Method and device for predicting course angle of vehicle, electronic equipment and automatic driving vehicle

Abstract

The invention provides a vehicle course angle prediction method and device, electronic equipment and an automatic driving vehicle. The method comprises the following steps: acquiring historical measurement data of a first inertial measurement unit and a second inertial measurement unit, performing zero offset compensation processing on the historical measurement data to obtain zero offset compensation data, and acquiring course angle data corresponding to the zero offset compensation data, the mounting directions of the horizontal shafts of the first inertial measurement unit and the second inertial measurement unit are opposite; constructing an initial course angle prediction model, and training the initial course angle prediction model by adopting the zero offset compensation data and the course angle data to obtain a course angle prediction model; and acquiring real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and inputting the real-time measurement data into the course angle prediction model to generate a course angle prediction value under the condition that the vehicle is detected to be in a static scene. The problem that in the prior art, the course angle obtained at low cost in the static scene of the vehicle is low in precision is solved.

Description

Technical Field

[0001] The present application relates to the technical field of vehicle heading angle prediction. Specifically, it relates to a method, device, computer-readable storage medium, electronic device, and autonomous vehicle for predicting a vehicle heading angle. Background Technique

[0002] For current L4-level autonomous driving applications (Robotaxi), the ability of high-precision positioning is essential. To achieve the goal of high-precision positioning, it is necessary to go through a positioning initialization process first, that is, to determine the initial attitude and initial position of the vehicle. Among them, the initial horizontal angle can be determined by the accelerometer of the inertial measurement unit, and the initial position can be obtained through GNSS. Only the initial heading cannot be obtained under the condition that the vehicle is not equipped with dual GNSS antennas. Since Robotaxi needs to be stationary to complete positioning initialization, although it can be obtained by matching lidar or camera with a high-precision feature map, this technology is limited by the surrounding environment and is easily affected by sparse map features and line-of-sight occlusion.

[0003] The prior art uses gyrocompass technology to rely on high-precision gyroscopes, which has the problem of high cost; the prior art also uses magnetometers to obtain the heading angle. Although the cost is very low, the magnetic field around the vehicle is complex and is greatly affected by surrounding vehicles, including hard magnetic fields and soft magnetic fields. Among them, hard magnetic fields need to be calibrated to compensate, which is not very convenient for in-vehicle use, and the problem of soft magnetic field interference is very serious, and the heading angle accuracy cannot be guaranteed. Summary of the Invention

[0004] The main purpose of the present application is to provide a method, device, computer-readable storage medium, electronic device, and autonomous vehicle for predicting a vehicle heading angle, so as to at least solve the problem that the heading angle obtained at low cost in the vehicle stationary scenario in the prior art has low accuracy.

[0005] To achieve the above object, according to one aspect of the present application, a method for predicting a vehicle heading angle is provided, including: obtaining historical measurement data of a first inertial measurement unit and a second inertial measurement unit, performing zero-bias compensation processing on the historical measurement data to obtain zero-bias compensation data, and obtaining heading angle data corresponding to the zero-bias compensation data, wherein the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same; constructing an initial heading angle prediction model, and training the initial heading angle prediction model at least using the zero-bias compensation data and the heading angle data to obtain a heading angle prediction model; obtaining real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, inputting the real-time measurement data into the heading angle prediction model to generate a heading angle prediction value.

[0006] Optionally, the historical measurement data includes gyroscope data, accelerometer data, and temperature data. After obtaining the historical measurement data of the first inertial measurement unit and performing zero-bias compensation processing on the historical measurement data to obtain zero-bias compensation data, the method further includes: taking the coordinate center point of the first inertial measurement unit as the origin, converting the zero-bias compensation data of the second inertial measurement unit to the coordinate system of the first inertial measurement unit to obtain the average value of the zero-bias compensation data of the first inertial measurement unit and the second inertial measurement unit, thereby obtaining fused zero-bias compensation data; performing an averaging process on the fused zero-bias compensation data with a sliding window of a preset length to obtain target zero-bias compensation data.

[0007] Optionally, during the process of training the initial heading angle prediction model using the zero-bias compensation data and the heading angle data, the method further includes: calculating the loss function gradient of the initial heading angle prediction model, updating the model parameters of the initial heading angle prediction model through an optimizer according to the loss function gradient, and using an evaluation index to determine whether the initial heading angle prediction model is trained, wherein the evaluation index includes the mean absolute error.

[0008] Optionally, before detecting that the vehicle is in a stationary scenario, the method further includes: obtaining data information of the vehicle, where the data information includes at least one of sensor data, vehicle speed data, and GPS data; determining whether the vehicle is in the stationary scenario according to the data information.

[0009] Optionally, before detecting that the vehicle is in a stationary scenario, the method further includes: when it is determined that the vehicle is in a non-stationary scenario, outputting the heading angle data of the vehicle by using multi-sensor fusion positioning.

[0010] Optionally, after inputting the real-time measurement data into the heading angle prediction model to generate a heading angle prediction value, the method further includes: before detecting that the vehicle shuts down and powers off, storing zero-offset data in the vehicle's memory, where the zero-offset data is obtained by multi-sensor fusion positioning, and the zero-offset data is used to perform zero-offset compensation processing on the real-time measurement data after the vehicle is restarted; training the initial heading angle prediction model at least using the zero-offset compensation data and the heading angle data, including: determining the gyroscope vibration data of the first inertial measurement unit and the second inertial measurement unit according to the historical measurement data, and training the initial heading angle prediction model using the gyroscope vibration data, the zero-offset compensation data, and the heading angle data.

[0011] Optionally, the initial heading angle prediction model includes a convolutional neural network and a BiLSTM network, and the heading angle prediction model is sequentially composed of an input layer, a one-dimensional convolutional layer, a max pooling layer, a bidirectional long short-term memory layer, an attention layer, a fully connected layer, and an output layer.

[0012] According to another aspect of the present application, a vehicle heading angle prediction device is provided, including: a first acquisition unit, configured to acquire historical measurement data of a first inertial measurement unit and a second inertial measurement unit, perform zero-offset compensation processing on the historical measurement data to obtain zero-offset compensation data, and acquire the heading angle data corresponding to the zero-offset compensation data, where the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same; a training unit, configured to construct an initial heading angle prediction model, and train the initial heading angle prediction model at least using the zero-offset compensation data and the heading angle data to obtain a heading angle prediction model; an input unit, configured to acquire the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and input the real-time measurement data into the heading angle prediction model to generate a heading angle prediction value when it is detected that the vehicle is in a stationary scenario.

[0013] According to still another aspect of the present application, a computer-readable storage medium is provided, where the computer-readable storage medium includes a stored program, and when the program runs, it controls the device where the computer-readable storage medium is located to execute any one of the vehicle heading angle prediction methods.

[0014] According to another aspect of the present application, an electronic device is provided, including: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include those for executing any one of the vehicle heading angle prediction methods.

[0015] According to another aspect of the present application, an autonomous vehicle is provided, including: a domain controller for executing any one of the vehicle heading angle prediction methods.

[0016] Applying the technical solution of the present application, historical measurement data of a first inertial measurement unit and a second inertial measurement unit are obtained, and zero-offset compensation processing is performed on the historical measurement data to obtain zero-offset compensation data, and heading angle data corresponding to the zero-offset compensation data is obtained, wherein the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same; an initial heading angle prediction model is constructed, and the initial heading angle prediction model is trained using the zero-offset compensation data and the heading angle data to obtain a heading angle prediction model; real-time measurement data of the first inertial measurement unit and the second inertial measurement unit are obtained, and when it is detected that the vehicle is in a stationary scenario, the real-time measurement data is input into the heading angle prediction model to generate a heading angle prediction value. By combining the data of two inertial measurement units with opposite directions, the accuracy and reliability of heading angle prediction are effectively improved. Especially in the vehicle stationary scenario, the heading angle prediction value generated by using the prediction model can overcome the influence of sensor zero-offset and improve the positioning accuracy of the autonomous vehicle in the initial startup stage. In addition, by storing the zero-offset data and performing compensation at the next startup, the robustness and continuity of the system are further enhanced, providing a solid foundation for the wide application of autonomous driving technology. This solution is not only applicable to autonomous vehicles, but also can be widely applied to various mobile devices and systems that require high-precision heading angle information, with significant technical and economic value. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The specification drawings constituting a part of the present application are used to provide a further understanding of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation of the present application. In the drawings:

[0018] Figure 1 A hardware structure block diagram of a mobile terminal for executing a vehicle heading angle prediction method provided in an embodiment of the present application is shown;

[0019] Figure 2 A flowchart showing a vehicle heading angle prediction method provided in an embodiment of the present application is shown;

[0020] Figure 3 Shows the installation schematic diagram of two inertial measurement units (IMUs) provided according to an embodiment of the present application;

[0021] Figure 4 Shows the schematic flow diagram of the training and prediction of the CNN + BiLSTM model provided according to an embodiment of the present application;

[0022] Figure 5 Shows the structural block diagram of a vehicle heading angle prediction device provided according to an embodiment of the present application.

[0023] Wherein, the above-mentioned drawings include the following reference numerals:

[0024] 102, processor; 104, memory; 106, transmission device; 108, input / output device. Detailed implementation manners

[0025] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0026] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0027] It should be noted that 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 do not necessarily need to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so as to describe the embodiments of the present application here. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0028] As introduced in the background art, there is a problem of low accuracy in the heading angle obtained at low cost in the vehicle stationary scenario. To solve the problem of low accuracy in the heading angle obtained at low cost in the vehicle stationary scenario, embodiments of the present application provide a method, an apparatus, a computer-readable storage medium, an electronic device, and an autonomous vehicle for predicting the vehicle heading angle.

[0029] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention.

[0030] The method embodiments provided in the embodiments of the present application can be executed on a mobile terminal, a computer terminal, or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 is a hardware structure block diagram of a mobile terminal for a method of predicting the vehicle heading angle according to an embodiment of the present invention. As Figure 1 shown, the mobile terminal may include one or more ( Figure 1 only one is shown in Figure 1 a processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Among them, the above mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown in Figure 1 is only schematic and does not limit the structure of the above mobile terminal. For example, the mobile terminal may further include more or fewer components than

[0031] The memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the vehicle heading angle prediction method in the embodiments of the present invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, the above-mentioned method is implemented. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories may be connected to the mobile terminal through a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the mobile terminal. In one instance, the transmission device 106 includes a network adapter (Network Interface Controller, abbreviated as NIC), which can be connected to other network devices through a base station and thus can communicate with the Internet. In one instance, the transmission device 106 may be a radio frequency (Radio Frequency, abbreviated as RF) module, which is used to communicate with the Internet wirelessly.

[0032] In this embodiment, a vehicle heading angle prediction method running on a mobile terminal, a computer terminal, or a similar computing device is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0033] Figure 2 It is a flowchart of the vehicle heading angle prediction method according to an embodiment of the present application. As Figure 2 shown, the method includes the following steps:

[0034] Step S201, obtain the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, perform zero-bias compensation processing on the above historical measurement data to obtain zero-bias compensation data, and obtain the heading angle data corresponding to the above zero-bias compensation data, where the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same;

[0035] Among them, the historical measurement data includes gyroscope and accelerometer data (angular velocity X-axis, angular velocity Y-axis, angular velocity Z-axis, acceleration X, acceleration Y, acceleration Z) and temperature data, and zero-bias compensation processing is performed on the gyroscope and accelerometer data; the horizontal axis of the inertial measurement unit IMU includes the X-axis and the Y-axis, and the vertical axis is the Z-axis. The specific installation schematic diagrams of the first inertial measurement unit IMU0 and the second inertial measurement unit IMU1 are as Figure 3 shown.

[0036] Specifically, the heading angle data is obtained through multi-sensor (RTK / IMU / LiDAR / Camera / HDMap) fusion positioning, and the zero bias is estimated online through multi-sensor fusion positioning. Through the complementarity of the dual IMU, the zero bias of the same type of IMU can cancel each other out, improving the accuracy of heading angle prediction when the vehicle is stationary, which is crucial for the positioning of autonomous vehicles in scenarios such as waiting at traffic lights and traffic congestion.

[0037] Step S202: Construct an initial heading angle prediction model, and train the initial heading angle prediction model at least using the above zero-bias compensation data and heading angle data to obtain a heading angle prediction model;

[0038] In addition, the vibration condition of the gyroscope can be determined by calculating the standard deviation of the angular velocity data of the X-axis, Y-axis, and Z-axis of the gyroscope. Combining the vibration condition of the gyroscope with the zero-bias compensation data and heading angle data to train the initial heading angle prediction model can effectively improve the training accuracy of the heading angle prediction model, so that the heading angle prediction value output by the heading angle prediction model is more accurate.

[0039] Specifically, the heading angle prediction model adopts the architecture of CNN+BiLSTM. Among them, the convolutional neural network CNN (Convolutional Neural Network) extracts the local features of the gyroscope time series data and inputs them to the BiLSTM (Bidirection Long-Short-Term Memory), that is, the bidirectional LSTM, to realize the long-term feature dependence relationship in the front and back directions, and the relationship between the gyroscope angular velocity and the heading angle can be learned. The reason why heading alignment can be achieved is that after learning the zero-bias instability law of the low-cost gyroscope through CNN+BiLSTM, the angular velocity of the earth's rotation is sensed, and then the heading angle is obtained.

[0040] Step S203: Obtain the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, input the real-time measurement data into the heading angle prediction model to generate a heading angle prediction value.

[0041] Specifically, by utilizing two inertial measurement units with opposite installation directions, the accuracy and stability of heading angle prediction are effectively improved. Especially in the vehicle stationary scenario, it can provide a more accurate heading angle prediction value, significantly enhancing the positioning accuracy of autonomous vehicles at low speeds or in a stationary state. In addition, by storing the bias data and using it for subsequent real-time measurement data processing, the system initialization time is reduced, and the adaptability of the vehicle in different environments is enhanced. This solution has broad application prospects in the fields of autonomous driving, intelligent transportation systems, etc., and can effectively improve the safety and driving experience of vehicles.

[0042] Through this embodiment, historical measurement data of the first inertial measurement unit and the second inertial measurement unit are obtained, and zero-bias compensation processing is performed on the historical measurement data to obtain zero-bias compensation data, and heading angle data corresponding to the zero-bias compensation data is obtained. Among them, the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same; an initial heading angle prediction model is constructed, and the initial heading angle prediction model is trained using the zero-bias compensation data and the heading angle data to obtain a heading angle prediction model; when the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit are obtained and it is detected that the vehicle is in a stationary scenario, the real-time measurement data are input into the heading angle prediction model to generate a heading angle prediction value. By combining the data of two inertial measurement units with opposite directions, the accuracy and reliability of heading angle prediction are effectively improved. Especially in the vehicle stationary scenario, the heading angle prediction value generated by using the prediction model can overcome the influence of sensor bias and improve the positioning accuracy of autonomous vehicles at the initial stage of startup. In addition, by storing the zero-bias data and compensating it at the next startup, the robustness and continuity of the system are further enhanced, providing a solid foundation for the wide application of autonomous driving technology. This solution is not only applicable to autonomous vehicles but also can be widely applied to various mobile devices and systems that require high-precision heading angle information, with significant technical and economic value.

[0043] In the specific implementation process, the above historical measurement data includes gyroscope data, accelerometer data, and temperature data. After obtaining the zero-bias compensation data by performing zero-bias compensation processing on the historical measurement data of the first inertial measurement unit and the above, the method further includes: taking the coordinate center point of the first inertial measurement unit as the origin, converting the zero-bias compensation data of the second inertial measurement unit into the coordinate system of the first inertial measurement unit to obtain the average value of the zero-bias compensation data of the first inertial measurement unit and the second inertial measurement unit, so as to obtain fused zero-bias compensation data; performing an averaging process on the fused zero-bias compensation data with a sliding window of a preset length to obtain target zero-bias compensation data.

[0044] This method comprehensively determines whether the current vehicle is stationary based on the vehicle wheel speed data and the speed data obtained from multi-sensor fusion positioning (for example, when the vehicle speed < 0.003 m / s, it is considered that the vehicle is stationary). If it is stationary, taking the center of IMU0 as the origin, first perform zero-bias compensation on the gyroscope and accelerometer data (angular velocity X-axis, angular velocity Y-axis, angular velocity Z-axis, acceleration X, acceleration Y, acceleration Z) of the two IMUs. The zero bias is obtained by online estimation through multi-sensor fusion positioning. Then, transform IMU 1 to the IMU0 coordinate system through the external parameters, and then average the angular velocity data and accelerometer data of the two IMUs to obtain the fused angular velocity and acceleration data. Similarly, obtain the temperature data. Then, take a 30-second time length as the sliding window to average the fused angular velocity, acceleration data, and temperature data of the two IMUs respectively, obtaining the gyroscope zero bias and accelerometer zero bias sequence data sampled by time, including temperature data. At the same time, in this sequence data, the heading angle output by multi-sensor fusion positioning at each moment is used as the ground truth for training. By fusing the data of different sensors, the error can be further reduced, and the stability and reliability of the prediction model can be improved, which has significant advantages in scenarios such as vehicle navigation systems and unmanned aerial vehicle flight control that require high-precision attitude information.

[0045] Specifically, in the process of training the above initial heading angle prediction model using the above zero-bias compensation data and heading angle data, the above method further includes: calculating the loss function gradient of the above initial heading angle prediction model, and according to the above loss function gradient, updating the model parameters of the above initial heading angle prediction model through an optimizer, and using an evaluation index to determine whether the above initial heading angle prediction model is trained. Among them, the above evaluation index includes the mean absolute error. This training method based on deep learning can automatically adjust the model parameters and optimize the prediction results, and is applicable to heading angle prediction in various complex environments, such as urban roads and mountain roads, improving the adaptability and safety of autonomous vehicles.

[0046] More specifically, before detecting that the above autonomous vehicle is in a stationary scenario, the above method further includes: obtaining the data information of the above vehicle, where the above data information includes at least one of sensor data, vehicle speed data, and GPS data; determining whether the above vehicle is in the above stationary scenario according to the above data information.

[0047] Among them, the sensors for determining whether the vehicle is in a stationary scenario include:

[0048] 1) Radar: It can detect the objects around the vehicle, including stationary and moving objects. By analyzing the reflected signals, it can be determined whether the vehicle is stationary;

[0049] 2) Lidar: By emitting laser beams and measuring the time it takes for the beams to reflect back, the distance between the vehicle and surrounding objects can be accurately measured, thereby determining whether the vehicle is stationary;

[0050] 3) Ultrasonic sensor: Suitable for short - distance object detection, it can sense whether the vehicle is approaching an obstacle or is stationary;

[0051] 4) Camera: Through image recognition technology, it can analyze the environment around the vehicle, including whether the vehicle is moving;

[0052] 5) Global Positioning System (GPS): By using GPS data, the position change of the vehicle can be detected, thereby determining whether the vehicle is moving.

[0053] Further, before detecting that the above - mentioned autonomous vehicle is in a stationary scenario, the above - mentioned method further includes: when determining that the above - mentioned autonomous vehicle is in a non - stationary scenario, using multi - sensor fusion positioning to output the heading angle data of the above - mentioned autonomous vehicle. This multi - sensor fusion positioning technology can provide high - precision heading angle information when the vehicle is driving at high speed. For autonomous vehicles on highways, it can ensure the safety and stability of their driving.

[0054] Even further, after inputting the above - mentioned real - time measurement data into the above - mentioned heading angle prediction model to generate a heading angle prediction value, the above - mentioned method further includes: before detecting that the above - mentioned vehicle shuts down and powers off, storing the zero - bias data in the memory of the above - mentioned vehicle, where the above - mentioned zero - bias data is obtained by multi - sensor fusion positioning, and the above - mentioned zero - bias data is used to perform zero - bias compensation processing on the real - time measurement data after the above - mentioned vehicle is restarted next time.

[0055] The storage and utilization of the zero - bias data by this method reduce the time for system recalibration, improve the response speed when the vehicle starts, and can significantly improve the operation efficiency for public transportation such as urban buses and taxis that start and stop frequently.

[0056] Training the above - mentioned initial heading angle prediction model using at least the above - mentioned zero - bias compensation data and heading angle data includes: determining the gyroscope vibration data of the above - mentioned first inertial measurement unit and the above - mentioned second inertial measurement unit according to the above - mentioned historical measurement data, and training the above - mentioned initial heading angle prediction model using the above - mentioned gyroscope vibration data, the above - mentioned zero - bias compensation data, and the heading angle data.

[0057] This method determines the vibration condition of the gyroscope by calculating the standard deviation of the angular velocity data of the X - axis, Y - axis, and Z - axis of the gyroscope. Combining the vibration condition of the gyroscope with the zero - bias compensation data and the heading angle data to train the initial heading angle prediction model can effectively improve the training accuracy of the heading angle prediction model, so that the heading angle prediction value output by the heading angle prediction model is more accurate.

[0058] Specifically, the above initial heading angle prediction model includes a convolutional neural network and a BiLSTM network. The above heading angle prediction model is sequentially composed of an input layer, a one-dimensional convolutional layer, a max pooling layer, a bidirectional long short-term memory layer, an attention layer, a fully connected layer, and an output layer connected together.

[0059] The process of predicting the heading angle using the heading angle prediction model specifically includes: inputting the gyroscope zero bias, accelerometer zero bias data, and temperature data of the time series, with the data dimension feature being 8-dimensional and the length being 450, into a 1D Cov1D convolutional neural network model through the input layer first. 64 filters are used, the filter size is 3, the activation function is "ReLU", and "same" padding is used to keep the output size the same as the input. Immediately followed is a max pooling layer (MaxPooling1D) with a pooling window size of 2, which is used to reduce the data dimension and extract features. The output of the pooling layer is input into a 3-layer BiLSTM (bidirectional LSTM) model. Among them, the first layer is a bidirectional LSTM layer (Bidirection) with 64 units, the second layer is a bidirectional LSTM layer with 32 units, and the third layer is a bidirectional LSTM layer with 16 units. After passing through the 3-layer BiLSTM, it passes through an attention layer, which helps the model pay more attention to the importance of different time steps when synthesizing information. Then, the output processed by the attention mechanism is concatenated with the output of the BiLSTM. In this way, the model can not only utilize the features extracted by the BiLSTM but also utilize the sequence information strengthened by the attention mechanism. Finally, the data passes through a fully connected layer (Dense) with 8 units and the activation function is "ReLU". Finally, there is an output layer, which is a fully connected layer with one unit and is used for the one-dimensional heading angle regression task.

[0060] This deep learning model structure can effectively extract and process the temporal features in historical data, improve the prediction accuracy, and has significant performance advantages for scenarios that require real-time prediction of the heading angle, such as automatic parking, path planning, etc.

[0061] In order to enable those skilled in the art to more clearly understand the technical solution of the present application, the implementation process of the vehicle heading angle prediction method of the present application will be described in detail below in combination with specific embodiments.

[0062] This embodiment relates to a specific vehicle heading angle prediction method, which specifically includes the following content:

[0063] Multi-sensor (RTK / IMU / LiDAR / Camera / HDMap) fusion positioning for driverless vehicles can obtain the heading angle of the vehicle. Through supervised learning, this heading angle is used as the ground truth to train the gyroscope data of a low-cost IMU (Inertial Measurement Unit). The training model adopts the architecture of CNN+BiLSTM. Among them, the CNN (Convolutional Neural Network) convolutional neural network extracts the local features of the gyroscope time series data and inputs them to the BiLSTM (Bidirectional Long Short-Term Memory), that is, the bidirectional LSTM, to achieve long-term feature dependence in the forward and backward directions, and the relationship between the gyroscope angular velocity and the heading angle can be learned. The reason for achieving heading alignment is that after learning the zero-bias instability law of the low-cost gyroscope through CNN+BiLSTM, the perception of the earth's angular velocity of rotation is realized, and then the heading angle is obtained.

[0064] The driverless autonomous vehicle uses two IMUs (IMU0 is the first inertial measurement unit and IMU1 is the second inertial measurement unit) as redundant. And in order to make the zero biases of the same model of IMUs cancel each other out, during the installation process, the X-axes of the two IMUs are installed in opposite directions, and the Y-axes of the two IMUs are installed in opposite directions. The specific installation schematic diagram is as Figure 3 shown.

[0065] Data preparation process:

[0066] Based on the vehicle wheel speed data and the speed data obtained from multi-sensor fusion positioning, it is comprehensively judged whether the current vehicle is stationary (for example, when the vehicle speed < 0.003 m / s, it is considered that the vehicle is stationary). If it is stationary, taking the center of IMU0 as the origin, first perform zero-bias compensation on the gyroscope and accelerometer data (angular velocity X-axis, angular velocity Y-axis, angular velocity Z-axis, acceleration X, acceleration Y, acceleration Z) of the two IMUs. The zero bias is obtained by online estimation through multi-sensor fusion positioning. Then, IMU1 is transformed to the IMU0 coordinate system through the extrinsic parameters. Then, the angular velocity data and accelerometer data of the two IMUs are averaged to obtain the fused angular velocity and accelerometer data. Similarly, the temperature data is obtained. Then, with a sliding window of 30 seconds in length, the fused angular velocity, accelerometer data, and temperature data of the two IMUs are averaged respectively to obtain the gyroscope zero-bias and accelerometer zero-bias sequence data sampled by time, including temperature data. At the same time, in this sequence data, the heading angle output by multi-sensor fusion positioning at each moment is used as the ground truth for training.

[0067] The model training process is as Figure 4 shown, and specifically includes the following content:

[0068] 1), By calculating the standard deviation of the angular velocity data of the X-axis, Y-axis, and Z-axis of the gyroscope, the vibration data of the gyroscope is determined. The zero-bias data of the gyroscope, the zero-bias data of the accelerometer, and the temperature data in the time series, with a data dimension feature of 8 dimensions and a length of 450, as well as the gyroscope vibration data are first input into a 1D Cov1D convolutional neural network model. 64 filters are used, the filter size is 3, the activation function is "ReLU", and "same" padding is used to keep the output size the same as the input. Immediately followed by a max pooling layer (MaxPooling1D) with a pooling window size of 2, which is used to reduce the data dimension and extract features. Among them, using the gyroscope vibration data for model training can improve the training accuracy of the model.

[0069] 2), The output of the pooling layer is input into a 3-layer BiLSTM (bidirectional LSTM) model. Among them, the first layer is a bidirectional LSTM layer (Bidirection) with 64 units, the second layer is a bidirectional LSTM layer with 32 units, and the third layer is a bidirectional LSTM layer with 16 units.

[0070] 3), After passing through 3 layers of BiLSTM, then passing through an attention layer, which helps the model pay more attention to the importance of different time steps when synthesizing information. Then, the output processed by the attention mechanism is concatenated with the output of BiLSTM. In this way, the model can not only utilize the features extracted by BiLSTM but also utilize the sequence information strengthened by the attention mechanism. Finally, the data passes through a fully connected layer (Dense) with 8 units, and the activation function is "ReLU".

[0071] 4), Finally, the output layer is a fully connected layer with one unit, which is used for a one-dimensional heading angle regression task.

[0072] Regarding the setting of the filters of the convolutional layer (Conv1D):

[0073] During the training process, the model uses the Adam optimizer, and its learning rate is adjusted through an exponential decay strategy. The initial learning rate is 0.001, and it decays once every 10000 steps with a decay rate of 0.96. The model uses a custom angular_mean_squared_error as the loss function and uses the mean absolute error (MAE) as the evaluation metric. The definition of the loss function is as follows:

[0074] Function AngularMeanSquaredError(y_true,y_predict):

[0075] err = sin(y_true - y_predict);

[0076] loss = mean(Rad2Deg(err) * Rad2Deg(err));

[0077] Regarding the relationship between the loss obtained by calculating the custom loss function and the MAE (Mean Absolute Error) evaluation metric: angular_mean_squared_error is a custom loss function used to calculate the difference between the model's predicted values and the true values. The output of this loss function (i.e., loss) will be used to guide the update of the model's parameters. Evaluation metrics specify that during the training and validation processes, in addition to the loss function, additional evaluation metrics need to be calculated. Here, it is the mean absolute error (MAE). MAE calculates the average of the absolute values of the differences between the predicted values and the true values, which provides another perspective on the model's performance.

[0078] Among them, the loss function is the main objective to be optimized during the training process. The model learns parameters by minimizing the loss function. It directly affects the update of the model's parameters. Evaluation metrics are additional tools for measuring the model's performance. They do not directly affect the update of the model's parameters but provide additional information about the model's performance. MAE is an evaluation metric that can help us understand the average error size of the model when predicting the heading angle.

[0079] During the training process, the value of the loss function will be used to calculate the gradient, and then the model's parameters will be updated through the optimizer. At the same time, evaluation metrics (such as MAE) will be calculated after each epoch to monitor the model's performance. If the value of MAE decreases, it means that the model's predictions are getting closer to the true values, even though it is not an indicator directly used for parameter update.

[0080] In summary, angular_mean_squared_error, as the loss function, directly participates in the optimization process of the model's parameters, while the evaluation metric MAE provides additional information about the model's performance, helping us evaluate and monitor the training effect of the model.

[0081] The model prediction process is as Figure 4 shown, and specifically includes the following content:

[0082] The model is completed through offline training. In real-time, the sequence data (angular velocity X-axis, angle Y-axis, angular velocity Z-axis, acceleration X-axis, acceleration Y-axis, acceleration Z-axis) and temperature data of the gyroscope and accelerometer obtained by averaging with a 30-second sliding window, with a length of 450, are input to the model, and the current heading angle can be predicted.

[0083] The embodiments of this application have the following technical effects:

[0084] 1), Training a low-cost MEMS IMU using big data from autonomous driverless vehicles to enable it to have heading alignment ability statically;

[0085] 2), Obtaining IMU data by symmetrically installing dual-redundant IMUs for static heading alignment;

[0086] 3), Only relying on a low-cost MEMS IMU can achieve an alignment accuracy of about 0.5 deg for the heading angle;

[0087] 4), Training the bias instability of a low-cost MEMS IMU with CNN + BiLSTM;

[0088] 5), Training a low-cost MEMS IMU to sense the earth's angular velocity of rotation with CNN + BiLSTM;

[0089] 6), CNN realizes the extraction of static local features of angular velocity and acceleration data;

[0090] 7), BiLSTM realizes the learning of the law of bias instability of a low-cost MEMS IMU.

[0091] The embodiment of the present application also provides a prediction device for the vehicle heading angle. It should be noted that the prediction device for the vehicle heading angle in the embodiment of the present application can be used to execute the prediction method for the vehicle heading angle provided in the embodiment of the present application. This device is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0092] The following introduces the prediction device for the vehicle heading angle provided in the embodiment of the present application.

[0093] Figure 5 is a schematic diagram of the prediction device for the vehicle heading angle according to the embodiment of the present application. As Figure 5 shown, the device includes:

[0094] A first acquisition unit 51, configured to acquire historical measurement data of a first inertial measurement unit and a second inertial measurement unit, perform bias compensation processing on the historical measurement data to obtain bias-compensated data, and acquire heading angle data corresponding to the bias-compensated data, wherein the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same;

[0095] A training unit 52 is configured to build an initial heading angle prediction model and train the initial heading angle prediction model with the above-mentioned zero-bias compensation data and heading angle data to obtain a heading angle prediction model.

[0096] An input unit 53 is configured to obtain real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and input the real-time measurement data into the heading angle prediction model to generate a heading angle prediction value when it is detected that the vehicle is in a stationary scenario.

[0097] In this embodiment, a first acquisition unit is configured to obtain historical measurement data of the first inertial measurement unit and the second inertial measurement unit, perform zero-bias compensation processing on the historical measurement data to obtain zero-bias compensation data, and obtain heading angle data corresponding to the zero-bias compensation data, where the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same; a training unit is configured to build an initial heading angle prediction model and train the initial heading angle prediction model with the above-mentioned zero-bias compensation data and heading angle data to obtain a heading angle prediction model; an input unit is configured to obtain real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and input the real-time measurement data into the heading angle prediction model to generate a heading angle prediction value when it is detected that the vehicle is in a stationary scenario. By combining data from two inertial measurement units with opposite directions, the accuracy and reliability of heading angle prediction are effectively improved. Especially in the vehicle stationary scenario, the heading angle prediction value generated by using the prediction model can overcome the influence of sensor zero bias and improve the positioning accuracy of the autonomous driving vehicle at the initial stage of startup. In addition, by storing the zero-bias data and compensating it at the next startup, the robustness and continuity of the system are further enhanced, providing a solid foundation for the wide application of autonomous driving technology.

[0098] As an optional solution, the device further includes a conversion unit and an averaging processing unit; the conversion unit is configured to, after obtaining the zero-bias compensation data by performing zero-bias compensation processing on the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, take the coordinate center point of the first inertial measurement unit as the origin and convert the zero-bias compensation data of the second inertial measurement unit into the coordinate system of the first inertial measurement unit to obtain the average value of the zero-bias compensation data of the first inertial measurement unit and the second inertial measurement unit, thereby obtaining fused zero-bias compensation data; the averaging processing unit is configured to perform averaging processing on the fused zero-bias compensation data with a sliding window of a preset length to obtain target zero-bias compensation data.

[0099] An alternative solution is that the device further includes a calculation unit, which is configured to calculate the loss function gradient of the initial heading angle prediction model during the process of training the initial heading angle prediction model using the above zero-bias compensation data and heading angle data, update the model parameters of the initial heading angle prediction model through an optimizer according to the loss function gradient, and determine whether the initial heading angle prediction model is trained completed by using an evaluation index, where the evaluation index includes the mean absolute error.

[0100] An alternative solution is that the device further includes a second acquisition unit and a determination unit; the second acquisition unit is configured to acquire the vehicle's data information before detecting that the above-mentioned autonomous vehicle is in a stationary scenario, where the data information includes at least one of sensor data, vehicle speed data, and GPS data; the determination unit is configured to determine whether the vehicle is in the above-mentioned stationary scenario according to the data information.

[0101] An alternative solution is that the device further includes an output unit, which is configured to output the heading angle data of the above-mentioned autonomous vehicle by using multi-sensor fusion positioning before detecting that the above-mentioned autonomous vehicle is in a stationary scenario and when it is determined that the above-mentioned autonomous vehicle is in a non-stationary scenario.

[0102] An alternative solution is that the device further includes a storage unit, which is configured to store zero-bias data in the vehicle's memory before detecting that the vehicle shuts down and powers off after generating a heading angle prediction value by inputting the above-mentioned real-time measurement data into the heading angle prediction model, where the zero-bias data is obtained by multi-sensor fusion positioning and is used to perform zero-bias compensation processing on the real-time measurement data after the vehicle is restarted next time. The training unit includes a determination module, which is configured to determine the gyroscope vibration data of the first inertial measurement unit and the second inertial measurement unit according to the historical measurement data, and train the initial heading angle prediction model by using the gyroscope vibration data, the zero-bias compensation data, and the heading angle data.

[0103] An alternative solution is that the above-mentioned initial heading angle prediction model includes a convolutional neural network and a BiLSTM network, and the heading angle prediction model is sequentially composed of an input layer, a one-dimensional convolutional layer, a max-pooling layer, a bidirectional long short-term memory layer, a fully connected layer, and an output layer.

[0104] The above-mentioned prediction device for the vehicle heading angle includes a processor and a memory. The first acquisition unit, the training unit, the input unit, etc. are all stored in the memory as program units, and the corresponding functions are implemented by the processor executing the above-mentioned program units stored in the memory. The above-mentioned modules are all located in the same processor; or, the above-mentioned each module is located in different processors in any combination form.

[0105] The processor contains a kernel, which retrieves the corresponding program unit from the memory. One or more kernels can be set, and by adjusting the kernel parameters, the problem of low accuracy in the heading angle obtained at low cost in the prior art in the vehicle stationary scenario can be solved.

[0106] The memory may include non - permanent memory in a computer - readable medium, forms such as random access memory (RAM) and / or non - volatile memory, such as read - only memory (ROM) or flash memory (flash RAM), and the memory includes at least one memory chip.

[0107] An embodiment of the present invention provides an autonomous vehicle, including: a domain controller for executing any one of the above - mentioned vehicle heading angle prediction methods.

[0108] An embodiment of the present invention provides a computer - readable storage medium. The above - mentioned computer - readable storage medium includes a stored program, wherein when the above - mentioned program runs, it controls the device where the above - mentioned computer - readable storage medium is located to execute the above - mentioned vehicle heading angle prediction method.

[0109] Specifically, the vehicle heading angle prediction method includes:

[0110] Step S201: Obtain the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, perform zero - bias compensation processing on the above - mentioned historical measurement data to obtain zero - bias compensation data, and obtain the heading angle data corresponding to the above - mentioned zero - bias compensation data, wherein the installation directions of the horizontal axes of the above - mentioned first inertial measurement unit and the above - mentioned second inertial measurement unit are opposite, and the installation directions of the vertical axes of the above - mentioned first inertial measurement unit and the above - mentioned second inertial measurement unit are the same;

[0111] Step S202: Construct an initial heading angle prediction model, and train the above - mentioned initial heading angle prediction model at least using the above - mentioned zero - bias compensation data and heading angle data to obtain a heading angle prediction model;

[0112] Step S203: Obtain the real - time measurement data of the above - mentioned first inertial measurement unit and the above - mentioned second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, input the above - mentioned real - time measurement data into the above - mentioned heading angle prediction model to generate a heading angle prediction value.

[0113] An embodiment of the present invention provides a processor, which is used to run a program, wherein when the above - mentioned program runs, it executes the above - mentioned vehicle heading angle prediction method.

[0114] Specifically, the vehicle heading angle prediction method includes:

[0115] Step S201: Obtain the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, perform zero-offset compensation processing on the above historical measurement data to obtain zero-offset compensation data, and obtain the heading angle data corresponding to the above zero-offset compensation data, wherein the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same;

[0116] Step S202: Construct an initial heading angle prediction model, and train the initial heading angle prediction model at least using the above zero-offset compensation data and heading angle data to obtain a heading angle prediction model;

[0117] Step S203: Obtain the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, input the above real-time measurement data into the heading angle prediction model to generate a heading angle prediction value.

[0118] An embodiment of the present invention provides an electronic device, which includes a processor, a memory, and a program stored on the memory and executable on the processor. When the processor executes the program, it implements at least the following steps:

[0119] Step S201: Obtain the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, perform zero-offset compensation processing on the above historical measurement data to obtain zero-offset compensation data, and obtain the heading angle data corresponding to the above zero-offset compensation data, wherein the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same;

[0120] Step S202: Construct an initial heading angle prediction model, and train the initial heading angle prediction model at least using the above zero-offset compensation data and heading angle data to obtain a heading angle prediction model;

[0121] Step S203: Obtain the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, input the above real-time measurement data into the heading angle prediction model to generate a heading angle prediction value.

[0122] The device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0123] The present application also provides a computer program product, which, when executed on a data processing device, is adapted to execute a program initialized with at least the following method steps:

[0124] Step S201: Obtain the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, perform zero-offset compensation processing on the above historical measurement data to obtain zero-offset compensation data, and obtain the heading angle data corresponding to the above zero-offset compensation data, wherein the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same;

[0125] Step S202: Construct an initial heading angle prediction model, and train the initial heading angle prediction model at least using the above zero-offset compensation data and heading angle data to obtain a heading angle prediction model;

[0126] Step S203: Obtain the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, input the above real-time measurement data into the heading angle prediction model to generate a heading angle prediction value.

[0127] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. They can be implemented by program codes executable by the computing device. Thus, they can be stored in a storage device and executed by the computing device. And in some cases, the steps shown or described herein can be executed in a different order, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. In this way, the present invention is not limited to any specific combination of hardware and software.

[0128] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0129] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, as well as the combination of flows and / or blocks in the flowchart and / or block diagram. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0130] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process Figure 1 one process or multiple processes and / or blocks Figure 1 or means for implementing the functions specified in multiple blocks.

[0132] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0133] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0134] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0135] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0136] From the above description, it can be seen that the above embodiments of the present application achieve the following technical effects:

[0137] 1) A method for predicting the vehicle heading angle of the present application includes: obtaining the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, performing zero-bias compensation processing on the above historical measurement data to obtain zero-bias compensation data, and obtaining the heading angle data corresponding to the above zero-bias compensation data, wherein the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same; constructing an initial heading angle prediction model, and training the initial heading angle prediction model with the above zero-bias compensation data and heading angle data to obtain a heading angle prediction model; obtaining the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, inputting the above real-time measurement data into the heading angle prediction model to generate a heading angle prediction value. By combining the data of two inertial measurement units with opposite directions, the accuracy and reliability of heading angle prediction are effectively improved. Especially in the vehicle stationary scenario, the heading angle prediction value generated by using the prediction model can overcome the influence of sensor zero bias and improve the positioning accuracy of the autonomous vehicle at the initial stage of startup.

[0138] 2) A device for predicting the vehicle heading angle of the present application includes: a first acquisition unit, configured to obtain the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, perform zero-bias compensation processing on the above historical measurement data to obtain zero-bias compensation data, and obtain the heading angle data corresponding to the above zero-bias compensation data, wherein the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same; a training unit, configured to construct an initial heading angle prediction model, and train the initial heading angle prediction model with the above zero-bias compensation data and heading angle data to obtain a heading angle prediction model; an input unit, configured to obtain the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, input the above real-time measurement data into the heading angle prediction model to generate a heading angle prediction value. By combining the data of two inertial measurement units with opposite directions, the accuracy and reliability of heading angle prediction are effectively improved. Especially in the vehicle stationary scenario, the heading angle prediction value generated by using the prediction model can overcome the influence of sensor zero bias and improve the positioning accuracy of the autonomous vehicle at the initial stage of startup.

[0139] The above are only the preferred embodiments of the present application and are not used to limit the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for predicting the heading angle of a vehicle, characterized in that, Including: Obtain the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, perform zero-offset compensation processing on the historical measurement data to obtain zero-offset compensation data, and obtain the heading angle data corresponding to the zero-offset compensation data, wherein the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same; Construct an initial heading angle prediction model, and train the initial heading angle prediction model at least using the zero-offset compensation data and the heading angle data to obtain a heading angle prediction model; Obtain the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, input the real-time measurement data into the heading angle prediction model to generate a heading angle prediction value.

2. The method according to claim 1, characterized in that, The historical measurement data includes gyroscope data, accelerometer data, and temperature data. After obtaining the first inertial measurement unit and performing zero-offset compensation processing on the historical measurement data to obtain zero-offset compensation data, the method further includes: Taking the coordinate center point of the first inertial measurement unit as the origin, convert the zero-offset compensation data of the second inertial measurement unit into the coordinate system of the first inertial measurement unit to obtain the average value of the zero-offset compensation data of the first inertial measurement unit and the second inertial measurement unit, and obtain fused zero-offset compensation data; Perform averaging processing on the fused zero-offset compensation data with a sliding window of a preset length to obtain target zero-offset compensation data.

3. The method according to claim 1, characterized in that During the process of training the initial heading angle prediction model using the zero-offset compensation data and the heading angle data, the method further includes: Calculate the loss function gradient of the initial heading angle prediction model, update the model parameters of the initial heading angle prediction model according to the loss function gradient through an optimizer, and use an evaluation index to determine whether the initial heading angle prediction model is trained, wherein the evaluation index includes the mean absolute error.

4. The method according to claim 1, characterized in that Before it is detected that the vehicle is in a stationary scenario, the method further includes: Obtain the data information of the vehicle, wherein the data information includes at least one of sensor data, vehicle speed data, and GPS data; Determine whether the vehicle is in the stationary scenario according to the data information.

5. The method according to claim 1, wherein Before it is detected that the vehicle is in a stationary scenario, the method further includes: When it is determined that the vehicle is in a non-stationary scenario, output the heading angle data of the vehicle using multi-sensor fusion positioning.

6. The method according to claim 1, wherein After inputting the real-time measurement data into the heading angle prediction model to generate a heading angle prediction value, the method further includes: before it is detected that the vehicle shuts down and powers off, store the zero-offset data in the memory of the vehicle, wherein the zero-offset data is obtained by multi-sensor fusion positioning, and the zero-offset data is used to perform zero-offset compensation processing on the real-time measurement data after the vehicle is restarted next time; At least use the zero-bias compensation data and the heading angle data to train the initial heading angle prediction model, including: determining the gyroscope vibration data of the first inertial measurement unit and the second inertial measurement unit according to the historical measurement data, and using the gyroscope vibration data, the zero-bias compensation data and the heading angle data to train the initial heading angle prediction model.

7. The method according to claim 1, characterized in that, The initial heading angle prediction model includes a convolutional neural network and a BiLSTM network, and the heading angle prediction model is sequentially composed of an input layer, a one-dimensional convolutional layer, a max pooling layer, a bidirectional long short-term memory layer, an attention layer, a fully connected layer and an output layer.

8. A prediction device for a vehicle heading angle, characterized in that, Including: A first acquisition unit, configured to acquire the historical measurement data of the first inertial measurement unit and the second inertial measurement unit, perform zero-bias compensation processing on the historical measurement data to obtain zero-bias compensation data, and acquire the heading angle data corresponding to the zero-bias compensation data, where the installation directions of the horizontal axes of the first inertial measurement unit and the second inertial measurement unit are opposite, and the installation directions of the vertical axes of the first inertial measurement unit and the second inertial measurement unit are the same; A training unit, configured to construct an initial heading angle prediction model, and at least use the zero-bias compensation data and the heading angle data to train the initial heading angle prediction model to obtain a heading angle prediction model; An input unit, configured to acquire the real-time measurement data of the first inertial measurement unit and the second inertial measurement unit, and when it is detected that the vehicle is in a stationary scenario, input the real-time measurement data into the heading angle prediction model to generate a heading angle prediction value.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein when the program runs, it controls the device where the computer-readable storage medium is located to execute the vehicle heading angle prediction method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, Including: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and the one or more programs include those for executing the vehicle heading angle prediction method according to any one of claims 1 to 7.

11. An autonomous vehicle, characterized in that, Including: A domain controller, configured to execute the vehicle heading angle prediction method according to any one of claims 1 to 7.

Images