture is poor. Therefore, after predicting the predefined driving action type of the driver, the embodiment of the present application can match the posture by the action type, and the change pattern of the posture when a specific driving action occurs can better predict the future posture.

[0108] 504. Input the first posture and the first input signal to a posture predictor that has been pre-trained with the second database.

[0109] like Figure 7 As shown, further, after the deployed vehicle obtains the first posture, because the first posture may not be accurate, the first posture and the first input signal can continue to be input into a posture predictor that has been pre-trained with a second database, and the posture predictor can more accurately predict the second posture of the deployed vehicle in the second time period.

[0110] Among them, the output of the posture predictor is the third posture of the first object in the second time period, the second database includes M data pairs, each of the M data pairs includes the second input signal of the second object in the third time period, and the second posture and fourth posture of the second object in the fourth time period, and the fourth posture of the second object in the fourth time period is determined based on the action classifier and the second input signal.

[0111] For example, referring to Table 4, when the second input signal is A2, the first action type output by the action classifier is B2, and the first posture is C3, the deployed vehicle can predict that in the second time period, the third posture of the deployed vehicle is C3. The specific implementation of the posture predictor can refer to the training method of the aforementioned neural network model, and the embodiments of this application will not be repeated here.

[0112] like Figure 8 As shown, after deploying the first database, the second database, the action classifier and / or the posture predictor, the deployed vehicle can predict the posture of the deployed vehicle in the next time period, that is, after predicting the first posture or the third posture of the first object in the second time period in the first time period, the first object determines the imaging area of ​​the HUD or the headlights based on the first posture or the third posture, and after the first object determines the imaging area of ​​the HUD or the headlights, after a system delay, the first object just lights up the HUD or the headlights in the second time period, and at this time the imaging area of ​​the HUD or the headlights just matches the body posture at this time.

[0113] Optionally, the sum of the first time period and the second time period is less than or equal to 300 milliseconds, and the sum of the first time period and the second time period is greater than or equal to 200 milliseconds, which can match the system delay of the vehicle's HUD or headlight lighting system, and the system delay of the vehicle's HUD or headlight lighting system is usually around 100 milliseconds. It should be understood that the sum of the first time period and the second time period can also be longer, and the embodiment of the present application is not limited to this.

[0114] Because posture prediction is a sequence prediction task, the posture predictor can further integrate the template postures of the first database and the second database and the first input signal collected in real time, the limited offline template posture and the current input signal for prediction, so as to obtain more accurate prediction results.

[0115] It should be understood that the posture prediction method and the training method of the neural network model provided in the embodiments of the present application can be applied not only to vehicles, but also to other vehicles that need to be equipped with IMU functions, such as airplanes and drones.

[0116] In the embodiment of the present application, the model of the first object and the second object is the same, that is, the model of the acquisition vehicle used to train the action classifier and the posture predictor is the same as the model of the deployment vehicle, and the body size and other vehicle data are the same, ensuring that the data used for training and the data actually used match. Optionally, the data acquisition accuracy of the second object is greater than that of the first object, that is, the equipment used for data acquisition in the acquisition vehicle can use a more expensive equipment to ensure the accuracy of the data and improve the accuracy of the neural network model, while the equipment used for data acquisition in the deployment vehicle can use a slightly lower cost equipment, and a large number of deployed vehicles can reuse the database and neural network model to reduce large-scale production costs.

[0117] Summarizing the above embodiments, it can be seen that the posture prediction method and the training method of the neural network model provided in the embodiments of the present application can bring beneficial effects including but not limited to the following three points:

[0118] (1) The action classifier trained by the database can predict the posture of the first object in the future time period, thereby improving the accuracy of the compensation data and improving the display quality of the vehicle head-up display imaging and headlight projection.

[0119] (2) The use of a posture predictor trained with a database combined with an action classifier can further improve the accuracy of posture prediction.

[0120] (3) The use of high-precision equipment in vehicles collecting data from the database can improve data accuracy, while the use of low-precision equipment in application vehicles can reduce production costs without reducing prediction accuracy.

[0121] The above introduces the posture prediction method and the training method of the neural network model provided in the embodiments of the present application. The following introduces the device provided in the embodiments of the present application in conjunction with the accompanying drawings.

[0122] like Fig. 9 As shown, the computing device 900 provided in the embodiment of the present application includes:

[0123] The acquisition unit 901 is used to acquire a first input signal generated by a first object within a first time period; the acquisition unit 901 can execute step 501 in the above method embodiment.

[0124] The first processing unit 902 is used to input the first input signal to an action classifier that has been pre-trained with the first database; wherein the output of the action classifier is the first action type of the first object in the first time period, the first database includes N data pairs, each of the N data pairs includes the second input signal and the second action type of the second object in the third time period, and the second posture of the second object in the fourth time period, the third time period is continuous with the fourth time period, the fourth time period is the subsequent time period of the third time period, and N is a positive integer greater than 1; the first processing unit 902 can execute step 502 in the above method embodiment.

[0125] The determining unit 903 is used to determine a first posture of the first object in a second time period, the first posture corresponds to the first action type in the first database, the first time period is continuous with the second time period, and the second time period is a subsequent time period of the first time period. The determining unit 903 can execute step 503 in the above method embodiment.

[0126] Optionally, the computing device 900 further includes a second processing unit 904, which is used to input the first posture and the first input signal into a posture predictor pre-trained by a second database; wherein the output of the posture predictor is a third posture of the first object in a second time period, the second database includes M data pairs, each of the M data pairs includes a second input signal of the second object in a third time period, and a second posture and a fourth posture of the second object in a fourth time period, and the fourth posture of the second object in the fourth time period is determined based on the action classifier and the second input signal. The second processing unit 904 can perform step 504 in the above method embodiment.

[0127] Optionally, the first object and the second object have the same model, and the data acquisition accuracy of the second object is greater than that of the first object.

[0128] Optionally, the first object is a vehicle; the first input signal includes at least one of inertial measurement information, vehicle speed, steering wheel information, positioning information and road surface information of the first object; the first action type includes at least one of low-speed start, high-speed start, braking, emergency braking, left turn, right turn, emergency left turn, emergency right turn and reversing of the first object; the first posture includes at least one of pitch angle information, horizontal angle information and roll angle information of the first object.

[0129] Optionally, the sum of the first time period and the second time period is less than or equal to 300 milliseconds.

[0130] Optionally, the action classifier adopts at least one of a support vector machine model, a random forest model and an artificial neural network model.

[0131] Optionally, the posture predictor adopts at least one of a long short-term memory neural network model, an attention-based transformer network model, and a gradient descent-based sequence prediction network model.

[0132] The computing device 900 provided in the embodiment of the present application can be understood by referring to the corresponding contents of the aforementioned posture prediction method embodiment part, which will not be repeated here.

[0133] like Fig.10 As shown, the computing device 1000 provided in the embodiment of the present application includes:

[0134] The first acquisition unit 1001 is used to acquire a first database, the first database includes N data pairs, each of the N data pairs includes a second input signal and a second action type of the second object in a third time period, and a second posture of the second object in a fourth time period, the third time period is continuous with the fourth time period, the fourth time period is a subsequent time period of the third time period, and N is a positive integer greater than 1; the first acquisition unit 1001 can execute step 301 in the above method embodiment.

[0135] The first training unit 1002 is used to train an action classifier based on the first database, the input of the action classifier is the second input signal, and the output of the action classifier is the second action type. The first training unit 1002 can execute step 302 in the above method embodiment.

[0136] Optionally, the computing device 1000 further includes a second acquisition unit 1003 and a second training unit 1004, the second acquisition unit 1003 is used to acquire a second database, the second database includes M data pairs, each of the M data pairs includes a second input signal of the second object in a third time period, and a second posture and a fourth posture of the second object in a fourth time period, the fourth posture of the second object in the fourth time period is determined based on the action classifier and the second input signal. The second acquisition unit 1003 can perform step 303 in the above method embodiment. The second training unit 1004 is used to train a posture predictor based on the second database, the input of the posture predictor is the second input signal and the fourth posture, and the output of the posture predictor is the second posture. The second training unit 1004 can perform step 304 in the above method embodiment.

[0137] The computing device 1000 provided in the embodiment of the present application can be understood by referring to the corresponding contents of the embodiment of the training method of the aforementioned neural network model, and will not be repeated here.

[0138] Fig.11As shown, a possible logical structure diagram of a computing device 1100 provided in an embodiment of the present application is shown. The computing device 1100 includes: a processor 1101, a communication interface 1102, a memory 1103 and a bus 1104. The processor 1101 may include a CPU, or a CPU and at least one of a GPU, an NPU and other types of processors. The processor 1101, the communication interface 1102 and the memory 1103 are interconnected via the bus 1104. In an embodiment of the present application, the processor 1101 is used to control and manage the actions of the computing device 1100. For example, the processor 1101 is used to execute the above Figures 3 to 8 The training method of the neural network model or the posture prediction method described in some embodiments, and / or other processes used in the technology described herein. The communication interface 1102 is used to support the computing device 1100 to communicate. The memory 1103 is used to store program codes and data of the computing device 1100.

[0139] Among them, the processor 1101 can be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It can implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the contents disclosed in this application. The processor can also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a digital signal processor and a microprocessor, and so on. The bus 1104 can be a peripheral component interconnect standard (Peripheral Component Interconnect, PCI) bus or an extended industry standard architecture (Extended Industry Standard Architecture, EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.11 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0140] The embodiment of the present application also provides a vehicle, which can be Figure 2 The deployment vehicle, or the vehicle includes Fig. 9 or Fig.11 The computing device shown.

[0141] The embodiment of the present application also provides a vehicle, which can be Figure 2 The collection vehicle, or the vehicle includes Fig.10 or Fig.11 The computing device shown.

[0142] In another embodiment of the present application, a computer-readable storage medium is further provided, wherein the computer-readable storage medium stores computer-executable instructions. When at least one processor of the device executes the computer-executable instructions, the device executes the above Figures 3 to 8 Some embodiments describe a neural network model training method or a posture prediction method.

[0143] In another embodiment of the present application, a computer program product is also provided. The computer program product includes computer-executable instructions, which are stored in a computer-readable storage medium; at least one processor of the device can read the computer-executable instructions from the computer-readable storage medium, and at least one processor executes the computer-executable instructions so that the device performs the above Figures 3 to 8 Some embodiments describe a neural network model training method or a posture prediction method.

[0144] In another embodiment of the present application, a chip system is further provided, the chip system comprising at least one processor and an interface, the interface being used to receive data and / or signals, and at least one processor being used to support the implementation of the above Figures 3 to 8 The training method of the neural network model or the posture prediction method described in some embodiments. In a possible design, the chip system may also include a memory, which is used to store program instructions and data necessary for the computer device. The chip system may be composed of a chip or may include a chip and other discrete devices.

[0145] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present application.

[0146] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0147] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0148] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0149] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0150] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, read-only memory), random access memory (RAM, random access memory), disk or optical disk and other media that can store program code.

Claims

1. A posture prediction method, characterized in that: The method comprises: Acquire a first input signal generated by a first object within a first time period; Inputting the first input signal into an action classifier pre-trained by a first database; The output of the action classifier is a first action type of the first object in a first time period, the first database includes N data pairs, each of the N data pairs includes a second input signal and a second action type of the second object in a third time period, and a second posture of the second object in a fourth time period, the third time period is continuous with the fourth time period, the fourth time period is a subsequent time period of the third time period, and N is a positive integer greater than 1; A first posture of the first object in a second time period is determined, the first posture corresponds to the first action type in the first database, the first time period is continuous with the second time period, and the second time period is a subsequent time period of the first time period.

2. The method according to claim 1, characterized in that The method further comprises: Inputting the first posture and the first input signal into a posture predictor pre-trained with a second database; Wherein, the output of the posture predictor is the third posture of the first object in the second time period, the second database includes M data pairs, each of the M data pairs includes the second input signal of the second object in the third time period, and the second posture and fourth posture of the second object in the fourth time period, and the fourth posture of the second object in the fourth time period is determined based on the action classifier and the second input signal.

3. The method according to claim 1 or 2, characterized in that: The first object and the second object have the same model, and the data acquisition accuracy of the second object is greater than that of the first object.

4. The method according to any one of claims 1 to 3, characterized in that The first object is a vehicle; The first input signal includes at least one of inertial measurement information, vehicle speed, steering wheel information, positioning information, and road surface information of the first object; The first action type includes at least one of low-speed start, high-speed start, braking, emergency braking, left turn, right turn, emergency left turn, emergency right turn and reversing of the first object; The first posture includes at least one of pitch angle information, horizontal angle information, and roll angle information of the first object.

5. The method according to any one of claims 1 to 4, characterized in that A sum of the first time period and the second time period is less than or equal to 300 milliseconds.

6. The method according to any one of claims 1 to 5, characterized in that The action classifier adopts at least one of a support vector machine model, a random forest model and an artificial neural network model.

7. The method according to any one of claims 2 to 6, characterized in that: The posture predictor adopts at least one of a long short-term memory neural network model, an attention-based transformer network model, and a gradient descent-based sequence prediction network model.

8. A method for training a neural network model, characterized in that: The method comprises: Acquire a first database, the first database includes N data pairs, each of the N data pairs includes a second input signal and a second action type of a second object in a third time period, and a second posture of the second object in a fourth time period, the third time period is continuous with the fourth time period, the fourth time period is a subsequent time period of the third time period, and N is a positive integer greater than 1; An action classifier is trained based on the first database, the input of the action classifier is the second input signal, and the output of the action classifier is the second action type.

9. The method according to claim 8, characterized in that The method further comprises: Acquire a second database, the second database comprising M data pairs, each of the M data pairs comprising a second input signal of the second object in the third time period, and a second posture and a fourth posture of the second object in the fourth time period, the fourth posture of the second object in the fourth time period being determined based on the action classifier and the second input signal; A posture predictor is trained based on the second database, the input of the posture predictor is the second input signal and the fourth posture, and the output of the posture predictor is the second posture.

10. A computing device, characterized in that: include: An acquisition unit, configured to acquire a first input signal generated by a first object within a first time period; A first processing unit, configured to input the first input signal into an action classifier pre-trained by a first database; The output of the action classifier is a first action type of the first object in a first time period, the first database includes N data pairs, each of the N data pairs includes a second input signal and a second action type of the second object in a third time period, and a second posture of the second object in a fourth time period, the third time period is continuous with the fourth time period, the fourth time period is a subsequent time period of the third time period, and N is a positive integer greater than 1; The determination unit is used to determine a first posture of the first object in a second time period, the first posture corresponds to the first action type in the first database, the first time period is continuous with the second time period, and the second time period is a subsequent time period of the first time period.

11. The computing device according to claim 10, characterized in that Also includes: a second processing unit, configured to input the first posture and the first input signal into a posture predictor pre-trained with a second database; Wherein, the output of the posture predictor is the third posture of the first object in the second time period, the second database includes M data pairs, each of the M data pairs includes the second input signal of the second object in the third time period, and the second posture and fourth posture of the second object in the fourth time period, and the fourth posture of the second object in the fourth time period is determined based on the action classifier and the second input signal.

12. The computing device according to claim 10 or 11, characterized in that The first object and the second object have the same model, and the data acquisition accuracy of the second object is greater than that of the first object.

13. The computing device according to any one of claims 10 to 12, characterized in that: The first object is a vehicle; The first input signal includes at least one of inertial measurement information, vehicle speed, steering wheel information, positioning information, and road surface information of the first object; The first action type includes at least one of low-speed start, high-speed start, braking, emergency braking, left turn, right turn, emergency left turn, emergency right turn and reversing of the first object; The first posture includes at least one of pitch angle information, horizontal angle information, and roll angle information of the first object.

14. The computing device according to any one of claims 10 to 13, characterized in that: A sum of the first time period and the second time period is less than or equal to 300 milliseconds.

15. The computing device according to any one of claims 10 to 14, characterized in that: The action classifier adopts at least one of a support vector machine model, a random forest model and an artificial neural network model.

16. The computing device according to any one of claims 11 to 15, characterized in that: The posture predictor adopts at least one of a long short-term memory neural network model, an attention-based transformer network model, and a gradient descent-based sequence prediction network model.

17. A computing device, characterized in that include: a first acquisition unit, configured to acquire a first database, wherein the first database includes N data pairs, each of the N data pairs includes a second input signal and a second action type of a second object in a third time period, and a second posture of the second object in a fourth time period, the third time period is continuous with the fourth time period, the fourth time period is a subsequent time period of the third time period, and N is a positive integer greater than 1; The first training unit is used to train an action classifier based on the first database, the input of the action classifier is the second input signal, and the output of the action classifier is the second action type.

18. The computing device according to claim 17, characterized in that Also includes: a second acquisition unit, configured to acquire a second database, the second database comprising M data pairs, each of the M data pairs comprising a second input signal of the second object in the third time period, and a second posture and a fourth posture of the second object in the fourth time period, the fourth posture of the second object in the fourth time period being determined based on the action classifier and the second input signal; The second training unit is used to train a posture predictor based on the second database, the input of the posture predictor is the second input signal and the fourth posture, and the output of the posture predictor is the second posture.

19. A vehicle, characterized in that: The vehicle is used to perform the method according to any one of claims 1-7.

20. A vehicle, characterized in that: The vehicle is used to perform the method according to claim 8 or 9.

21. A computing device, characterized in that The method comprises a processor on which a computer-executable instruction runs. When the computer-executable instruction is executed by the processor, the processor executes the method according to any one of claims 1 to 9.

22. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

Images