Method and device for predicting pose information of vehicle and computer readable storage medium

Abstract

The embodiment of the invention provides a method and device for predicting pose information of a vehicle and a computer readable storage medium, and the method comprises the steps: obtaining the first predicted pose information of the vehicle at the current moment, and obtaining the first predicted pose information of the vehicle at the current moment according to the actual observation data of the vehicle at the current moment; the initial prediction attitude information of the vehicle is obtained through updating; the first prediction attitude information is input into a target Kalman filtering model, second prediction attitude information of the vehicle at the next discrete moment of the current moment is obtained through prediction, and the target Kalman filtering model is obtained by updating a state covariance matrix in the initial Kalman filtering model based on actual observation data; and inputting the first predicted attitude information and the first predicted position information of the vehicle at the current moment into a kinematic model, and predicting to obtain second predicted position information of the vehicle at the next discrete moment. The technical problem of inaccurate vehicle positioning is solved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of autonomous driving, and in particular, to a method, an apparatus, and a computer-readable storage medium for predicting the pose information of a vehicle. Background Art

[0002] With the rapid development of autonomous driving technology, the automatic parking function has become an important part of vehicles. During the automatic parking process, the vehicle needs to accurately locate its position and attitude in the parking lot environment to achieve safe and efficient parking operations.

[0003] In the prior art, during the automatic parking process, vehicle positioning mainly relies on the measurement data of the vehicle's wheel speedometer, steering wheel angle sensor, and inertial measurement unit to predict the vehicle's pose information and position information. For the measurement errors of the sensors, simple filtering or compensation methods are usually adopted, but this method is difficult to effectively cope with the fluctuations of the sensor performance in a complex environment, which further leads to the technical problem of inaccurate vehicle positioning.

[0004] Regarding the above technical problem of inaccurate vehicle positioning, no effective solution has been proposed yet. Summary of the Invention

[0005] The embodiments of the present application provide a method, an apparatus, and a computer-readable storage medium for predicting the pose information of a vehicle, aiming to improve the technical problem of inaccurate vehicle positioning.

[0006] According to one aspect of the embodiments of the present application, a method for predicting the pose information of a vehicle is provided, including: obtaining first predicted pose information of the vehicle at the current moment, where the first predicted pose information is obtained by updating the initial predicted pose information of the vehicle according to the actual observation data at the current moment, and the initial predicted pose information is obtained by predicting the pose of the vehicle using an initial Kalman filter model; inputting the first predicted pose information into a target Kalman filter model to predict second predicted pose information of the vehicle at the next discrete moment at the current moment, where the target Kalman filter model is obtained by updating the state covariance matrix in the initial Kalman filter model based on the actual observation data; and inputting the first predicted pose information and the first predicted position information of the vehicle at the current moment into a kinematic model to predict second predicted position information of the vehicle at the next discrete moment.

[0007] In the above embodiments of the present application, when predicting the attitude information of the vehicle, by fusing the observed data and the predicted attitude information of the Kalman filter model in real time, the attitude information of the vehicle is updated and optimized, which can significantly improve the accuracy of attitude prediction of the vehicle in the automatic parking scenario. This process not only corrects the attitude estimation deviation caused by sensor measurement errors, environmental factors, and model inaccuracies, but also ensures that the model can more accurately reflect the true attitude of the vehicle by dynamically adjusting the state estimation covariance matrix of the Kalman filter model. In addition, by combining the optimized predicted attitude information with the kinematic model, the position information of the vehicle at the next discrete moment is predicted. This strategy significantly improves the accuracy of the vehicle position information. During the automatic parking process, the complexity of the vehicle's dynamic behavior and road surface conditions requires the prediction model to accurately capture and reflect the impact of changes in the vehicle's attitude on the position. By using the optimized attitude information as the input, the kinematic model can more accurately predict the future position of the vehicle, thereby achieving high-precision tracking of the vehicle's movement in automatic parking control, improving the positioning accuracy and stability of the vehicle in a complex environment, achieving the technical effect of improving the positioning accuracy of the vehicle, and further solving the technical problem of inaccurate vehicle positioning.

[0008] Optionally, inputting the first predicted attitude information into the target Kalman filter model to predict the second predicted attitude information of the vehicle at the next discrete moment of the current moment includes: converting the first predicted attitude information into a first quaternion state vector; inputting the first quaternion state vector into the target Kalman filter model to predict the first predicted quaternion state vector of the vehicle at the next discrete moment; and converting the first predicted quaternion state vector to obtain the second predicted attitude information.

[0009] In the above embodiments of the present application, by representing the first predicted attitude information in the quaternion state space, the target Kalman filter model can efficiently and accurately process the state prediction of the vehicle's attitude, and not only considers the attitude information of the vehicle during the prediction process, but also combines the influence of sensor noise and vehicle dynamics characteristics, improving the prediction accuracy of the attitude information.

[0010] Optionally, inputting the first predicted attitude information and the first predicted position information of the vehicle at the current moment into the kinematic model to predict the second predicted position information of the vehicle at the next discrete moment includes: inputting the first predicted attitude information and the first predicted position information into the kinematic model to predict the first predicted displacement, where the first predicted displacement is used to represent the displacement of the vehicle from the current moment to the next discrete moment according to the first predicted attitude information; and predicting the second predicted position information based on the position information of the vehicle at the current moment and the first predicted displacement.

[0011] In the above embodiments of the present application, by combining the attitude information of the vehicle and the position information of the vehicle, the position information of the vehicle at subsequent discrete moments can be predicted, which can greatly improve the accuracy of vehicle position prediction and ensure that the vehicle can safely and accurately park in a predetermined parking space in an automatic parking scenario.

[0012] Optionally, obtaining the first predicted attitude information of the vehicle at the current moment includes: obtaining the actual observation data of the vehicle at the current moment; determining the Kalman gain of the initial Kalman filter model based on the actual observation data, where the Kalman gain is used to represent the accuracy of the actual observation data; updating the initial predicted attitude information based on the Kalman gain and the actual observation data to obtain the first predicted attitude information.

[0013] In the above embodiments of the present application, through the fusion of real-time observation data and prediction information, the Kalman filter model can continuously calibrate and correct its estimation of the vehicle attitude, so as to provide a more stable and accurate vehicle attitude information prediction in dynamic environments such as automatic parking. By updating the state estimation and state covariance matrix using the latest observation data at each discrete moment, the Kalman filter ensures the real-time accuracy and long-term stability of the vehicle attitude information, meeting the strict requirements of automatic parking control for the accuracy of the relative position information of the vehicle itself.

[0014] Optionally, updating the initial predicted attitude information based on the Kalman gain and the actual observation data to obtain the first predicted attitude information includes: determining the update weight of the actual observation data based on the Kalman gain, where the update weight is used to represent the contribution degree of the actual observation data to updating the initial predicted attitude information to the first predicted attitude information; updating the initial predicted attitude information based on the update weight and the actual observation data to obtain the first predicted attitude information.

[0015] In the above embodiments of the present application, by dynamically adjusting the Kalman gain, the update weight of the actual observation data is adaptively determined, ensuring the reasonable use of the actual observation data in the initial attitude prediction, avoiding over-reliance on any single data source, and thus obtaining a more accurate and reliable attitude estimation.

[0016] Optionally, updating the initial predicted attitude information based on the update weight and the actual observation data to obtain the first predicted attitude information includes: updating the pitch angle and roll angle in the initial predicted attitude information based on the update weight and the acceleration sensor measurement data in the actual observation data to obtain the updated pitch angle and the updated roll angle; updating the heading angle in the initial predicted attitude information based on the update weight and the steering wheel angle measurement data in the actual observation data to obtain the updated heading angle.

[0017] In the above embodiments of the present application, by updating the pitch angle, roll angle, and heading angle of the initial predicted attitude information using the data of the acceleration sensor and the steering wheel angle sensor respectively, the accuracy and real-time performance of vehicle attitude information prediction can be ensured in the automatic parking scenario.

[0018] Optionally, the method for predicting the pose information of the vehicle further includes: respectively constructing a process noise covariance matrix, an observation noise covariance matrix, and a state estimation covariance matrix based on the sensor measurement errors of the vehicle; constructing an initial Kalman filter model based on the process noise covariance matrix, the observation noise covariance matrix, and the state estimation covariance matrix.

[0019] In the above embodiments of the present application, by combining the process noise covariance matrix, the observation noise covariance matrix, and the state estimation covariance matrix, the initial Kalman filter model can make a preliminary estimate of the vehicle's attitude information, and continuously optimize this estimate through subsequent observation updates to meet the high requirements for vehicle positioning accuracy during the automatic parking process.

[0020] Optionally, the method for predicting the pose information of the vehicle further includes: obtaining the initial attitude information of the vehicle at the initial moment, where the initial moment is the previous discrete moment of the current moment; inputting the initial attitude information into the initial Kalman filter model to predict the initial predicted attitude information of the vehicle at the current moment; updating the state covariance matrix in the initial Kalman filter model based on the actual observation data of the vehicle at the current moment to obtain the target Kalman filter model.

[0021] In the above embodiments of the present application, after predicting the vehicle's attitude information according to the initial Kalman filter model, the predicted attitude information can also be updated according to the actual observation data of the vehicle to improve the accuracy of the vehicle's attitude information.

[0022] Optionally, updating the state covariance matrix in the initial Kalman filter model based on the actual observation data of the vehicle at the current moment to obtain the target Kalman filter model includes: determining the Kalman gain of the initial Kalman filter model based on the actual observation data; updating the state covariance matrix in the initial Kalman filter model based on the Kalman gain and the fading factor parameter to obtain the target Kalman filter model, where the fading factor parameter is used to control the convergence speed of the state covariance matrix during the update process of the state covariance matrix.

[0023] In the above embodiments of the present application, by using the Kalman gain and the fading factor parameter to update the state covariance matrix in the initial Kalman filter model to obtain the target Kalman filter model, the accuracy of the target Kalman filter model in predicting the attitude information at subsequent discrete moments can be improved.

[0024] Optionally, after obtaining the second predicted pose information of the vehicle at the next discrete moment, the method for predicting the pose information of the vehicle further includes: updating the second predicted pose information based on the actual observation data of the vehicle at the next discrete moment to obtain the target predicted pose information of the vehicle at the next discrete moment; and updating the state covariance matrix in the target Kalman filter model based on the actual observation data at the next discrete moment to obtain the updated target Kalman filter model.

[0025] In the above embodiments of the present application, after obtaining the second predicted pose information of the vehicle at the next discrete moment, the second predicted pose information can also be updated according to the actual observation data of the vehicle at the next discrete moment to obtain the target predicted pose information of the vehicle at the next discrete moment, so as to improve the prediction accuracy of the pose information of the vehicle. In addition, the state covariance matrix in the target Kalman filter model can also be updated according to the actual observation data of the vehicle at the next discrete moment to obtain the updated target Kalman filter model. This iterative update method can improve the prediction accuracy of the Kalman filter model.

[0026] Optionally, the method for predicting the pose information of the vehicle further includes: predicting the pose information of the vehicle at subsequent discrete moments by using the updated target Kalman filter model.

[0027] In the above embodiments of the present application, the updated target Kalman filter model can be used to predict the pose information of the vehicle at subsequent discrete moments, so as to predict the pose information of the vehicle at each discrete moment.

[0028] According to another aspect of the embodiments of the present application, there is provided a device for predicting the pose information of a vehicle. The device includes: an acquisition module, configured to acquire the first predicted pose information of the vehicle at the current moment, where the first predicted pose information is obtained by updating the initial predicted pose information of the vehicle according to the actual observation data of the vehicle at the current moment, and the initial pose information is obtained by predicting the pose of the vehicle by using an initial Kalman filter model; a prediction module, configured to input the first predicted pose information into the target Kalman filter model to predict and obtain the second predicted pose information of the vehicle at the next discrete moment of the current moment, where the target Kalman filter model is obtained by updating the state covariance matrix in the initial Kalman filter model based on the actual observation data; and inputting the first predicted pose information and the first predicted position information of the vehicle at the current moment into the kinematic model to predict and obtain the second predicted position information of the vehicle at the next discrete moment.

[0029] According to another aspect of the embodiments of the present application, there is also provided an electronic device, including a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to execute the instructions to implement the above method.

[0030] According to another aspect of the embodiments of the present application, there is provided a computer-readable storage medium, in which a computer program is stored, wherein the computer program is set to execute the above method when being run by a processor.

[0031] According to another aspect of the embodiments of the present application, there is provided a computer program product, including a computer program, and the computer program implements the above method when being executed by a processor.

[0032] It should be noted that the above general description and the following detailed description are only for exemplifying and explaining the present application, and do not constitute a limitation to the present application. Description of the Drawings

[0033] Figure 1 is a flowchart of a method for predicting the pose information of a vehicle provided by an embodiment of the present application;

[0034] Figure 2 is an interaction diagram of another method for predicting the pose information of a vehicle provided by an embodiment of the present application;

[0035] Figure 3 is a schematic diagram of a device for predicting the pose information of a vehicle provided by an embodiment of the present application;

[0036] Figure 4 is a structural diagram of an electronic device provided by an embodiment of the present application. Detailed Embodiments

[0037] In order to make the technical problems, technical solutions and beneficial effects solved by the present application clearer, the present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0038] During the process of automatic parking for vehicle positioning, related technologies mainly rely on the measurement data of the vehicle's wheel speedometer, steering wheel angle sensor, and inertial measurement unit to predict the vehicle's attitude information and position information. For the measurement errors of sensors, simple filtering or compensation methods are usually adopted, but this method is difficult to effectively cope with the fluctuations of sensor performance in complex environments, thereby resulting in inaccurate vehicle positioning.

[0039] However, the embodiments of the present application provide a method for predicting the pose information of a vehicle, which includes obtaining the first predicted pose information of the vehicle at the current moment. The first predicted pose information is obtained by updating the initial predicted pose information of the vehicle based on the actual observation data of the vehicle at the current moment. The initial predicted pose information is obtained by using an initial Kalman filter model to predict the pose of the vehicle; inputting the first predicted pose information into a target Kalman filter model to predict the second predicted pose information of the vehicle at the next discrete moment at the current moment, where the target Kalman filter model is obtained by updating the state covariance matrix in the initial Kalman filter model based on the actual observation data; and inputting the first predicted pose information and the first predicted position information of the vehicle at the current moment into a kinematic model to predict the second predicted position information of the vehicle at the next discrete moment.

[0040] The above method for predicting the pose information of a vehicle provided by the embodiments of the present application achieves the following technical effects: When predicting the pose information of a vehicle, by fusing the observation data and the predicted pose information of the Kalman filter model in real time, the pose information of the vehicle is updated and optimized, which can significantly improve the accuracy of pose prediction of the vehicle in the automatic parking scenario. This process not only corrects the pose estimation deviation caused by sensor errors, environmental factors, and low model accuracy, but also ensures that the model can more accurately reflect the true pose of the vehicle by dynamically adjusting the state estimation covariance matrix of the Kalman filter model. In addition, by combining the optimized predicted pose information with the kinematic model, the position information of the vehicle at the next discrete moment is predicted. This strategy significantly improves the accuracy of the vehicle position information. During the automatic parking process, the dynamic behavior of the vehicle and the complexity of the road surface conditions require the prediction model to accurately capture and reflect the influence of the change in the vehicle pose on the position. By using the optimized pose information as the input, the kinematic model can more accurately predict the future position of the vehicle, thereby achieving high-precision tracking of the vehicle movement in the automatic parking control, improving the positioning accuracy and stability of the vehicle in a complex environment, achieving the technical effect of improving the positioning accuracy of the vehicle, and further solving the technical problem of low positioning accuracy of the vehicle.

[0041] Embodiment 1

[0042] The embodiments of the present application provide a method for predicting the pose information of a vehicle, which can be applied to the automatic parking scenario. Please refer to Figure 1 , and the method may include the following steps:

[0043] S110: Obtain the first predicted pose information of the vehicle at the current moment. The first predicted pose information is obtained by updating the initial predicted pose information of the vehicle based on the actual observation data of the vehicle at the current moment. The initial predicted pose information is obtained by using an initial Kalman filter model to predict the pose of the vehicle.

[0044] S120: Input the first predicted attitude information into the target Kalman filter model to predict the second predicted attitude information of the vehicle at the next discrete moment of the current moment, where the target Kalman filter model is obtained by updating the state covariance matrix in the initial Kalman filter model based on actual observation data; and input the first predicted attitude information and the first predicted position information of the vehicle at the current moment into the kinematic model to predict the second predicted position information of the vehicle at the next discrete moment.

[0045] In this embodiment, the current moment can be the next discrete moment after the initial moment, and the initial moment can be the moment when the vehicle starts. At the initial moment, the attitude information of the vehicle can be predicted by the initial Kalman filter model to obtain the initial predicted attitude information of the vehicle at the current moment. Among them, the initial Kalman filter model can be an error-state Kalman filter model, and one of the core ideas of this model is to combine the predicted attitude information and the actual observation information to obtain a more accurate attitude estimation.

[0046] Optionally, when predicting the initial attitude information of the vehicle relying on the initial Kalman filter model, due to the inaccuracy of the data used in the prediction process and the changes in the external environment, there are often deviations between the initially predicted attitude information predicted by the model and the true attitude information of the vehicle. In this case, the initially predicted attitude information corresponding to the current moment can be updated according to the actual observation data of the vehicle at the current moment, so that the updated first predicted attitude information is more accurate, where the actual observation data at the current moment is the attitude data of the vehicle collected by the vehicle's sensors.

[0047] For example, since the vehicle is in a dynamic driving process, based on this, after obtaining the initially predicted attitude information of the vehicle at the current moment, data from multiple vehicle sensors can be further collected as actual observation information, including but not limited to the accelerometer and gyroscope data in the six-axis inertial measurement unit sensor, as well as the vehicle's wheel speedometer and steering wheel angle information. These actual observation data reflect the true attitude and motion state of the vehicle at the current moment. After obtaining the actual observation data of the vehicle at the current moment, the initially predicted attitude information can be updated according to the actual observation data of the vehicle at the current moment, and then the first predicted attitude information of the vehicle at the current moment can be obtained. Among them, the first predicted attitude information can be represented as a set of Euler angles. For example, the set of Euler angles can be the pitch angle, roll angle, and yaw angle of the vehicle at the current moment.

[0048] Optionally, the initial Kalman filter model at least includes a state covariance matrix, where the state covariance matrix is a mathematical representation used in Kalman filtering to describe the uncertainty of system state estimation. As the vehicle moves and sensor data is collected, the uncertainty of the system is affected by dynamic environmental changes and sensor errors. Therefore, updating the state covariance matrix in the initial Kalman filter model according to the actual observation data to obtain the target Kalman filter model can improve the prediction accuracy of the target Kalman filter model for the attitude information of the vehicle at the next discrete moment.

[0049] Optionally, the state covariance matrix in the initial Kalman filter model can be updated based on the actual observation data and the fading factor parameter, and then the target Kalman filter model can be obtained.

[0050] Optionally, after obtaining the first predicted attitude information of the vehicle at the current moment and the target Kalman filter model, the first predicted attitude information can be input into the target Kalman filter model for prediction to obtain the second predicted attitude information of the vehicle at the next discrete moment.

[0051] For example, in Kalman filtering, the state transition equation is used in the prediction stage. Based on this, the first predicted attitude information of the vehicle at the current moment can be converted into a quaternion state vector using the Euler angle quaternion conversion formula, and then the quaternion state vector can be used as the input state vector and input into the target state Kalman filter model for prediction to obtain the second predicted attitude information of the vehicle at the next discrete moment of the current moment.

[0052] Optionally, when using the target Kalman filter model to predict the second attitude information of the vehicle at the next discrete moment, the error of the vehicle's gyroscope signal can also be considered. That is, the error of the vehicle's gyroscope signal is also converted into a quaternion state vector, combined with the quaternion state vector converted from the first predicted information, and input into the target Kalman filter model for prediction to improve the prediction accuracy of the target Kalman filter model for the attitude information of the vehicle.

[0053] Optionally, after predicting the second predicted attitude information of the vehicle at the next discrete moment of the current moment, the second predicted attitude information can also be updated according to the actual observation data when the vehicle travels to the next discrete moment according to the method described above to obtain more accurate vehicle attitude information.

[0054] Optionally, after obtaining the second predicted attitude information of the vehicle at the next discrete moment of the current moment, the method described above can be referred to for predicting the attitude information of the vehicle at subsequent discrete moments, which will not be elaborated here.

[0055] Optionally, the above-introduced prediction process of the vehicle's attitude information is followed by an introduction to the prediction process of the vehicle's position information.

[0056] Optionally, when predicting the vehicle's position information, a kinematic model can be used to predict the vehicle's position information. Here, taking the prediction of the vehicle's position information at the next discrete moment as an example, the prediction process of the vehicle's position information is introduced. Among them, the kinematic model can be represented by the vehicle's kinematic equation, and by considering the wheel speed, steering angle of the vehicle during driving, and the attitude information of the vehicle at the current moment, the position change of the vehicle at the next discrete moment is predicted. For example, based on the dynamic parameters of the vehicle at the current moment, such as wheel speed, steering angle, etc., combined with the first predicted attitude information of the vehicle at the current moment, the position increment (ΔX, ΔY, ΔZ) of the vehicle at the next moment is calculated, where the position increment is calculated by projecting the displacement in the coordinate system determined by the attitude angle corresponding to the first predicted attitude information, so as to obtain the second predicted position information of the vehicle at the next moment of the current moment.

[0057] Optionally, after obtaining the second predicted position information of the vehicle at the next discrete moment of the current moment, the position information of the vehicle at subsequent discrete moments can be determined with reference to this method, which will not be elaborated here.

[0058] Optionally, after obtaining the second predicted attitude information and the second predicted position information of the vehicle at the next discrete moment of the current moment, the second predicted attitude information and the second predicted position information can be used as the predicted pose information of the vehicle at the next discrete moment of the current moment.

[0059] In the above steps S110 to S120 of this application, when predicting the attitude information of the vehicle, by fusing the observation data and the predicted attitude information of the Kalman filter model in real time, the attitude information of the vehicle is updated and optimized, which can significantly improve the accuracy of attitude prediction of the vehicle in the automatic parking scenario. This process not only corrects the attitude estimation deviation caused by sensor errors, environmental factors, and model inaccuracies, but also ensures that the model can more accurately reflect the true attitude of the vehicle by dynamically adjusting the state estimation covariance matrix of the Kalman filter model. In addition, by combining the optimized predicted attitude information with the kinematic model, the position information of the vehicle at the next discrete moment is predicted. This strategy significantly improves the accuracy of the vehicle position information. During the automatic parking process, the complexity of the vehicle's dynamic behavior and road surface conditions requires the prediction model to accurately capture and reflect the impact of changes in the vehicle's attitude on the position. By using the optimized attitude information as input, the kinematic model can more accurately predict the future position of the vehicle, thereby achieving high-precision tracking of the vehicle's movement in automatic parking control, improving the positioning accuracy and stability of the vehicle in complex environments, achieving the technical effect of improving the positioning accuracy of the vehicle, and further solving the technical problem of inaccurate vehicle positioning.

[0060] The above method of this embodiment will be further introduced below.

[0061] As an optional implementation manner, in step S120, inputting the first predicted attitude information into the target Kalman filter model to predict the second predicted attitude information of the vehicle at the next discrete moment of the current moment includes: converting the first predicted attitude information into a first quaternion state vector; inputting the first quaternion state vector into the target Kalman filter model to predict the first predicted quaternion state vector of the vehicle at the next discrete moment; and converting the first predicted quaternion state vector to obtain the second predicted attitude information.

[0062] In this embodiment, in the automatic parking scenario, the attitude information of the vehicle can be represented by a set of Euler angles (including heading angle, pitch angle, roll angle). However, this representation has some limitations in mathematical operations, especially in representing the continuity of rotation and avoiding singularities (such as the four-axis roll problem). Based on this, the attitude information of the vehicle can be converted into a quaternion state vector for subsequent attitude prediction. It should be noted that the conversion of the quaternion state vector is essentially encoding the information in the Euler angles into the four components of the quaternion to support the calculation of the Kalman filter model.

[0063] For example, through the Euler angle quaternion conversion formula, the Euler angles corresponding to the first predicted attitude information of the vehicle at the current moment are converted into a quaternion state vector.

[0064] Optionally, after converting the first predicted attitude information of the vehicle at the current moment into a quaternion state vector, the quaternion state vector can be input into the target Kalman filter model to predict the first predicted quaternion state vector of the vehicle at the next discrete moment.

[0065] Optionally, when the target Kalman filter model predicts the attitude information of the vehicle at the next discrete moment, it will also consider factors such as vehicle dynamics, road conditions, and sensor noise, and their influence on the attitude information of the vehicle at the next discrete moment. That is to say, the prediction here is actually to estimate the prior distribution of the vehicle's attitude information at the next discrete moment through the mathematical operations of the state transition matrix and process noise in the target Kalman filter model.

[0066] Optionally, after obtaining the first predicted quaternion state vector of the vehicle at the next discrete moment through prediction by the target Kalman filter model, in order to make the first predicted quaternion state vector more intuitively applicable to vehicle control and path planning, it is usually necessary to convert it back to the Euler angle representation to obtain the second predicted attitude information of the vehicle at the next discrete moment. This conversion process uses the conversion formula from quaternion to Euler angle, ensuring the continuity of mathematical representation and the intuitiveness of physical understanding.

[0067] In the above steps, by representing the first predicted attitude information in the quaternion state space, the target Kalman filter model can efficiently and accurately process the state prediction of the vehicle's attitude, and not only considers the attitude information of the vehicle during the prediction process, but also combines the influence of sensor noise and vehicle dynamics characteristics, improving the accuracy of the predicted attitude information.

[0068] In the automatic parking scenario, inputting the first predicted attitude information and the first predicted position information into the kinematic model to predict the second predicted position information at the next discrete moment is an important step in realizing precise vehicle navigation and positioning. This step will be introduced next.

[0069] As an optional implementation manner, in step S120, inputting the first predicted attitude information and the first predicted position information of the vehicle at the current moment into the kinematic model to predict the second predicted position information of the vehicle at the next discrete moment includes: inputting the first predicted attitude information and the first predicted position information into the kinematic model to predict the first predicted displacement, where the first predicted displacement is used to represent the displacement of the vehicle from the current moment to the next discrete moment according to the first predicted attitude information; and predicting the second predicted position information based on the position information of the vehicle at the current moment and the first predicted displacement.

[0070] In this embodiment, the first predicted attitude information and the first predicted position information are the optimal attitude and position estimates obtained at the current moment based on sensor data (such as inertial measurement unit, wheel speedometer, steering wheel angle, etc.) and the Kalman filtering algorithm. These information reflect the precise state of the vehicle at the current moment, including its heading angle, pitch angle, roll angle, and position coordinates. After obtaining the first predicted attitude information and the first predicted position information, the first predicted attitude information and the first predicted position information can be input into the kinematic model to predict the displacement of the vehicle from the current moment to the next discrete moment according to the first predicted attitude information, that is, the first predicted displacement. The first predicted displacement not only includes the movement of the vehicle in the two-dimensional plane but also takes into account the height change due to pitch and roll in the three-dimensional space.

[0071] Optionally, after obtaining the first predicted displacement, the second predicted position information can be predicted according to the position information of the vehicle at the current moment and the first predicted displacement. This prediction process integrates vehicle dynamics information and sensor data, which can more accurately take into account the actual motion trajectory of the vehicle, especially in the complex path tracking and positioning process of automatic parking, providing a more reliable position prediction for the vehicle.

[0072] Optionally, by continuously iterating the above process, the attitude information and position information of the vehicle can be updated in real time, and then based on these information, the attitude information and position information of the vehicle at subsequent discrete moments can be predicted, so that the vehicle can continuously obtain its precise position during the automatic parking process, thereby achieving high-precision tracking of the parking path.

[0073] In this step, by combining the attitude information of the vehicle and the position information of the vehicle, the position information of the vehicle at subsequent discrete moments can be predicted, which can greatly improve the position prediction accuracy of the vehicle and ensure that the vehicle can safely and accurately park in the predetermined parking space in the automatic parking scenario.

[0074] As an alternative implementation manner, step S110 of obtaining the first predicted attitude information of the vehicle at the current moment includes: obtaining the actual observation data of the vehicle at the current moment; based on the actual observation data, determining the Kalman gain of the initial Kalman filter model, where the Kalman gain is used to represent the accuracy of the actual observation data; based on the Kalman gain and the actual observation data, updating the initial predicted attitude information to obtain the first predicted attitude information.

[0075] In this embodiment, obtaining the first predicted attitude information of the vehicle at the current moment is achieved through an iterative process. First, the actual observation data of the vehicle at the current moment is obtained through the vehicle's sensors (such as six-axis inertial measurement unit inertial navigation components, wheel speedometers, and steering wheel angle sensors, etc.). These data include but are not limited to the three-axis angular velocity of the gyroscope, the three-axis acceleration of the accelerometer, wheel speedometer information, and steering wheel angle information. These actual observation data are important representations of the vehicle's dynamic state, such as position, speed, and attitude, etc.

[0076] Optionally, after obtaining the actual observation data of the vehicle, the Kalman gain of the initial Kalman filter model can be determined based on the obtained actual observation data. Among them, the Kalman gain is a key parameter for how the initial Kalman filter model combines prediction information with actual observation information, reflecting the credibility or accuracy of the actual observation data for state estimation. In the framework of Kalman filtering, the Kalman gain is jointly determined by the state covariance matrix in the prediction stage, the observation model matrix, and the observation noise covariance matrix. Among them, the Kalman gain is used to guide how the filter effectively fuses prediction and observation information and corrects the state estimation to achieve the purpose of optimal estimation.

[0077] Optionally, after determining the Kalman gain, the initial predicted attitude information is updated using the Kalman gain and the actual observation data. This update process follows the principle of error-state Kalman filtering and involves the fusion of predicted states and observation data.

[0078] For example, the quaternion state equation is used for the prior estimation of the attitude angle, that is, the update of the predicted state. Then, the prior estimation is corrected using the accelerometer and steering wheel angle information. In this step, the Kalman gain will calculate the optimal correction amount according to the difference between the current observation data and the predicted state, and is used to update the initial predicted attitude information to obtain the first predicted attitude information.

[0079] In this step, through the fusion of real-time observation data and prediction information, the Kalman filter model can continuously calibrate and correct its estimation of the vehicle's attitude, so as to provide more stable and accurate vehicle attitude information prediction in dynamic environments such as automatic parking. By updating the state estimation and state covariance matrix using the latest observation data at each discrete moment, the Kalman filter ensures the real-time accuracy and long-term stability of the vehicle attitude information, meeting the strict requirements of the automatic parking control for the accuracy of the relative position information of the vehicle itself.

[0080] As an alternative implementation, the initial predicted attitude information is updated based on the Kalman gain and actual observation data to obtain the first predicted attitude information, including: determining an update weight for the actual observation data based on the Kalman gain, where the update weight is used to represent the contribution degree of the actual observation data to updating the initial predicted attitude information to the first predicted attitude information; updating the initial predicted attitude information based on the update weight and the actual observation data to obtain the first predicted attitude information.

[0081] In this embodiment, since the basic function of the Kalman gain is to balance the confidence between the prior estimate (initial predicted attitude information) and the actual observation data. For example, if the actual observation data is very reliable, then the Kalman gain will be large, which means that the actual observation data will have a greater impact on the initial predicted attitude information. On the contrary, if the actual observation data is unreliable or the prior estimate is already very accurate, the Kalman gain will be small, thereby reducing the correction effect of the actual observation data on the initial predicted attitude information. Based on this, the update weight of the actual observation data on the initial predicted attitude information can be determined according to the determined Kalman gain, and then the correction amplitude of the actual observation data on the prior state estimate can be determined.

[0082] Optionally, after determining the update weight of the actual observation data based on the Kalman gain, the initial predicted attitude information can be updated according to the update weight of the actual observation data and the actual observation data, and then the first predicted attitude information can be obtained.

[0083] In this step, by dynamically adjusting the Kalman gain, the update weight of the actual observation data is adaptively determined, ensuring the reasonable use of the actual observation data in the initial attitude prediction, avoiding over-reliance on any single data source, and thus obtaining a more accurate and reliable attitude estimate.

[0084] As an alternative implementation, the initial predicted attitude information is updated based on the update weight and the actual observation data to obtain the first predicted attitude information, including: updating the pitch angle and roll angle in the initial predicted attitude information respectively based on the update weight and the acceleration sensor measurement data in the actual observation data to obtain the updated pitch angle and the updated roll angle; updating the heading angle in the initial predicted attitude information based on the update weight and the steering wheel angle measurement data in the actual observation data to obtain the updated heading angle.

[0085] In this embodiment, when using the actual observation data to update the initial predicted attitude information, the update process of the heading angle will be separated from the update processes of the pitch angle and the roll angle to avoid damaging the accuracy of the heading angle during the correction process.

[0086] Optionally, in actual measurement data, the acceleration sensor can provide acceleration information of the vehicle in three-dimensional space, and this information is crucial for predicting the pitch angle and roll angle of the vehicle. Based on this, when the vehicle is in a dynamic driving state, the measurement data of the acceleration sensor is used to update the state vector in the Kalman filter, especially the part related to the pitch angle and roll angle. By fusing the measurement data of the sensor with the predicted state of the filter, the attitude prediction deviation caused by prediction errors or sensor noise can be corrected, so as to obtain the updated pitch angle and roll angle.

[0087] Optionally, the update of the heading angle usually depends on the measurement data of the vehicle's steering wheel angle. The measurement data of the steering wheel angle can reflect the actual steering action of the vehicle, thus providing an important basis for the change of the heading angle. In the observation update stage of the Kalman filter, the measurement data of the steering wheel angle is also given the corresponding update weight to correct the predicted state related to the heading angle in the filter. By optimizing the fusion strategy, it can ensure that even under complex driving conditions, the prediction of the heading angle can maintain high accuracy, avoiding the problem of path deviation that may occur due to inaccurate attitude prediction during the automatic parking process of the vehicle.

[0088] Optionally, after updating the pitch angle, roll angle, and heading angle in the initial attitude information, the updated pitch angle, roll angle, and heading angle can be used as the first predicted attitude information.

[0089] In this step, by updating the pitch angle, roll angle, and heading angle of the initial predicted attitude information using the data of the acceleration sensor and the steering wheel angle sensor respectively, it can ensure the accuracy and real-time performance of the vehicle attitude information prediction in the automatic parking scenario.

[0090] As an optional implementation manner, the method for predicting the pose information of the vehicle further includes: constructing a process noise covariance matrix, an observation noise covariance matrix, and a state estimation covariance matrix respectively based on the sensor measurement errors of the vehicle; constructing an initial Kalman filter model based on the process noise covariance matrix, the observation noise covariance matrix, and the state estimation covariance matrix.

[0091] In this embodiment, the process noise covariance matrix describes the random errors that may be introduced during the change process of the system state, that is, the inaccuracy of the dynamic model itself. In the automatic parking scenario, the movement of the vehicle may be affected by various factors, such as the unevenness of the road surface, wind resistance, tire slip or side slip, etc. These factors will cause a difference between the actual movement of the vehicle and the model prediction. When constructing the process noise covariance matrix, it is necessary to estimate according to the statistical characteristics (such as mean value, variance) of these uncertainties to ensure that the model can reasonably consider the random effects in the system dynamic process.

[0092] Optionally, the observation noise covariance matrix reflects the uncertainty of sensor measurements, which includes the measurement errors of sensors (such as the zero bias of the gyroscope and the quantization error of the accelerometer) and the influence of environmental interference on the measurement results. In the automatic parking scenario, there will be errors in the measurement results of the inertial measurement unit, wheel speed sensor, and steering wheel sensor. These errors may be caused by factors such as the inherent inaccuracy of the sensors, temperature changes, and electromagnetic interference. When constructing the observation noise covariance matrix, it is necessary to analyze the measurement errors of each sensor and determine the elements of the matrix according to their statistical characteristics (such as variance).

[0093] Optionally, the state estimation covariance matrix quantifies the uncertainty of the Kalman filter's estimation of the system state. At the initial stage of the automatic parking scenario, due to the lack of observations, the state estimation covariance matrix is often initialized to a relatively large value to indicate a high level of uncertainty in the initial estimation. As sensor data is continuously acquired and the Kalman filter algorithm is iteratively updated, the state estimation covariance matrix will gradually decrease to reflect the improvement in the filter's confidence in estimating the vehicle state.

[0094] Optionally, based on the sensor errors of the vehicle, a process noise covariance matrix, an observation noise covariance matrix, and a state estimation covariance matrix can be constructed, and then an initial Kalman filter model can be constructed based on the process noise covariance matrix, the observation noise covariance matrix, and the state estimation covariance matrix.

[0095] In this step, by combining the process noise covariance matrix, the observation noise covariance matrix, and the state estimation covariance matrix, the initial Kalman filter model can make a preliminary prediction of the vehicle's attitude information, and through subsequent observation updates, continuously optimize this prediction result to meet the high requirements for vehicle positioning accuracy during the automatic parking process.

[0096] As an alternative implementation, the method for predicting the pose information of the vehicle further includes: obtaining the initial pose information of the vehicle at the initial moment, where the initial moment is the previous discrete moment of the current moment; inputting the initial pose information into the initial Kalman filter model to predict the initial predicted pose information of the vehicle at the current moment; and updating the state covariance matrix in the initial Kalman filter model based on the actual observation data of the vehicle at the current moment to obtain the target Kalman filter model.

[0097] In this embodiment, the initial moment can be the moment when the vehicle starts. The initial pose information at this moment includes the vehicle's heading angle, pitch angle, and roll angle. After obtaining the initial pose information at the vehicle start moment, the initial pose information can be input into the pre-constructed initial Kalman filter model, and then based on the initial Kalman filter model, predict the initial predicted pose information of the vehicle at the current moment.

[0098] Optionally, since the vehicle is in a dynamic driving process during parking, after obtaining the initial predicted attitude information of the vehicle at the current moment, the initially predicted attitude information can be updated according to the actual observation data of the vehicle at the current moment to obtain the first predicted attitude information of the vehicle at the current moment.

[0099] In this step, after predicting the attitude information of the vehicle according to the initial Kalman filter model, the predicted attitude information can also be updated according to the actual observation data of the vehicle to improve the accuracy of the attitude information of the vehicle.

[0100] As an alternative implementation, the method for predicting the pose information of the vehicle further includes: determining the Kalman gain of the initial Kalman filter model based on the actual observation data; updating the state covariance matrix in the initial Kalman filter model based on the Kalman gain and the fading factor parameter to obtain a target Kalman filter model, where the fading factor parameter is used to control the convergence speed of the state covariance matrix during the update process of the state covariance matrix.

[0101] In this embodiment, the Kalman gain is a core parameter in the Kalman filter algorithm and is used to determine the weight of the actual observation data of the sensor in the state update. The fading factor is used to control the convergence speed of the state covariance matrix during the update process of the state covariance matrix to avoid excessive convergence of the state covariance matrix during the update process, thereby ensuring that the filter can still effectively update the state prediction result when the environmental conditions change slowly. In the scenarios of automatic parking and dynamic positioning, the introduction of the fading factor parameter helps the filter better adapt to environmental factors in reality such as temperature changes, vehicle aging, and sensor performance fluctuations, avoiding these long-term and small changes being ignored, so as to maintain the accuracy of the state prediction. The use of the fading factor adjusts the update process of the state covariance matrix and ensures that the prediction of the vehicle state does not ignore the information of the actual observation data.

[0102] Optionally, the state covariance matrix in the initial Kalman filter model is a key parameter for quantifying the uncertainty of the state prediction result in the filter, and its update is the core of the observation update stage in the Kalman filter algorithm. After determining the Kalman gain, the state covariance matrix in the initial Kalman model can be updated in combination with the actual observation data and the fading factor parameter, and then a target Kalman filter model can be obtained.

[0103] In this step, by using the Kalman gain and the fading factor parameter to update the state covariance matrix in the initial Kalman filter model to obtain a target Kalman filter model, the accuracy of the target Kalman filter model in predicting the attitude information at subsequent discrete moments can be improved.

[0104] As an alternative implementation, after obtaining the second predicted pose information of the vehicle at the next discrete moment, the method for predicting the pose information of the vehicle further includes: updating the second predicted pose information based on the actual observation data of the vehicle at the next discrete moment to obtain the target predicted pose information of the vehicle at the next discrete moment; and updating the state covariance matrix in the target Kalman filter model based on the actual observation data at the next discrete moment to obtain the updated target Kalman filter model.

[0105] In this embodiment, after obtaining the second predicted pose information of the vehicle at the next discrete moment, the second predicted pose information can also be updated based on the actual observation data of the vehicle at the next discrete moment to correct the second pose information. In addition, the target Kalman filter model can also be updated according to the actual observation data at the next discrete moment to obtain the updated target Kalman filter model.

[0106] Optionally, the update process of the target Kalman filter model can be performed iteratively, that is, when predicting the pose information at each subsequent discrete moment, the target Kalman filter model can also be updated according to the actual observation data at each subsequent discrete moment to continuously improve the prediction accuracy of the target Kalman filter model.

[0107] As an alternative implementation, the method for predicting the pose information of the vehicle further includes: predicting the pose information of the vehicle at subsequent discrete moments by using the updated target Kalman filter model.

[0108] In this embodiment, once the initial Kalman filter model is updated by fusing the actual observation data and the fading factor parameter to obtain the target Kalman filter model, the target Kalman filter model will contain a more accurate state estimate and an updated state covariance matrix. At subsequent discrete time points, the filter will use the latest state information and model parameters in the target model to predict the pose information of the vehicle at the next moment to improve the prediction accuracy of the vehicle's pose information.

[0109] The technical solution of the embodiment of the present invention will be illustrated below in conjunction with preferred implementations.

[0110] Figure 2 It is a flowchart of another scheduling method for compilation tasks according to an embodiment of the present application. As Figure 2 shown, the method includes the following steps:

[0111] Step S201, determine the reference coordinate system to be used.

[0112] In this embodiment, the reference coordinate system used in the calculation is determined. The navigation coordinate reference system used is generally specified to be the northeast celestial reference system, wherein the front, left, and top of the vehicle body in the navigation coordinate system correspond to the directions of the x, y, and z axes, and the Euler angle rotation order is yaw, pitch, and roll, which correspond to the rotation of the z, y, and x axes.

[0113] Step S202: When the vehicle is in a stationary state, predict the initial posture information of the vehicle.

[0114] In this embodiment, when the vehicle is initially stationary, the accelerometer observation data is used as a calculation basis for vehicle posture prediction. For example, the vehicle's Euler angles are calculated using the accelerometer observation data as initial vehicle posture information.

[0115] For example, the pitch angle pitch0 of the vehicle can be calculated by the following formula (1), the roll angle roll0 of the vehicle can be calculated by the following formula (2), and the yaw angle yaw0 of the vehicle can be calculated by the following formula (3), where the initial value of the yaw angle yaw0 can be set to 0.

[0116]

[0117] yaw0=0 (3)

[0118] Among them, a x 、a y 、a z The vehicle acceleration in different directions is measured by the accelerometer.

[0119] Step S203: converting the initial posture information of the vehicle into an initial state value.

[0120] In this embodiment, in the reference frame previously determined for use, the Euler angle quaternion conversion formula is used to convert the initial attitude angle (Euler angle) into a quaternion state vector, and the quaternion state vector is used as the initial state value, and the process noise covariance, observation noise covariance matrix and initialization state estimation covariance matrix are defined, and the process noise covariance, observation noise covariance matrix and initialization state estimation covariance matrix are used to construct an initial Kalman filter model.

[0121] Alternatively, the quaternion state vector obtained using the Euler angle quaternion conversion formula can be expressed as q0 = (w0, x0, y0, z0). The defined process noise covariance matrix can be expressed by the following formula (4), the observation noise covariance matrix can be expressed by the following formula (5), and the initialized state estimate covariance matrix can be expressed by the following formula (6).

[0122]

[0123] R = [ω a (5)

[0124] P = I (6)

[0125] Step S204, perform a priori prediction on the attitude information of the vehicle.

[0126] In this embodiment, a quaternion state space in discrete time form is used, and the quaternion state value is expressed as In addition, considering the errors of the three gyroscope signals extended to the state space, the space state can be expressed as:

[0127] Optionally, the extended state transition matrix can be expressed by the following formula (7).

[0128]

[0129] Among them, the above A 11 , A 12 , A 12 , A 21 are respectively expressed by the following formulas (8)-(11).

[0130]

[0131] A 12 = 0(4×3) (9)

[0132] A 21 = 0(3×4) (10)

[0133] A 22 = I(3) (11)

[0134] Optionally, the Jacobian matrix corresponding to the state transition matrix can be expressed by the following formula (12).

[0135]

[0136] Optionally, after obtaining the Jacobian matrix corresponding to the state transition matrix, the covariance matrix in the a priori prediction can be determined by the following formula (13).

[0137]

[0138] Step S205, perform posterior update on the attitude information of the vehicle.

[0139] In this embodiment, after prior prediction of the vehicle's attitude information, the prior prediction result can be posteriorly updated using the vehicle's actual observation data. In this update process, first, the gain of the error extended Kalman filter model can be determined using the vehicle's actual observation data, and then the weight for updating the vehicle's body attitude information using the vehicle's accelerometer signal can be determined using the Kalman gain. Furthermore, based on this weight and the vehicle's accelerometer signal, the predicted body attitude information can be updated.

[0140] Optionally, since there may be errors in the vehicle's actual observation data, based on this, when determining the gain of the error extended Kalman filter model using the vehicle's actual observation data, the credibility of the vehicle's actual observation data can be determined first through the following formulas (14)-(16).

[0141]

[0142] Optionally, after determining the credibility of the vehicle's actual observation data through the above formulas (14)-(16), the gain of the error extended Kalman filter model can be calculated through the following formula (17).

[0143]

[0144] Optionally, after obtaining the gain of the error extended Kalman filter model, the posterior state and covariance matrix can be updated based on this gain. Among them, in the update process, the update process of the heading angle in the body attitude information can be calculated separately from the update processes of the pitch angle and roll angle. For example, since accelerometer data can usually only calculate the pitch angle and roll angle, the pitch angle (pitch) and roll angle (roll) can be updated using the acceleration measurement data in the actual observation data. The heading angle (yaw) in the body attitude information can be updated using the vehicle's steering wheel data in the actual observation data.

[0145] Optionally, to avoid damaging the accuracy of the heading angle during the correction process, when updating the pitch angle and roll angle, the vehicle's heading angle can be defined as 0. Among them, the correction amount M can be defined through the following formula (18), and the vehicle's lateral angle can be defined as 0 using this correction amount M.

[0146]

[0147] Optionally, the posterior update state in Kalman filtering is:

[0148] Optionally, a fading factor is also introduced during the update process to prevent the covariance matrix from converging excessively during the update process, enabling slow changes caused by environmental factors such as temperature to affect the filtering. Among them, the updated state estimation covariance matrix can be expressed by the following formula (19).

[0149]

[0150] Optionally, when updating the vehicle's heading angle, to avoid the influence of the pitch angle and roll angle on the heading angle, the correction amount M' can also be defined to exclude the influence of the vehicle's pitch angle and roll angle.

[0151] Step S206, update the error state Kalman filter model.

[0152] In this embodiment, the error state Kalman filter model can be updated according to the actual observation data of the vehicle to improve the prediction accuracy of the error state Kalman filter model, and then the updated error state Kalman filter model is used to predict the attitude information of the vehicle at subsequent discrete moments.

[0153] Step S207, predict the position information of the vehicle.

[0154] In this implementation, since the prediction is carried out within discrete time, the change of the vehicle body attitude angle is discrete. At each calculation moment, the vehicle body attitude angle forms a definite plane in space. Assuming that the vehicle body attitude angle does not change before the next calculation moment, the vehicle will only move in the plane determined at this moment, and the vehicle body attitude will change when reaching the next calculation moment. By calculating the average wheel speed pulse of the vehicle's rear wheels between two moments, the vehicle displacement obtained can be projected onto the vehicle body attitude angle plane, and then the coordinate increment of the vehicle can be obtained. By accumulating the coordinate increments, the position of the vehicle body at each discrete moment can be obtained.

[0155] In the above steps S201 to S207, through the fusion of multi-sensor data and the optimized Kalman filter algorithm, high-precision real-time prediction of the vehicle's six-degree-of-freedom pose information is achieved. The extended error Kalman filter technology is used to accurately estimate the vehicle body attitude angle, effectively dealing with the influence of sensor errors and environmental changes on positioning. Subsequently, the heading angle estimation is further optimized through the steering wheel angle information, enhancing the vehicle's dynamic response ability. Finally, by combining the optimized attitude information with the wheel speed data, the vehicle position is accurately predicted, significantly improving the navigation performance during the automatic parking process, breaking through the limitations of traditional two-dimensional positioning, and being able to provide more stable and accurate pose information in three-dimensional space, providing a solid foundation for path tracking and control of the vehicle in complex parking environments, effectively solving the problem of inaccurate positioning in automatic parking, and improving the parking safety and efficiency.

[0156] Embodiment 2

[0157] The embodiment of the present application further provides a prediction device 30 for the pose information of a vehicle for implementing the prediction method of the pose information of the vehicle. Please refer to Figure 3 , including: an acquisition module 301, configured to acquire the first predicted pose information of the vehicle at the current moment, where the first predicted pose information is obtained by updating the initial predicted pose information of the vehicle according to the actual observation data of the vehicle at the current moment, and the initial pose information is obtained by using an initial Kalman filter model to predict the pose of the vehicle; a prediction module 302, configured to input the first predicted pose information into the target Kalman filter model to predict the second predicted pose information of the vehicle at the next discrete moment at the current moment, where the target Kalman filter model is obtained by updating the state covariance matrix in the initial Kalman filter model based on the actual observation data; and input the first predicted pose information and the first predicted position information of the vehicle at the current moment into the kinematic model to predict the second predicted position information of the vehicle at the next discrete moment.

[0158] In the above-mentioned prediction device for the pose information of the vehicle in the present application, when predicting the pose information of the vehicle, by fusing the observation data and the predicted pose information of the Kalman filter model in real time, the pose information of the vehicle is updated and optimized, which can significantly improve the accuracy of the pose prediction of the vehicle in the automatic parking scenario. This process not only corrects the pose estimation deviation caused by sensor errors, environmental factors, and model inaccuracies, but also ensures that the model can more accurately reflect the true pose of the vehicle by dynamically adjusting the state estimation covariance matrix of the Kalman filter model. In addition, by combining the optimized predicted pose information with the kinematic model, the position information of the vehicle at the next discrete moment is predicted. This strategy significantly improves the accuracy of the vehicle position information. During the automatic parking process, the complexity of the vehicle's dynamic behavior and road surface conditions requires the prediction model to accurately capture and reflect the influence of the change in the vehicle's pose on the position. By using the optimized pose information as the input, the kinematic model can more accurately predict the future position of the vehicle, thereby achieving high-precision tracking of the vehicle's movement in automatic parking control, improving the positioning accuracy and stability of the vehicle in a complex environment, achieving the technical effect of improving the positioning accuracy of the vehicle, and further solving the technical problem of inaccurate vehicle positioning.

[0159] The embodiment of the present application further provides an electronic device 40. Please refer to Figure 4 , including a memory 410 and a processor 420, where the memory 410 is used to store a computer program; the processor 420 is configured to execute the program stored on the memory 410 to implement the prediction method of the pose information of the vehicle introduced in any embodiment of the present application.

[0160] An embodiment of the present application also provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, it implements the method for predicting the pose information of a vehicle introduced in any embodiment of the present application.

[0161] An embodiment of the present application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the above-mentioned method for predicting the pose information of a vehicle.

[0162] In the present application, "a plurality" refers to two or more.

[0163] In the present application, unless otherwise clearly defined, terms should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the communication inside two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific situations.

[0164] The terms "first", "second", "third", "fourth", etc. (if any) in the present application are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence.

[0165] The term "and / or" in the present application is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in the present application generally represents an "or" relationship between the front and back associated objects.

[0166] If there is no special description, all steps of the present application can be carried out in sequence or randomly. For example, the method includes steps A and B, indicating that the method can include steps A and B carried out in sequence, or can also include steps B and A carried out in sequence. For example, it is mentioned that the method may further include step C, indicating that step C can be added to the method in any order. For example, the method can include steps A, B, and C, or can also include steps A, C, and B, or can also include steps C, A, and B, etc.

[0167] The above are only the preferred embodiments of the present application, and are not intended to limit the present application. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for predicting the pose information of a vehicle, characterized in that, Including: Obtain the first predicted attitude information of the vehicle at the current moment, where the first predicted attitude information is obtained by updating the initial predicted attitude information of the vehicle according to the actual observation data of the vehicle at the current moment, and the initial predicted attitude information is obtained by predicting the attitude of the vehicle using an initial Kalman filter model; Input the first predicted attitude information into a target Kalman filter model to predict the second predicted attitude information of the vehicle at the next discrete moment at the current moment, where the target Kalman filter model is obtained by updating the state covariance matrix in the initial Kalman filter model based on the actual observation data; and input the first predicted attitude information and the first predicted position information of the vehicle at the current moment into a kinematic model to predict the second predicted position information of the vehicle at the next discrete moment.

2. The method according to claim 1, characterized in that, Inputting the first predicted attitude information into a target Kalman filter model to predict the second predicted attitude information of the vehicle at the next discrete moment at the current moment includes: Convert the first predicted attitude information into a first quaternion state vector; Input the first quaternion state vector into the target Kalman filter model to predict the first predicted quaternion state vector of the vehicle at the next discrete moment; Convert the first predicted quaternion state vector to obtain the second predicted attitude information.

3. The method according to claim 1, characterized in that, Inputting the first predicted attitude information and the first predicted position information of the vehicle at the current moment into the kinematic model to predict the second predicted position information of the vehicle at the next discrete moment includes: Input the first predicted attitude information and the first predicted position information into the kinematic model to predict a first predicted displacement, where the first predicted displacement is used to represent the displacement of the vehicle traveling from the current moment to the next discrete moment according to the first predicted attitude information; Based on the position information of the vehicle at the current moment and the first predicted displacement, predict the second predicted position information.

4. The method according to claim 1, characterized in that, Obtain the first predicted attitude information of the vehicle at the current moment, including: Obtain the actual observation data of the vehicle at the current moment; Based on the actual observation data, determine the Kalman gain of the initial Kalman filter model, where the Kalman gain is used to represent the accuracy of the actual observation data; Based on the Kalman gain and the actual observation data, update the initial predicted attitude information to obtain the first predicted attitude information.

5. The method according to claim 4, wherein Based on the Kalman gain and the actual observation data, updating the initial predicted attitude information to obtain the first predicted attitude information includes: Based on the Kalman gain, determine the update weight of the actual observation data, where the update weight is used to represent the contribution degree of the actual observation data to updating the initial predicted attitude information to the first predicted attitude information; Update the initial predicted attitude information based on the updated weight and the actual observation data to obtain the first predicted attitude information.

6. The method according to claim 5, characterized in that, Updating the initial predicted attitude information based on the updated weight and the actual observation data to obtain the first predicted attitude information includes: Based on the updated weight and the acceleration sensor measurement data in the actual observation data, update the pitch angle and the roll angle in the initial predicted attitude information respectively to obtain the updated pitch angle and the updated roll angle; Based on the updated weight and the steering wheel angle measurement data in the actual observation data, update the heading angle in the initial predicted attitude information to obtain the updated heading angle.

7. The method according to claim 1, characterized in that, The method further includes: Construct a process noise covariance matrix, an observation noise covariance matrix, and a state estimation covariance matrix respectively based on the sensor measurement error of the vehicle; Construct the initial Kalman filter model based on the process noise covariance matrix, the observation noise covariance matrix, and the state estimation covariance matrix.

8. The method according to claim 7, characterized in that, The method further includes: Obtain the initial attitude information of the vehicle at the initial moment, where the initial moment is the previous discrete moment of the current moment; Input the initial attitude information into the initial Kalman filter model to predict the initial predicted attitude information of the vehicle at the current moment; Update the initial predicted attitude information based on the actual observation data of the vehicle at the current moment to obtain the first predicted attitude information.

9. The method according to claim 8, characterized in that, Updating the state covariance matrix in the initial Kalman filter model based on the actual observation data of the vehicle at the current moment to obtain the target Kalman filter model includes: Determine the Kalman gain of the initial Kalman filter model based on the actual observation data; Update the state covariance matrix in the initial Kalman filter model based on the Kalman gain and the fading factor parameter, where the fading factor parameter is used to control the convergence speed of the state covariance matrix during the update process of the state covariance matrix to obtain the target Kalman filter model.

10. The method according to claim 1, wherein After obtaining the second predicted attitude information of the vehicle at the next discrete moment, the method further includes: Update the second predicted attitude information based on the actual observation data of the vehicle at the next discrete moment to obtain the target predicted attitude information of the vehicle at the next discrete moment; and, Update the state covariance matrix in the target Kalman filter model based on the actual observation data at the next discrete moment to obtain the updated target Kalman filter model.

11. The method according to claim 10, characterized in that, The method further includes: Use the updated target Kalman filter model to predict the attitude information of the vehicle at subsequent discrete moments.

12. A prediction device for the pose information of a vehicle, characterized in that, Including: An acquisition module, configured to acquire first predicted attitude information of a vehicle at a current moment, where the first predicted attitude information is obtained by updating initial predicted attitude information of the vehicle according to actual observation data of the vehicle at the current moment, and the initial attitude information is obtained by predicting the attitude of the vehicle using an initial Kalman filter model; A prediction module, configured to input the first predicted attitude information into a target Kalman filter model to predict second predicted attitude information of the vehicle at the next discrete moment of the current moment, where the target Kalman filter model is obtained by updating a state covariance matrix in the initial Kalman filter model based on the actual observation data; and input the first predicted attitude information and first predicted position information of the vehicle at the current moment into a kinematic model to predict second predicted position information of the vehicle at the next discrete moment.

13. An electronic device, characterized in that, Comprising a processor and a memory, wherein The memory is used for storing a computer program; The processor is configured to execute the program stored on the memory to implement the method according to any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1-11 is implemented.

Images