CN113252060A

CN113252060A - Vehicle track calculation method and device based on neural network model

Info

Publication number: CN113252060A
Application number: CN202110597498.4A
Authority: CN
Inventors: 费再慧; 贾双成; 朱磊; 李成军; 李倩
Original assignee: Zhidao Network Technology Beijing Co Ltd
Current assignee: Zhidao Network Technology Beijing Co Ltd
Priority date: 2021-05-31
Filing date: 2021-05-31
Publication date: 2021-08-13
Anticipated expiration: 2041-05-31
Also published as: CN113252060B

Abstract

The application relates to a vehicle track estimation method and device based on a neural network model. The method comprises the following steps: dividing neurons of a neural network model to be trained into a plurality of layers of neurons according to a set rule; outputting an IMU estimation data sequence by using a plurality of layers of neurons in the neural network model to be trained layer by layer in sequence according to a measurement data sequence of an inertia measurement unit, calculating to obtain a track point calculation position data sequence of the vehicle according to the IMU estimation data sequence, calculating the position data sequence according to the track point of the vehicle and a position data sequence of a positioning module, and obtaining parameters of the plurality of layers of neurons in sequence layer by layer to finish the training of the neural network model; and calculating to obtain a track point calculated position sequence of the vehicle according to the IMU estimation data sequence output by the trained neural network model, thereby obtaining the motion trail of the vehicle. According to the scheme provided by the application, the neural network model can be used for reducing the accumulated error when the vehicle track is calculated by using the measurement data of the inertia measurement unit.

Description

Vehicle track calculation method and device based on neural network model

Technical Field

The application relates to the technical field of navigation, in particular to a vehicle track calculation method and device based on a neural network model.

Background

In the related art, vehicle navigation often depends on a satellite Positioning module such as a GPS (Global Positioning System) satellite Positioning module. However, in some situations, such as under bridges, culverts, tunnels, dense buildings, etc., the positioning deviation of the satellite positioning module of the related art is very large, and even the positioning result cannot be provided. The Inertial navigation system including an Inertial Measurement Unit (IMU) can calculate accurate speed, attitude, and position information of the vehicle by using Measurement data of the Inertial Measurement Unit.

The inertial navigation system estimates speed, attitude and position information of the vehicle using measurement data of an accelerometer and a gyroscope of the inertial measurement unit. However, when the inertial navigation system estimates the vehicle trajectory by using the inertial measurement unit, the accuracy is good only near the initial time, and the accuracy of estimating the vehicle trajectory by using the measurement data of the inertial measurement unit is reduced due to the influence of accumulated errors as time increases. Therefore, how to reduce the accumulated error when the vehicle trajectory is estimated by using the measurement data of the inertial measurement unit is an urgent problem to be solved.

Disclosure of Invention

In order to solve or partially solve the problems in the related art, the present application provides a vehicle trajectory estimation method and device based on a neural network model, which can reduce the accumulated error when estimating the vehicle trajectory by using the measurement data of an inertial measurement unit by using the neural network model, and improve the accuracy of estimating the vehicle trajectory by using the measurement data of the inertial measurement unit.

The application provides a vehicle track estimation method based on a neural network model in a first aspect, and the method comprises the following steps:

inputting a measurement data sequence of an inertia measurement unit and a position data sequence of a positioning module in the same time period to a neural network model to be trained;

dividing neurons of the full connection layer of the neural network model to be trained into a plurality of layers of neurons according to a set rule;

outputting an IMU estimation data sequence by the neural network model to be trained by utilizing the multiple layers of neurons in sequence layer by layer according to the measurement data sequence, calculating to obtain a track point calculation position data sequence of the vehicle according to the IMU estimation data sequence, and obtaining parameters of the multiple layers of neurons in sequence layer by layer according to the track point calculation position data sequence of the vehicle and the position data sequence of the positioning module, so as to finish the training of the neural network model to be trained;

inputting a measurement data sequence of the inertial measurement unit to the trained neural network model;

and enabling the trained neural network model to output an IMU estimation data sequence according to the measurement data sequence, and calculating according to the IMU estimation data sequence to obtain a track point calculation position sequence of the vehicle, thereby obtaining the motion track of the vehicle.

Preferably, the dividing neurons of the fully-connected layer of the neural network model to be trained into multiple layers of neurons according to a set rule includes:

and sequentially dividing the neurons of the full connection layer of the neural network model to be trained into n layers of neurons, wherein the number of the i-th layer of neurons is the power i of 2, the i + 1-th layer of neurons comprises the i-th layer of neurons, n is a positive integer, and i =1,2,3, …, n.

Preferably, the dividing the neurons of the fully-connected layer of the neural network model to be trained into n layers of neurons sequentially, the number of the i layer of neurons being 2 to the power i, the i +1 layer of neurons including the i layer of neurons, where n is a positive integer, i =1,2,3, …, n includes:

setting an acceleration data model and an angular velocity data model in the neural network model to be trained;

sequentially dividing neurons of a full connection layer of the acceleration data model into n layers of neurons, sequentially dividing neurons of a full connection layer of the angular velocity data model into n layers of neurons, wherein the number of the i-th layer of neurons is the power i of 2, and the i + 1-th layer of neurons comprises the i-th layer of neurons, wherein n is a positive integer, i =1,2,3, …, n.

Preferably, the step of enabling the neural network model to be trained to sequentially and layer-by-layer utilize the multiple layers of neurons to output an IMU estimation data sequence according to the measurement data sequence, calculate and obtain a trajectory point calculation position data sequence of a vehicle according to the IMU estimation data sequence, sequentially and layer-by-layer obtain parameters of the multiple layers of neurons according to the trajectory point calculation position data sequence of the vehicle and the position data sequence of the positioning module, thereby completing the training of the neural network model to be trained includes:

enabling the acceleration data model to sequentially utilize the i-th layer neuron of the full connection layer to output an acceleration estimation data sequence according to the acceleration sequence of the measurement data sequence, and calculating to obtain the track point calculated position of the vehicle at each moment of the same time period according to the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, so as to obtain the track point calculated position data sequence of the vehicle;

calculating and obtaining the ith group of errors of the track point calculation position data sequence of the vehicle and the position data sequence of the positioning module;

obtaining parameters of the i-th layer neuron of the full connection layer of the acceleration data model by enabling the i-th group of error convergence to be smaller than a preset threshold value;

outputting an angular velocity estimation data sequence by the angular velocity data model according to the angular velocity sequence of the measurement data sequence by sequentially utilizing the i-th layer neuron of the full connection layer, and calculating to obtain the track point calculated position of the vehicle at each moment of the same time period according to the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, thereby obtaining the track point calculated position data sequence of the vehicle;

and acquiring parameters of the i-th layer neuron of the full connection layer of the angular velocity data model by enabling the i-th group of error convergence to be smaller than a preset threshold value.

Preferably, the method further comprises:

setting the parameter of the i-th layer neuron of the full connection layer of the acceleration data model as the initial parameter of the i + 1-th layer neuron;

enabling the acceleration data model to utilize neurons of an i +1 th layer of a full connection layer to output an acceleration estimation data sequence according to the acceleration sequence of the measurement data sequence, and calculating and obtaining a track point calculated position of the vehicle at each moment of the same time period according to the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, so as to obtain a track point calculated position data sequence of the vehicle;

calculating and obtaining the i +1 group error of the track point calculation position data sequence of the vehicle and the position data sequence of the positioning module;

obtaining parameters of neurons of the (i + 1) th layer of the full connection layer of the acceleration data model by enabling the (i + 1) th group of error convergence to be smaller than a preset threshold value;

setting the parameter of the i layer neuron of the full connection layer of the angular velocity data model as the initial parameter of the i +1 layer neuron;

enabling the angular velocity data model to utilize neurons of an i +1 th layer of a full connection layer to output an angular velocity estimation data sequence according to the angular velocity sequence of the measurement data sequence, and calculating to obtain a track point calculated position of the vehicle at each moment of the same time period according to the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, so as to obtain a track point calculated position data sequence of the vehicle;

and acquiring parameters of neurons of the (i + 1) th layer of the full-connection layer of the angular velocity data model by enabling the i +1 th group of error convergence to be smaller than a preset threshold value.

The second aspect of the present application provides a vehicle trajectory estimation device based on a neural network model, the device including:

6. a vehicle trajectory estimation device based on a neural network model is characterized by comprising:

the first input module is used for inputting a measurement data sequence of the inertia measurement unit and a position data sequence of the positioning module in the same time period to a neural network model to be trained;

the layering module is used for dividing the neurons of the full connection layer of the neural network model to be trained into a plurality of layers of neurons according to a set rule;

the training module is used for enabling the neural network model to be trained to sequentially utilize the multiple layers of neurons to output an IMU estimation data sequence according to the measurement data sequence input by the first input module layer by layer, obtaining a track point calculation position data sequence of a vehicle according to the IMU estimation data sequence, and sequentially obtaining the parameters of the multiple layers of neurons layer by layer according to the track point calculation position data sequence of the vehicle and the position data sequence of the positioning module input by the first input module, so that the training of the neural network model to be trained is completed;

the second input module is used for inputting the measurement data sequence of the inertial measurement unit to the trained neural network model;

and the track calculation module is used for enabling the trained neural network model to output an IMU estimation data sequence according to the measurement data sequence input by the second input module, and calculating and obtaining a track point calculation position sequence of the vehicle according to the IMU estimation data sequence, so that the motion track of the vehicle is obtained.

Preferably, the apparatus further comprises:

the model setting module is used for setting an acceleration data model and an angular velocity data model on the neural network model to be trained;

the layering module is specifically configured to sequentially divide neurons of a full connection layer of the acceleration data model into n layers of neurons, sequentially divide neurons of a full connection layer of the angular velocity data model into n layers of neurons, where the number of i-th layer neurons is i-th power of 2, and the i + 1-th layer neurons include i-th layer neurons, where n is a positive integer, i =1,2,3, …, n.

Preferably, the training module comprises:

the first track submodule is used for enabling the acceleration data model to sequentially utilize the i-th layer neuron of the full connection layer to output an acceleration estimation data sequence according to the acceleration sequence of the measurement data sequence input by the first input module, and calculating and obtaining a track point calculated position of the vehicle at each moment of the same time period according to the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, so that a track point calculated position data sequence of the vehicle is obtained;

the first error calculation submodule is used for calculating and obtaining the ith group of errors of the track point calculated position data sequence of the vehicle obtained by the first track submodule and the position data sequence of the positioning module input by the first input module;

a first parameter obtaining submodule, configured to obtain parameters of an ith layer neuron of a full connection layer of the acceleration data model by making the i-th group of error convergence calculated and obtained by the first error calculation submodule smaller than a preset threshold;

the second track submodule is used for enabling the angular velocity data model to sequentially utilize the i-th layer neuron of the full-connection layer to output an angular velocity estimation data sequence according to the angular velocity sequence of the measurement data sequence input by the first input module, and calculating and obtaining the track point calculated position of the vehicle at each moment of the same time period according to the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, so that the track point calculated position data sequence of the vehicle is obtained;

the second error calculation submodule is used for calculating and obtaining the ith group of errors between the track point calculated position data sequence of the vehicle obtained by the second track submodule and the position data sequence of the positioning module input by the first input module;

and the second parameter acquisition submodule is used for acquiring parameters of the i-th layer neuron of the full connection layer of the angular velocity data model by enabling the i-th group of error convergence calculated and acquired by the second error calculation submodule to be smaller than a preset threshold value.

Preferably, the training module further comprises:

the first parameter setting submodule is used for setting the parameter of the i-th layer neuron of the full connection layer of the acceleration data model, which is obtained by the first parameter obtaining submodule, as the initial parameter of the i + 1-th layer neuron;

the first track submodule is further configured to enable the acceleration data model to output an acceleration estimation data sequence according to the acceleration sequence of the measurement data sequence input by the first input module by using an i +1 th layer neuron of a full connection layer, and calculate and obtain a track point calculated position of the vehicle at each time of the same time period according to the acceleration estimation data sequence and an angular velocity sequence of the measurement data sequence, so as to obtain a track point calculated position data sequence of the vehicle;

the first error calculation submodule is further configured to calculate and obtain an i +1 th group of errors between the track point calculated position data sequence of the vehicle obtained by the first track submodule and the position data sequence of the positioning module input by the first input module;

the first parameter obtaining submodule is further configured to obtain parameters of neurons in an i +1 th layer of the full-link layer of the acceleration data model by making the i +1 th group of error convergence calculated and obtained by the first error calculating submodule smaller than a preset threshold;

a second parameter setting submodule, configured to set the parameter of the i-th layer neuron of the full connection layer of the angular velocity data model, obtained by the second parameter obtaining submodule, as an initial parameter of the i + 1-th layer neuron;

the second trajectory submodule is further configured to enable the angular velocity data model to output an angular velocity estimation data sequence according to the angular velocity sequence of the measurement data sequence input by the first input module by using the i +1 th layer neuron of the full connection layer, and calculate and obtain a trajectory point calculated position of the vehicle at each time of the same time period according to the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, so as to obtain a trajectory point calculated position data sequence of the vehicle;

the second error calculation submodule is also used for calculating and obtaining the i +1 group of errors between the track point calculated position data sequence of the vehicle obtained by the second track submodule and the position data sequence of the positioning module input by the first input module;

the second parameter obtaining submodule is further configured to obtain parameters of neurons in an i +1 th layer of the full-link layer of the angular velocity data model by making the i +1 th group of error convergence calculated and obtained by the second error calculating submodule smaller than a preset threshold.

A third aspect of the present application provides an electronic device comprising:

a processor; and

a memory having executable code stored thereon, which when executed by the processor, causes the processor to perform the method as described above.

A fourth aspect of the present application provides a non-transitory machine-readable storage medium having stored thereon executable code, which when executed by a processor of an electronic device, causes the processor to perform a method as described above.

The technical scheme provided by the application can comprise the following beneficial effects:

according to the technical scheme, neurons of a full connection layer of a neural network model to be trained are divided into multiple layers of neurons according to a set rule, a measurement data sequence of an inertial measurement unit and a position data sequence of a positioning module are taken as samples, parameters of the multiple layers of neurons of the full connection layer of the neural network model to be trained are sequentially obtained layer by layer, training of each layer of neurons in the multiple layers of neurons of the neural network model to be trained is completed layer by layer and in turn, the training difficulty of the neural network model is reduced, the parameters of the neurons of the full connection layer of the neural network model to be trained can be rapidly obtained, and the training efficiency of the neural network model is improved; meanwhile, the trained neural network model is used for outputting an IMU estimation data sequence according to the measurement data sequence of the inertia measurement unit, the motion track of the vehicle is obtained according to the IMU estimation data sequence, the accumulated error caused by the fact that the vehicle track is calculated by the measurement data sequence of the inertia measurement unit can be reduced by the neural network model, and the accuracy of calculating the vehicle track by the measurement data sequence of the inertia measurement unit is improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.

Drawings

The foregoing and other objects, features and advantages of the application will be apparent from the following more particular descriptions of exemplary embodiments of the application, as illustrated in the accompanying drawings wherein like reference numbers generally represent like parts throughout the exemplary embodiments of the application.

FIG. 1 is a schematic flow chart of a vehicle trajectory estimation method based on a neural network model according to an embodiment of the present application;

FIG. 2 is a schematic flow chart diagram illustrating a neural network model-based vehicle trajectory estimation method according to another embodiment of the present application;

FIG. 3 is a schematic flow chart diagram illustrating a neural network model-based vehicle trajectory estimation method according to another embodiment of the present application;

FIG. 4 is a schematic structural diagram of a vehicle trajectory estimation device based on a neural network model according to an embodiment of the present application;

FIG. 5 is a schematic structural diagram of a vehicle trajectory estimation device based on a neural network model according to an embodiment of the present application;

fig. 6 is a schematic structural diagram of an electronic device shown in an embodiment of the present application.

Detailed Description

Embodiments of the present application will be described in more detail below with reference to the accompanying drawings. While embodiments of the present application are illustrated in the accompanying drawings, it should be understood that the present application may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this application and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

It should be understood that although the terms "first," "second," "third," etc. may be used herein to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present application. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the present application, "a plurality" means two or more unless specifically limited otherwise.

The embodiment of the application provides a vehicle track estimation method based on a neural network model, which can reduce the accumulated error when the vehicle track is estimated by using the measurement data of an inertia measurement unit by using the neural network model, and improve the accuracy of vehicle track estimation by using the measurement data of the inertia measurement unit.

The technical solutions of the embodiments of the present application are described in detail below with reference to the accompanying drawings.

The first embodiment is as follows:

fig. 1 is a schematic flowchart of a vehicle trajectory estimation method based on a neural network model according to an embodiment of the present application.

Referring to fig. 1, a vehicle trajectory estimation method based on a neural network model includes:

in step 101, a measurement data sequence of an inertial measurement unit and a position data sequence of a positioning module in the same time period are input into a neural network model to be trained.

In one specific implementation mode, the vehicle is provided with an inertia measurement unit, a positioning module and an on-board intelligent device loaded with a neural network model. The inertial measurement unit comprises an accelerometer and a gyroscope, and the measurement data sequence of the inertial measurement unit comprises an acceleration data sequence of the accelerometer of the inertial measurement unit and an angular velocity data sequence of the gyroscope. The vehicle acceleration data sequence obtained by an accelerometer of the inertial measurement unit and the vehicle angular velocity data sequence obtained by a gyroscope of the inertial measurement unit. The positioning module may include, but is not limited to, at least one of a GPS module, a beidou satellite positioning module, an RTK (Real Time Kinematic) positioning module, and the like. With the positioning module, positioning information of the vehicle may be obtained, which may include, but is not limited to, position information, speed information, and attitude information. The location information includes, but is not limited to, latitude and longitude coordinate information describing the location, and the attitude information includes, but is not limited to, heading angle information describing the heading. The position data sequence of the vehicle can be obtained according to the positioning information of the positioning module.

In one embodiment, a position data sequence for a time period of a vehicle is obtained where a vehicle RTK positioning module signal is available. When the position data sequence of the vehicle in one time period is acquired through the RTK positioning module, the measurement data sequence of the vehicle in the same time period is acquired through the inertial measurement unit. It will be appreciated that the position data sequence acquired by the RTK positioning module and the measurement data sequence acquired by the inertial measurement unit are both aligned in time to avoid inaccuracies in the results due to the difference in the reference times.

In one specific embodiment, the position data sequence and the measurement data sequence of the vehicle are input to the neural network model to be trained for the same time period. And the neural network model to be trained takes the position data sequence and the measurement data sequence of the vehicle in the same time period as samples to finish the training of the neural network model to be trained.

In step 102, dividing the neurons of the full connection layer of the neural network model to be trained into multiple layers of neurons according to a set rule.

In one specific embodiment, the neurons of the full connection layer of the neural network model to be trained are divided into a plurality of layers of neurons according to a set rule, and each layer comprises a set number of neurons.

In step 103, the neural network model to be trained sequentially uses multiple layers of neurons to output an IMU estimation data sequence according to the measurement data sequence layer by layer, calculates to obtain a trajectory point calculation position data sequence of the vehicle according to the IMU estimation data sequence, calculates the position data sequence according to the trajectory point of the vehicle, and obtains parameters of multiple layers of neurons sequentially layer by layer according to the position data sequence of the positioning module, thereby completing the training of the neural network model to be trained.

In one embodiment, the neural network model to be trained sequentially utilizes each layer of neurons of the multiple layers of neurons of the full connection layer, and outputs an IMU estimation data sequence according to the measurement data sequence of the inertial measurement unit; estimating a data sequence by the neural network model according to the IMU, and calculating to obtain a track point calculation position data sequence of the vehicle; calculating errors of the track point calculated position data sequence of the vehicle and the position data sequence of the positioning module according to the track point calculated position data sequence of the vehicle and the position data sequence of the positioning module; and sequentially acquiring the parameters of each layer of neurons in the multiple layers of neurons of the full connection layer of the neural network model layer by making the error convergence smaller than a preset threshold value, thereby finishing the training of the neural network model to be trained.

In one specific embodiment, a track point estimated position data sequence of the vehicle is estimated according to an IMU estimated data sequence output by a neural network model to be trained; calculating errors of the track point calculation position data sequence of the vehicle and the position data sequence of the positioning module; if the error convergence is smaller than the preset threshold value, the neural network model to be trained can output an IMU estimation data sequence for reducing the accumulated error, and the training of the neural network model to be trained is completed.

In step 104, the measurement data sequence of the inertial measurement unit is input to the trained neural network model.

In one embodiment, in the event that a vehicle RTK positioning module signal is unavailable, a measurement data sequence of an inertial measurement unit is input to the trained neural network model, the measurement data sequence including measurement data at each time instant.

In step 105, the trained neural network model outputs an IMU estimation data sequence according to the measurement data sequence, and calculates a trajectory point calculated position sequence of the vehicle according to the IMU estimation data sequence, thereby obtaining a motion trajectory of the vehicle.

In one embodiment, the trained neural network model is enabled to output IMU estimation data at each moment according to the input measurement data at each moment; according to positioning information before the positioning module signal is unavailable, obtaining the initial pose of the vehicle, and the position information, the course angle information and the speed information of the vehicle at the initial moment of the initial pose of the vehicle; and calculating the track point calculated position of each moment of the vehicle by the trained neural network model according to the initial pose of the vehicle and the IMU estimation data of each moment, and obtaining a track point calculated position sequence of the vehicle, thereby obtaining the motion track of the vehicle.

According to the vehicle trajectory calculation method based on the neural network model, neurons of a full connection layer of the neural network model to be trained are divided into multiple layers of neurons according to a set rule, a measurement data sequence of an inertia measurement unit and a position data sequence of a positioning module are used as samples, parameters of the multiple layers of neurons of the full connection layer of the neural network model to be trained are sequentially obtained layer by layer, training of each layer of neurons in the multiple layers of neurons of the neural network model to be trained is completed sequentially layer by layer, training difficulty of the neural network model is reduced, the parameters of the neurons of the full connection layer of the neural network model to be trained can be rapidly obtained, and training efficiency of the neural network model is improved; meanwhile, the trained neural network model is used for outputting an IMU estimation data sequence according to the measurement data sequence of the inertia measurement unit, the motion track of the vehicle is obtained according to the IMU estimation data sequence, the accumulated error caused by the fact that the vehicle track is calculated by the measurement data sequence of the inertia measurement unit can be reduced by the neural network model, and the accuracy of calculating the vehicle track by the measurement data sequence of the inertia measurement unit is improved.

Example two:

fig. 2 is a schematic flowchart of a vehicle trajectory estimation method based on a neural network model according to another embodiment of the present application.

Referring to fig. 2, a vehicle trajectory estimation method based on a neural network model includes:

in step 201, a measurement data sequence of the inertial measurement unit and a position data sequence of the positioning module in the same time period are input into the neural network model to be trained.

This step can be referred to the description of step 101, and is not described here.

In step 202, the neurons of the fully-connected layer of the neural network model to be trained are sequentially divided into n layers of neurons, the number of the i-th layer of neurons is the power i of 2, the i + 1-th layer of neurons includes the i-th layer of neurons, where n is a positive integer, and i =1,2,3, …, n.

In one embodiment, the neurons of the fully-connected layer of the neural network model to be trained are sequentially divided into n layers of neurons, the number of the neurons in each layer is the power i of 2, wherein n is a positive integer, i =1,2, …, n, and the neurons in the latter layer include the neurons in the former layer. For example, there are 100 neurons in the fully-connected layer of the neural network model to be trained, the 100 neurons are sequentially divided into a plurality of layers of neurons, the first layer of neurons includes 2 (power of 1 of 2) neurons from 1 st to 2 nd, the second layer of neurons includes 4 (power of 2) neurons from 1 st to 4 th, the third layer of neurons includes 8 (power of 3 of 2) neurons from 1 st to 8 th, the fourth layer of neurons includes 16 (power of 4 of 2) neurons from 1 st to 16 th, the fifth layer of neurons includes 32 (power of 5 of 2) neurons from 1 st to 32 th, the sixth layer of neurons includes 64 (power of 6 of 2) neurons from 1 st to 64 th, and the seventh layer of neurons includes 100 (power of 7 of 2 =128> 100) neurons from 1 st to 100 th.

In another embodiment, the neurons of the full-connectivity layer of the neural network model to be trained may be sequentially divided into n layers of neurons, the number of the i-th layer of neurons may be i-th power of 3, and the i + 1-th layer of neurons includes the i-th layer of neurons, where n is a positive integer, i =1,2,3, …, n.

In step 203, the neural network model to be trained sequentially uses the i-th layer neuron of the full connection layer to output an IMU estimation data sequence according to the measurement data sequence, and calculates and obtains the trajectory point calculated position of the vehicle at each moment in the same time period according to the IMU estimation data sequence, thereby obtaining the trajectory point calculated position data sequence of the vehicle.

In a specific embodiment, taking the first layer of neurons and the second layer of neurons of the neural network model to be trained as an example, the first layer of neurons of the neural network model to be trained is trained to obtain the parameters of the first layer of neurons. The neural network model to be trained can utilize the first layer neuron of the full connection layer to predict and output an IMU estimation data sequence according to the measurement data sequence of the inertia measurement unit in a time period; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module in the same time period; and the neural network model to be trained can calculate the track point calculated position of the vehicle at each moment in a time period by utilizing the first layer of neurons of the full connection layer according to the initial pose of the vehicle and the IMU estimated data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

In step 204, an ith group of errors of the track point estimated position data sequence of the vehicle and the position data sequence of the positioning module are calculated and obtained.

In a specific embodiment, according to a track point estimated position data sequence of a vehicle obtained by a neuron of a first layer of a full connection layer of a neural network model, a difference value between track point estimated position data of the vehicle and position data of a positioning module at each same time in the same time period of the track point estimated position data sequence of the vehicle and the position data of the positioning module is calculated, so that a first group of errors between the track point estimated position data sequence of the vehicle and the position data sequence of the positioning module in the same time period are calculated.

In step 205, parameters of the i-th layer neurons of the fully-connected layer of the neural network model to be trained are obtained by making the i-th group of error convergence smaller than a preset threshold.

In a specific embodiment, parameters of the first layer neurons of the full-connection layer of the neural network model to be trained can be continuously optimized by adjusting the parameters of the first layer neurons of the full-connection layer of the neural network model to be trained, so that a first group of errors between a track point estimated position data sequence of a vehicle and a position data sequence of a positioning module in the same time period are the smallest as possible, and convergence is smaller than a preset threshold value, that is, the track point estimated position data sequence of the vehicle in the same time period is converged towards the position data sequence of the positioning module, so that the parameters of the first layer neurons of the neural network model to be trained are obtained.

In step 206, the parameters of the i-th layer neurons of the fully-connected layer of the neural network model to be trained are set as the initial parameters of the i + 1-th layer neurons.

In one embodiment, the parameters of a first layer of neurons of a fully-connected layer of the neural network model to be trained may be set as initial parameters of a second layer of neurons.

In step 207, the neural network model to be trained outputs an IMU estimation data sequence according to the measurement data sequence by using the i +1 th layer neuron of the full connection layer, and estimates and obtains the trajectory point estimation position of the vehicle at each moment in the same time period according to the IMU estimation data sequence, thereby obtaining the trajectory point estimation position data sequence of the vehicle.

In one embodiment, the neural network model to be trained may be enabled to predict an output IMU estimation data sequence according to a measurement data sequence of an inertial measurement unit for a time period by using neurons of a second layer of the full connection layer; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module in the same time period; and the neural network model to be trained can calculate the track point calculated position of the vehicle at each moment in a time period by utilizing a second layer neuron of the full connection layer according to the initial pose of the vehicle and the IMU estimated data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

In step 208, the i +1 th group of errors between the track point estimated position data sequence of the vehicle and the position data sequence of the positioning module are calculated.

In a specific embodiment, according to a track point estimated position data sequence of the vehicle obtained by a neural network model of a second layer of the full connection layer by using neurons of the second layer, a difference value between track point estimated position data of the vehicle and position data of the positioning module at each same moment in the same time period is calculated, so that a second group of errors between the track point estimated position data sequence of the vehicle and the position data sequence of the positioning module in the same time period are calculated.

In step 209, the parameters of the i +1 th layer neurons of the fully-connected layer of the neural network model to be trained are obtained by making the i +1 th group of error convergence smaller than a preset threshold.

In a specific embodiment, parameters of neurons in the second layer of the fully-connected layer of the neural network model to be trained can be continuously optimized by adjusting the parameters of neurons in the second layer of the fully-connected layer of the neural network model to be trained, so that the error between the track point estimated position data sequence of the vehicle in the same time period and the second group of error between the position data sequence of the positioning module are the smallest as possible, and the convergence is smaller than a preset threshold value, that is, the track point estimated position data sequence of the vehicle in the same time period is converged to the position data sequence of the positioning module, so that the parameters of neurons in the second layer of the neural network model to be trained are obtained.

In a specific embodiment, and so on,

steps

206, 207, 208, and 209 are executed in a loop, and the parameters of each layer of neurons in the n layers of neurons in the full connection layer of the neural network model to be trained are obtained sequentially layer by layer, so as to complete the training of the neural network model to be trained.

It should be noted that the preset threshold values in step 205 and step 209 may be the same or different.

In step 210, the measurement data sequence of the inertial measurement unit is input to the trained neural network model.

This step can be referred to the description of step 104, and is not described here.

In step 211, the trained neural network model outputs an IMU estimation data sequence according to the measurement data sequence, and calculates a trajectory point calculated position sequence of the vehicle according to the IMU estimation data sequence, thereby obtaining a motion trajectory of the vehicle.

This step can be referred to the description of step 105, and is not described here.

According to the vehicle trajectory calculation method based on the neural network model, neurons of a full connection layer of the neural network model to be trained are divided into multiple layers of neurons according to a set rule, a measurement data sequence of an inertia measurement unit and a position data sequence of a positioning module are used as samples, parameters of the multiple layers of neurons of the full connection layer of the neural network model to be trained are sequentially obtained layer by layer, training of each layer of neurons in the multiple layers of neurons of the neural network model to be trained is completed sequentially layer by layer, training difficulty of the neural network model is reduced, the parameters of the neurons of the full connection layer of the neural network model to be trained can be rapidly obtained, and training efficiency of the neural network model is improved; meanwhile, the trained neural network model is used for outputting an IMU estimation data sequence according to the measurement data sequence of the inertia measurement unit, the motion track of the vehicle is obtained according to the IMU estimation data sequence, the accumulated error caused by the fact that the vehicle track is calculated by the measurement data of the inertia measurement unit can be reduced by the neural network model, and the accuracy of calculating the vehicle track by the measurement data of the inertia measurement unit is improved.

Further, in the vehicle trajectory estimation method based on the neural network model provided in the embodiment of the present application, the parameter of the i-th layer neuron of the full connection layer of the neural network model to be trained is set as the initial parameter of the i + 1-th layer neuron, so that the difficulty of training the neural network model can be reduced, the samples of training the neural network model can be reduced, the parameter of the full connection layer neuron of the neural network model to be trained can be quickly obtained, and the training of the neural network model can be further accelerated.

Example three:

fig. 3 is a schematic flowchart of a vehicle trajectory estimation method based on a neural network model according to another embodiment of the present application.

Referring to fig. 3, a vehicle trajectory estimation method based on a neural network model includes:

in step 301, a measurement data sequence of the inertial measurement unit and a position data sequence of the positioning module in the same time period are input into the neural network model to be trained.

In step 302, an acceleration data model and an angular velocity data model are set in a neural network model to be trained.

In one embodiment, an acceleration data model and an angular velocity data model are set in a neural network model to be trained based on different characteristics of an accelerometer and a gyroscope of an inertial measurement unit. By respectively training the acceleration data model and the angular velocity data model, the training complexity of the neural network model can be reduced, and the training of the neural network model to be trained can be completed more quickly and accurately.

In step 303, the neurons of the full connection layer of the acceleration data model are sequentially divided into n layers of neurons, the neurons of the full connection layer of the angular velocity data model are sequentially divided into n layers of neurons, the number of the i-th layer of neurons is i power of 2, the i + 1-th layer of neurons includes the i-th layer of neurons, where n is a positive integer, and i =1,2,3, …, n.

In one embodiment, the neurons of the full connectivity layer of the acceleration data model are sequentially divided into n layers of neurons, the number of the neurons in each layer is the power i of 2, wherein n is a positive integer, i =1,2, …, n, and the neurons in the latter layer include the neurons in the former layer. For example, the fully-connected layer of the acceleration data model has 100 neurons, the 100 neurons are sequentially divided into a plurality of layers of neurons, the first layer of neurons includes 2 (powers of 1 of 2) from 1 st to 2 nd, the second layer of neurons includes 4 (powers of 2) from 1 st to 4 th, the third layer of neurons includes 8 (powers of 3 of 2) from 1 st to 8 th, the fourth layer of neurons includes 16 (powers of 4 of 2) from 1 st to 16 th, the fifth layer of neurons includes 32 (powers of 5 of 2) from 1 st to 32 th, the sixth layer of neurons includes 64 (powers of 6 of 2) from 1 st to 64 th, and the seventh layer of neurons includes 100 (powers of 7 of 2 =128> 100) from 1 st to 100 th.

In one embodiment, the neurons of the full connectivity layer of the angular velocity data model are sequentially divided into n layers of neurons, the number of the neurons in each layer is the power i of 2, wherein n is a positive integer, i =1,2, …, n, and the neurons in the later layer include the neurons in the former layer. For example, the fully-connected layer of the angular velocity data model has 100 neurons, the 100 neurons are sequentially divided into a plurality of layers of neurons, the first layer of neurons includes 2 (powers of 1 of 2) from 1 st to 2 nd, the second layer of neurons includes 4 (powers of 2) from 1 st to 4 th, the third layer of neurons includes 8 (powers of 3 of 2) from 1 st to 8 th, the fourth layer of neurons includes 16 (powers of 4 of 2) from 1 st to 16 th, the fifth layer of neurons includes 32 (powers of 5 of 2) from 1 st to 32 th, the sixth layer of neurons includes 64 (powers of 6 of 2) from 1 st to 64 th, and the seventh layer of neurons includes 100 (powers of 7 of 2 =128> 100) from 1 st to 100 th.

In step 304, the acceleration data model sequentially outputs an acceleration estimation data sequence according to the acceleration sequence of the measurement data sequence by using the i-th layer neuron of the full connection layer, and estimates and obtains the track point estimated position of the vehicle at each time of the same time period according to the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, thereby obtaining the track point estimated position data sequence of the vehicle.

In one embodiment, taking the first layer of neurons and the second layer of neurons of the acceleration data model as an example, the first layer of neurons of the acceleration data model is trained to obtain the parameters of the first layer of neurons. The acceleration data model can utilize the first layer neuron of the full connection layer to predict and output an acceleration estimation data sequence according to the acceleration sequence of the measurement data sequence of the inertia measurement unit in a time period; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module in the same time period; and the acceleration data model can calculate the track point calculated position of the vehicle at each moment in a time period by using the first layer neuron of the full connection layer according to the initial pose of the vehicle, the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

In step 305, an ith set of errors of the track point estimated position data sequence of the vehicle and the position data sequence of the positioning module are calculated and obtained.

In a specific embodiment, according to the trajectory point estimated position data sequence of the vehicle obtained by the neuron in the first layer of the full connection layer of the acceleration data model, a difference value between the trajectory point estimated position data of the vehicle and the position data of the positioning module at each same time in the same time period of the trajectory point estimated position data sequence of the vehicle and the position data of the positioning module is calculated, so that a first group of errors between the trajectory point estimated position data sequence of the vehicle and the position data sequence of the positioning module in the same time period are calculated.

In step 306, the parameters of the i-th layer neurons of the fully-connected layer of the acceleration data model are obtained by making the i-th group error convergence smaller than a preset threshold.

In a specific embodiment, the parameters of the neurons in the first layer of the acceleration data model full-link layer can be continuously optimized by adjusting the parameters of the neurons in the first layer of the acceleration data model full-link layer, so that the first group of errors between the trajectory point estimation position data sequence of the vehicle in the same time period and the position data sequence of the positioning module are the smallest as possible, and the convergence is smaller than the preset threshold value, that is, the trajectory point estimation position data sequence of the vehicle in the same time period converges to the position data sequence of the positioning module, so as to obtain the parameters of the neurons in the first layer of the acceleration data model.

In step 307, the parameters of the i-th layer neurons of the fully-connected layer of the acceleration data model are set as the initial parameters of the i + 1-th layer neurons.

In one embodiment, the parameters of the first layer neurons of the acceleration data model full connectivity layer may be set as initial parameters of the second layer neurons.

In step 308, the acceleration data model outputs an acceleration estimation data sequence according to the acceleration sequence of the measurement data sequence by using the i +1 th layer neuron of the full connection layer, and estimates and obtains the track point estimated position of the vehicle at each time of the same time period according to the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, thereby obtaining the track point estimated position data sequence of the vehicle.

In one embodiment, the acceleration data model may be caused to predict an output acceleration estimation data sequence from an acceleration sequence of measurement data sequences of the inertial measurement units for a time period using neurons of a second layer of the fully-connected layer; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module in the same time period; and the acceleration data model can calculate the track point calculated position of the vehicle at each moment in a time period by using a second layer neuron of the full connection layer according to the initial pose of the vehicle, the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

In step 309, a group i +1 of errors between the trajectory point estimated position data sequence of the vehicle and the position data sequence of the positioning module are calculated.

In a specific embodiment, according to the trajectory point estimated position data sequence of the vehicle obtained by the neuron of the second layer of the full connection layer of the acceleration data model, a difference value between the trajectory point estimated position data of the vehicle and the position data of the positioning module at each same moment in the same time period of the trajectory point estimated position data sequence of the vehicle and the position data of the positioning module is calculated, so that a second group of errors between the trajectory point estimated position data sequence of the vehicle and the position data sequence of the positioning module in the same time period are calculated.

In step 310, parameters of i +1 layer neurons of the fully-connected layer of the acceleration data model are obtained by making the i +1 group error convergence smaller than a preset threshold.

In a specific embodiment, the parameters of the neurons in the second layer of the acceleration data model full connection layer can be continuously optimized by adjusting the parameters of the neurons in the second layer of the acceleration data model full connection layer, so that the second group of errors between the trajectory point estimation position data sequence of the vehicle in the same time period and the position data sequence of the positioning module are the smallest as possible, and the convergence is smaller than the preset threshold value, that is, the trajectory point estimation position data sequence of the vehicle in the same time period converges to the position data sequence of the positioning module, so as to obtain the parameters of the neurons in the second layer of the acceleration data model.

In one embodiment, and so on,

steps

307, 308, 309, and 310 are executed in a loop, and the parameters of each layer of neurons in the n layers of neurons in the full connection layer of the acceleration data model are obtained sequentially layer by layer, so as to complete the training of the acceleration data model.

It should be noted that the preset threshold values in step 306 and step 310 may be the same or different.

In step 311, the angular velocity data model sequentially uses the i-th layer neuron of the full connection layer to output the angular velocity estimation data sequence according to the angular velocity sequence of the measurement data sequence, and calculates and obtains the estimated position of the track point of the vehicle at each time in the same time period according to the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, thereby obtaining the estimated position data sequence of the track point of the vehicle.

In a specific embodiment, taking the first layer of neurons and the second layer of neurons of the angular velocity data model as an example, the first layer of neurons of the angular velocity data model is trained to obtain the parameters of the first layer of neurons. The angular velocity data model can utilize the neuron of the first layer of the full connection layer to predict and output the angular velocity estimation data sequence according to the angular velocity sequence of the measurement data sequence of the inertial measurement unit of a time period; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module in the same time period; and the angular velocity data model can calculate the track point calculated position of the vehicle at each moment in a time period by utilizing the first layer neuron of the full connection layer according to the initial pose of the vehicle, the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

In step 312, an ith group of errors of the track point estimated position data sequence of the vehicle and the position data sequence of the positioning module are calculated and obtained.

In a specific embodiment, according to the track point estimated position data sequence of the vehicle obtained by using the neurons in the first layer of the full connection layer according to the angular velocity data model, the difference value between the track point estimated position data of the vehicle and the position data of the positioning module at each same moment in the same time period of the track point estimated position data sequence of the vehicle and the position data of the positioning module is calculated, so that a first group of errors of the track point estimated position data sequence of the vehicle and the position data sequence of the positioning module in the same time period are calculated and obtained.

In step 313, the parameters of the i-th layer neurons of the fully-connected layer of the angular velocity data model are obtained by making the i-th group error convergence smaller than a preset threshold.

In a specific embodiment, the parameters of the neurons in the first layer of the full-link layer of the angular velocity data model can be continuously optimized by adjusting the parameters of the neurons in the first layer of the full-link layer of the angular velocity data model, so that the first group of errors between the trajectory point estimation position data sequence of the vehicle and the position data sequence of the positioning module in the same time period are the smallest as possible, and the convergence is smaller than the preset threshold value, that is, the trajectory point estimation position data sequence of the vehicle in the same time period is converged towards the position data sequence of the positioning module, so as to obtain the parameters of the neurons in the first layer of the angular velocity data model.

In step 314, the parameters for the i-th layer neurons of the fully-connected layer of the angular velocity data model are set to the initial parameters for the i + 1-th layer neurons.

In one embodiment, the parameters of the first layer neurons of the full connectivity layer of the angular velocity data model may be set as initial parameters of the second layer neurons.

In step 315, the angular velocity data model outputs an angular velocity estimation data sequence according to the angular velocity sequence of the measurement data sequence by using the i +1 th layer neuron of the full connection layer, and estimates and obtains the estimated position of the track point of the vehicle at each time of the same time period according to the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, thereby obtaining the estimated position data sequence of the track point of the vehicle.

In one embodiment, the angular velocity data model may be caused to predict an output angular velocity estimation data sequence from an angular velocity sequence of measurement data sequences of inertial measurement units for a time period using neurons of a second layer of the fully-connected layer; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module in the same time period; and the angular velocity data model can calculate the track point calculated position of the vehicle at each moment in a time period by using a second layer neuron of the full connection layer according to the initial pose of the vehicle, the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

In step 316, the i +1 th group of errors between the track point estimated position data sequence of the vehicle and the position data sequence of the positioning module are calculated.

In a specific embodiment, according to the track point estimated position data sequence of the vehicle obtained by using neurons in the second layer of the full connection layer according to the angular velocity data model, a difference value between the track point estimated position data of the vehicle and the position data of the positioning module at each same moment in the same time period of the track point estimated position data sequence of the vehicle and the position data of the positioning module is calculated, so that a second group of errors of the track point estimated position data sequence of the vehicle and the position data sequence of the positioning module in the same time period are calculated and obtained.

In step 317, parameters of i +1 layer neurons of the fully-connected layer of the angular velocity data model are obtained by making the i +1 group error convergence smaller than a preset threshold.

In a specific embodiment, the parameters of the neurons in the second layer of the full-link layer of the angular velocity data model can be continuously optimized by adjusting the parameters of the neurons in the second layer of the full-link layer of the angular velocity data model, so that the second group of errors between the trajectory point estimation position data sequence of the vehicle and the position data sequence of the positioning module in the same time period are the smallest as possible, and the convergence is smaller than the preset threshold value, that is, the trajectory point estimation position data sequence of the vehicle in the same time period is converged towards the position data sequence of the positioning module, so as to obtain the parameters of the neurons in the second layer of the angular velocity data model.

In one embodiment, and so on,

steps

314, 315, 316, and 317 are executed in a loop, and the parameters of each layer of neurons in the n layers of neurons in the full connection layer of the angular velocity data model are obtained sequentially layer by layer, so as to complete the training of the angular velocity data model.

It should be noted that the preset threshold values in step 313 and step 317 may be the same or different.

In step 318, the trained neural network model including the acceleration data model and the angular velocity data model is input with the measurement data sequence of the inertial measurement unit.

In step 319, the trained neural network model outputs an IMU estimation data sequence according to the measurement data sequence, and calculates a trajectory point calculation position sequence of the vehicle according to the IMU estimation data sequence, thereby obtaining a motion trajectory of the vehicle.

In a specific embodiment, the trained acceleration data model outputs acceleration estimation data at each moment according to the input acceleration of the vehicle accelerometer at each moment to obtain an acceleration estimation data sequence; enabling the trained angular velocity data model to output angular velocity estimation data at each moment according to the input angular velocity of the vehicle gyroscope at each moment, and obtaining an angular velocity estimation data sequence; acquiring an initial pose of the vehicle according to positioning information before the positioning module signal is unavailable, wherein the initial pose of the vehicle comprises position information, course angle information and speed information of the vehicle at the initial moment; and calculating the track point calculated position of each moment of the vehicle by the trained neural network model according to the initial pose, the acceleration estimated data sequence and the angular velocity estimated data sequence of the vehicle, and obtaining the track point calculated position sequence of the vehicle, thereby obtaining the motion track of the vehicle.

According to the vehicle trajectory calculation method based on the neural network model, neurons of a full connection layer of the neural network model to be trained are divided into multiple layers of neurons according to a set rule, a measurement data sequence of an inertia measurement unit and a position data sequence of a positioning module are used as samples, parameters of the multiple layers of neurons of the full connection layer of the neural network model to be trained are sequentially obtained layer by layer, training of each layer of neurons in the multiple layers of neurons of the neural network model to be trained is completed sequentially layer by layer, training difficulty of the neural network model is reduced, the parameters of the neurons of the full connection layer of the neural network model to be trained can be rapidly obtained, and training efficiency of the neural network model is improved; meanwhile, the trained neural network model is used for outputting an IMU estimation data sequence according to the measurement data sequence of the inertia measurement unit, the motion track of the vehicle is obtained according to the IMU estimation data sequence, the accumulated error caused by the fact that the vehicle track is calculated by the measurement data sequence of the inertia measurement unit can be reduced by the neural network model, and the accuracy of calculating the vehicle track by the measurement data of the inertia measurement unit is improved.

Further, in the vehicle trajectory estimation method based on the neural network model provided by the embodiment of the application, an acceleration data model and an angular velocity data model are set in the neural network model to be trained; sequentially dividing neurons of a full connection layer of the acceleration data model into n layers of neurons, sequentially dividing neurons of a full connection layer of the angular velocity data model into n layers of neurons, wherein the number of the i-th layer of neurons is the power i of 2, and the i + 1-th layer of neurons comprises the i-th layer of neurons, wherein n is a positive integer, and i =1,2,3, …, n. By respectively training the acceleration data model and the angular velocity data model, the training complexity of the neural network model can be reduced, and the training of the neural network model to be trained can be completed more quickly and accurately.

Further, according to the vehicle trajectory estimation method based on the neural network model, the parameter of the i-th layer neuron of the full connection layer is set as the initial parameter of the i + 1-th layer neuron, so that the training difficulty of the acceleration data model and the angular velocity data model can be reduced, samples for training the acceleration data model and the angular velocity data model can be reduced, the parameter of the full connection layer neuron of the acceleration data model and the angular velocity data model can be quickly obtained, and the training of the neural network model can be further accelerated.

Example four:

corresponding to the embodiment of the application function implementation method, the application also provides a vehicle track reckoning device based on the neural network model, electronic equipment and a corresponding embodiment.

Fig. 4 is a schematic structural diagram of a vehicle trajectory estimation device based on a neural network model according to an embodiment of the present application.

Referring to fig. 4, a vehicle trajectory estimation device based on a neural network model includes a first input module 401, a layering module 402, a training module 403, a second input module 404, and a trajectory estimation module 405.

The first input module 401 is configured to input a measurement data sequence of the inertial measurement unit and a position data sequence of the positioning module in the same time period to the neural network model to be trained.

In one specific implementation mode, the vehicle is provided with an inertia measurement unit, a positioning module and an on-board intelligent device loaded with a neural network model. The inertial measurement unit comprises an accelerometer and a gyroscope, and the measurement data sequence of the inertial measurement unit comprises an acceleration data sequence of the accelerometer of the inertial measurement unit and an angular velocity data sequence of the gyroscope. The first input module 401 may obtain a sequence of acceleration data of the vehicle through an accelerometer of the inertial measurement unit, and a sequence of angular velocity data of the vehicle through a gyroscope of the inertial measurement unit. The positioning module may include, but is not limited to, at least one of a GPS module, a beidou satellite positioning module, an RTK positioning module, and the like. With the positioning module, positioning information of the vehicle may be obtained, which may include, but is not limited to, position information, speed information, and attitude information. The location information includes, but is not limited to, latitude and longitude coordinate information describing the location, and the attitude information includes, but is not limited to, heading angle information describing the heading. The first input module 401 may obtain a sequence of position data of the vehicle according to the positioning information of the positioning module.

In one embodiment, the first input module 401 obtains a sequence of position data for a time period of the vehicle where the vehicle RTK positioning module signal is available. While the position data sequence of the vehicle for one time period is acquired by the RTK positioning module, the first input module 401 acquires the measurement data sequence of the vehicle for the same time period through the inertial measurement unit. It will be appreciated that the position data sequence acquired by the RTK positioning module acquired by the first input module 401 and the measurement data sequence acquired by the inertial measurement unit are both aligned in time to avoid inaccurate results due to the difference in reference times.

In one embodiment, the first input module 401 inputs the position data sequence and the measurement data sequence of the vehicle for the same time period to the neural network model to be trained. And the neural network model to be trained takes the position data sequence and the measurement data sequence of the vehicle in the same time period as samples to finish the training of the neural network model to be trained.

And the layering module 402 is configured to divide neurons in a full connection layer of the neural network model to be trained into multiple layers of neurons according to a set rule.

In one embodiment, the layering module 402 divides the neurons of the fully connected layer of the neural network model to be trained into a plurality of layers of neurons according to a set rule, wherein each layer comprises a set number of neurons. The layering module 402 may sequentially divide the neurons of the full connection layer of the neural network model to be trained into n layers of neurons, where the number of the i layer of neurons is the power i of 2, and the i +1 layer of neurons includes the i layer of neurons, where n is a positive integer, i =1,2,3, …, n. For example, there are 100 neurons in the fully-connected layer of the neural network model to be trained, the layering module 402 sequentially divides the 100 neurons into a plurality of layers of neurons, the first layer of neurons includes 2 (powers of 1 of 2) from 1 st to 2 nd, the second layer of neurons includes 4 (powers of 2) from 1 st to 4 th, the third layer of neurons includes 8 (powers of 3 of 2) from 1 st to 8 th, the fourth layer of neurons includes 16 (powers of 4 of 2) from 1 st to 16 th, the fifth layer of neurons includes 32 (powers of 5 of 2) from 1 st to 32 th, the sixth layer of neurons includes 64 (powers of 6 of 2) from 1 st to 64 th, and the seventh layer of neurons includes 100 (powers of 7 of 2 =128> 100) from 1 st to 100 th.

The training module 403 is configured to enable the neural network model to be trained to sequentially and layer-by-layer utilize multiple layers of neurons to output an IMU estimation data sequence according to the measurement data sequence input by the first input module 401, obtain a trajectory point estimation position data sequence of the vehicle according to the IMU estimation data sequence, and sequentially and layer-by-layer obtain parameters of the multiple layers of neurons according to the trajectory point estimation position data sequence of the vehicle and the position data sequence of the positioning module input by the first input module 401, thereby completing training of the neural network model to be trained.

In one embodiment, the training module 403 makes the neural network model to be trained sequentially use each layer of neurons in the multiple layers of neurons in the full connection layer, and outputs an IMU estimation data sequence according to the measurement data sequence of the inertial measurement unit input by the first input module 401; calculating to obtain a track point calculation position data sequence of the vehicle according to the IMU estimation data sequence; calculating errors of the track point calculated position data sequence of the vehicle and the position data sequence of the positioning module according to the track point calculated position data sequence of the vehicle and the position data sequence of the positioning module input by the first input module 401; and sequentially acquiring the parameters of each layer of neurons of the multiple layers of neurons of the full connection layer of the neural network model layer by making the error convergence smaller than a preset threshold value, thereby finishing the training of the neural network model to be trained.

In one embodiment, the training module 403 includes a trajectory submodule 413, an error calculation submodule 423, a parameter acquisition submodule 433, and a parameter setting submodule 443.

The trajectory sub-module 413 is configured to enable the neural network model to be trained to sequentially output an IMU estimation data sequence according to the measurement data sequence input by the first input module 401 by using the i-th layer neuron of the full connection layer, and to obtain a trajectory point estimation position of the vehicle at each time of the same time period according to the IMU estimation data sequence, so as to obtain a trajectory point estimation position data sequence of the vehicle.

In a specific embodiment, taking the first layer of neurons and the second layer of neurons as an example, the first layer of neurons of the neural network model is trained to obtain parameters of the first layer of neurons. The trajectory sub-module 413 may enable the neural network model to be trained to utilize the first layer neurons of the full connection layer to predict and output an IMU estimation data sequence according to the measurement data sequence of the inertial measurement unit of a time period input by the first input module 401; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module of the same time period input by the first input module 401; and the neural network model to be trained can calculate the track point calculated position of the vehicle at each moment in a time period by utilizing the first layer of neurons of the full connection layer according to the initial pose of the vehicle and the IMU estimated data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

The error calculating submodule 423 is configured to calculate an i-th group of errors between the track point estimated position data sequence of the vehicle obtained by the obtained track submodule 413 and the position data sequence of the positioning module input by the first input module 401.

In one specific embodiment, the error calculation sub-module 423 calculates a difference between the track point estimated position data of the vehicle obtained by the track sub-module 413 and the position data of the positioning module at each same time in the same time period and the position data of the positioning module input by the first input module 401, so as to calculate a first set of errors between the track point estimated position data of the vehicle obtained in the same time period and the position data of the positioning module.

And the parameter obtaining submodule 433 is configured to obtain the parameter of the i-th layer neuron of the full connection layer of the neural network model to be trained by making the i-th group of error convergence calculated and obtained by the error calculating submodule 423 smaller than a preset threshold.

In a specific embodiment, the parameter obtaining submodule 433 may continuously optimize the parameter of the first layer neuron of the fully-connected layer of the neural network model to be trained by adjusting the parameter of the first layer neuron of the fully-connected layer of the neural network model to be trained, so that the first group of errors calculated by the error calculating submodule 423 is as minimum as possible, and the convergence is smaller than the preset threshold, that is, the parameter of the first layer neuron of the neural network model to be trained is obtained by converging the trajectory point calculated position data sequence of the vehicle obtained by the trajectory submodule 413 to the position data sequence of the positioning module input by the first input module 401.

The parameter setting submodule 443 is configured to set the parameter of the i-th layer neuron of the full connection layer of the neural network model to be trained, which is obtained by the parameter obtaining submodule 433, as an initial parameter of the i + 1-th layer neuron.

In a specific embodiment, the parameter setting sub-module 443 may set the parameters of the first layer neurons of the fully-connected layer of the neural network model to be trained, which are obtained by the parameter obtaining sub-module 433, as the initial parameters of the second layer neurons.

The trajectory submodule 413 is further configured to enable the neural network model to be trained to output an IMU estimation data sequence according to the measurement data sequence input by the first input module 401 by using the i +1 th layer neuron of the full connection layer, and calculate and obtain a trajectory point estimated position of the vehicle at each time of the same time period according to the IMU estimation data sequence, so as to obtain a trajectory point estimated position data sequence of the vehicle.

In one embodiment, the trajectory sub-module 413 may enable the neural network model to be trained to utilize the second layer neurons of the full connection layer to predict and output an IMU estimation data sequence according to the measurement data sequence of the inertial measurement unit of a time period input by the first input module 401; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module of the same time period input by the first input module 401; and the neural network model to be trained can calculate the track point calculated position of the vehicle at each moment in a time period by utilizing a second layer neuron of the full connection layer according to the initial pose of the vehicle and the IMU estimated data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

The error calculating submodule 423 is further configured to calculate an i +1 th group of errors between the track point estimated position data sequence of the vehicle obtained by the obtained track submodule 413 and the position data sequence of the positioning module input by the first input module 401.

In one specific embodiment, the error calculation sub-module 423 calculates a difference between the track point estimated position data of the vehicle obtained by the track sub-module 413 and the position data of the positioning module at each same time in the same time period and the position data of the positioning module input by the first input module 401, so as to calculate a second set of errors between the track point estimated position data of the vehicle obtained in the same time period and the position data of the positioning module.

The parameter obtaining submodule 433 is further configured to obtain parameters of the i +1 th layer neuron of the full connection layer of the neural network model to be trained by making the i +1 th group of errors calculated by the error calculating submodule 423 converge to be smaller than a preset threshold.

In a specific embodiment, the parameter obtaining submodule 433 may continuously optimize the parameter of the second layer neuron of the fully-connected layer of the neural network model to be trained by adjusting the parameter of the second layer neuron of the fully-connected layer of the neural network model to be trained, so that the second group of errors calculated by the error calculating submodule 423 is as minimum as possible, and the convergence is smaller than the preset threshold, that is, the position data sequence of the positioning module input to the first input module 401 by the trajectory point calculated position data sequence of the vehicle obtained by the trajectory submodule 413 is converged, thereby obtaining the parameter of the second layer neuron of the neural network model to be trained.

In a specific embodiment, and so on, the trajectory sub-module 413, the error calculation sub-module 423, the parameter obtaining sub-module 433, and the parameter setting sub-module 443 of the training module 403 are executed in a loop, and the parameters of each layer of neurons in n layers of neurons in the full connection layer of the neural network model to be trained are obtained sequentially layer by layer, so as to complete the training of the neural network model to be trained.

And a second input module 404, configured to input the measurement data sequence of the inertial measurement unit to the trained neural network model.

In one embodiment, the second input module 404 inputs a sequence of measurement data of the inertial measurement unit to the trained neural network model in the event that the vehicle RTK positioning module signal is unavailable.

And a track calculating module 405, configured to enable the trained neural network model to output an IMU estimation data sequence according to the measurement data sequence input by the second input module 404, and calculate a track point calculated position sequence of the vehicle according to the IMU estimation data sequence, so as to obtain a motion track of the vehicle.

In one embodiment, the trajectory estimation module 405 causes the trained neural network model to output IMU estimation data at each time point according to the measurement data at each time point input by the second input module 404; according to positioning information before the positioning module signal is unavailable, obtaining the initial pose of the vehicle, and the position information, the course angle information and the speed information of the vehicle at the initial moment of the initial pose of the vehicle; and calculating the track point calculated position of each moment of the vehicle by the trained neural network model according to the initial pose of the vehicle and the IMU estimation data of each moment, and obtaining a track point calculated position sequence of the vehicle, thereby obtaining the motion track of the vehicle.

According to the technical scheme provided by the embodiment of the application, neurons of a full connection layer of a neural network model to be trained are divided into multiple layers of neurons according to a set rule, a measurement data sequence of an inertial measurement unit and a position data sequence of a positioning module are taken as samples, parameters of the multiple layers of neurons of the full connection layer of the neural network model to be trained are sequentially obtained layer by layer, training of the neurons of the neural network model to be trained is completed sequentially layer by layer, the training difficulty of the neural network model is reduced, the parameters of the neurons of the full connection layer of the neural network model to be trained can be rapidly obtained, and the training efficiency of the neural network model is improved; meanwhile, the trained neural network model is used for outputting an IMU estimation data sequence according to the measurement data sequence of the inertia measurement unit, the motion track of the vehicle is obtained according to the IMU estimation data sequence, the accumulated error caused by the fact that the vehicle track is calculated by the measurement data sequence of the inertia measurement unit can be reduced by the neural network model, and the accuracy of calculating the vehicle track by the measurement data sequence of the inertia measurement unit is improved.

Further, according to the technical scheme provided by the embodiment of the application, the parameter of the i-th layer neuron of the full connection layer of the neural network model to be trained is set as the initial parameter of the i + 1-th layer neuron, so that the difficulty of training the neural network model can be reduced, samples for training the neural network model can be reduced, the parameter of the full connection layer neuron of the neural network model to be trained can be quickly obtained, and the training of the neural network model is further accelerated.

Example five:

fig. 5 is a schematic structural diagram of a vehicle trajectory estimation device based on a neural network model according to another embodiment of the present application.

Referring to fig. 5, a vehicle trajectory estimation device based on a neural network model includes a first input module 401, a layering module 402, a training module 403, a second input module 404, a trajectory estimation module 405, and a model setting module 501.

The functions of the first input module 401 and the second input module 404 can be seen in fig. 4.

And the model setting module 501 is configured to set an acceleration data model and an angular velocity data model in the neural network model to be trained.

In one embodiment, the model setting module 501 sets an acceleration data model and an angular velocity data model in the neural network model to be trained based on different characteristics of an accelerometer and a gyroscope of the inertial measurement unit. The training module 403 trains the acceleration data model and the angular velocity data model respectively, so that the training complexity of the neural network model can be reduced, and the training of the neural network model to be trained can be completed more quickly and accurately.

The layering module 402 is specifically configured to sequentially divide neurons of a full connection layer of the acceleration data model into n layers of neurons, and sequentially divide neurons of a full connection layer of the angular velocity data model into n layers of neurons, where the number of i-th layer neurons is i power of 2, and the i + 1-th layer neurons include i-th layer neurons, where n is a positive integer, i =1,2,3, …, n.

In one embodiment, the layering module 402 sequentially divides the neurons of the fully-connected layer of the acceleration data model into n layers of neurons, each layer of neurons being 2 to the power i, where n is a positive integer, i =1,2, …, n, and the latter layer of neurons includes the former layer of neurons. For example, the fully-connected layer of the acceleration data model has 100 neurons, the layering module 402 sequentially divides the 100 neurons into a plurality of layers of neurons, the first layer of neurons includes 2 (powers of 1 of 2) total 1 to 2, the second layer of neurons includes 4 (powers of 2) total 1 to 4, the third layer of neurons includes 8 (powers of 3 of 2) total 1 to 8, the fourth layer of neurons includes 16 (powers of 4 of 2) total 1 to 16, the fifth layer of neurons includes 32 (powers of 5 of 2) total 1 to 32, the sixth layer of neurons includes 64 (powers of 6 of 2) total 1 to 64, and the seventh layer of neurons includes 100 (powers of 7 of 2 =128> 100) total 1 to 100.

In one embodiment, the layering module 402 sequentially divides the neurons of the fully-connected layer of the angular velocity data model into n layers of neurons, each layer of neurons being 2 to the power i, where n is a positive integer, i =1,2, …, n, and the latter layer of neurons includes the former layer of neurons. For example, the fully-connected layer of the angular velocity data model has 100 neurons, the layering module 402 sequentially divides the 100 neurons into a plurality of layers of neurons, the first layer of neurons includes 2 (power of 1 of 2) total neurons from 1 st to 2 nd, the second layer of neurons includes 4 (power of 2) total neurons from 1 st to 4 th, the third layer of neurons includes 8 (power of 3 of 2) total neurons from 1 st to 8 th, the fourth layer of neurons includes 16 (power of 4 of 2) total neurons from 1 st to 16 th, the fifth layer of neurons includes 32 (power of 5 of 2 total neurons from 1 st to 32 th), the sixth layer of neurons includes 64 (power of 6 of 2 total neurons from 1 st to 64 th, and the seventh layer of neurons includes 100 (power of 7 of 2 =128>100 total neurons from 1 to 100).

In one embodiment, the training module 403 includes a trajectory submodule 413, an error calculation submodule 423, a parameter acquisition submodule 433, and a parameter setting submodule 443. The trajectory submodule 413 includes a first trajectory submodule 4131, a second trajectory submodule 4132; the error calculation submodule 423 includes a first error calculation submodule 4231 and a second error calculation submodule 4232; the parameter obtaining submodule 433 includes a first parameter obtaining submodule 4331 and a second parameter obtaining submodule 4332; the parameter setting sub-module 443 includes a first parameter setting sub-module 4431 and a second parameter setting sub-module 4432.

The first trajectory submodule 4131 is configured to enable the acceleration data model to sequentially output an acceleration estimation data sequence according to the acceleration sequence of the measurement data sequence input by the first input module 401 by using the i-th layer neuron of the full connection layer, and estimate and obtain a trajectory point estimated position of the vehicle at each time of the same time period according to the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, thereby obtaining a trajectory point estimated position data sequence of the vehicle.

In a specific embodiment, taking a first layer of neurons and a second layer of neurons of an acceleration data model as an example, the first layer of neurons of the acceleration data model is trained to obtain parameters of the first layer of neurons. The first trajectory submodule 4131 can enable the acceleration data model to predict and output an acceleration estimation data sequence by using a first layer neuron of the full connection layer according to the acceleration sequence of the measurement data sequence of the inertial measurement unit of a time period input by the first input module 401; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module of the same time period input by the first input module 401; and the acceleration data model can calculate the track point calculated position of the vehicle at each moment in a time period by using the first layer neuron of the full connection layer according to the initial pose of the vehicle, the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

The first error calculation submodule 4231 is configured to calculate an i-th group of errors between the track point estimated position data sequence of the vehicle obtained by the first track submodule 4131 and the position data sequence of the positioning module input by the first input module 401.

In one specific embodiment, the first error calculation submodule 4231 calculates a difference between the trajectory point estimated position data of the vehicle obtained by the first trajectory submodule 4131 and the position data of the positioning module at each same time in the same time period and the position data of the positioning module input by the first input module 401, so as to calculate a first set of errors between the trajectory point estimated position data of the vehicle and the position data of the positioning module in the same time period.

The first parameter obtaining submodule 4331 is configured to obtain parameters of i-th layer neurons of the full connection layer of the acceleration data model by making the i-th group of error convergence calculated and obtained by the first error calculating submodule 4231 smaller than a preset threshold.

In a specific embodiment, the first parameter obtaining submodule 4331 may continuously optimize the parameters of the first layer of neurons in the acceleration data model full connection layer by adjusting the parameters of the first layer of neurons in the acceleration data model full connection layer, so that the first group of errors calculated by the first error calculating submodule 4231 is the smallest as possible, and the convergence is smaller than the preset threshold, that is, the first parameter obtaining submodule 4331 obtains the parameters of the first layer of neurons in the acceleration data model by converging the trajectory point calculated position data sequence of the vehicle obtained by the first trajectory submodule 4131 to the position data sequence of the positioning module input by the first input module 401.

The first parameter setting submodule 4431 is configured to set the parameter of the i-th layer neuron of the full connection layer of the acceleration data model, which is obtained by the first parameter obtaining submodule 4331, as the initial parameter of the i + 1-th layer neuron.

In one embodiment, the first parameter setting submodule 4431 may set the parameters of the first layer neurons of the acceleration data model full-link layer to the initial parameters of the second layer neurons.

The first trajectory sub-module 4131 is further configured to enable the acceleration data model to output an acceleration estimation data sequence according to the acceleration sequence of the measurement data sequence input by the first input module 401 by using the i +1 th layer neuron of the full connection layer, and estimate and obtain a trajectory point estimated position of the vehicle at each time of the same time period according to the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, thereby obtaining a trajectory point estimated position data sequence of the vehicle.

In one embodiment, the first trajectory sub-module 4131 may enable the acceleration data model to predict an output acceleration estimation data sequence using neurons of the second layer of the fully-connected layer from the acceleration sequence of the measurement data sequence of the inertial measurement unit for a time period input by the first input module 401; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module of the same time period input by the first input module 401; and the acceleration data model can calculate the track point calculated position of the vehicle at each moment in a time period by using a second layer neuron of the full connection layer according to the initial pose of the vehicle, the acceleration estimation data sequence and the angular velocity sequence of the measurement data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

The first error calculating submodule 4231 is further configured to calculate an i +1 th group of errors between the track point estimated position data sequence of the vehicle obtained by the first track submodule 4131 and the position data sequence of the positioning module input by the first input module 401.

In a specific embodiment, the first error calculation submodule 4231 calculates a difference between the trajectory point estimated position data of the vehicle obtained by the first trajectory submodule 4131 and the position data of the positioning module at each same time in the same time period and the position data of the positioning module input by the first input module 401, so as to calculate a second set of errors between the trajectory point estimated position data of the vehicle and the position data of the positioning module in the same time period.

The first parameter obtaining submodule 4331 is further configured to obtain parameters of neurons in the i +1 th layer of the full-link layer of the acceleration data model by making the i +1 th group of errors calculated by the first error calculating submodule 4231 converge to be smaller than a preset threshold.

In a specific embodiment, the first parameter obtaining submodule 4331 may continuously optimize the parameters of the neurons in the second layer of the acceleration data model full-link layer by adjusting the parameters of the neurons in the second layer of the acceleration data model full-link layer, so that the second group of errors calculated by the first error calculating submodule 4231 is the smallest as possible, and the convergence is smaller than the preset threshold, that is, the parameters of the neurons in the second layer of the acceleration data model are obtained by converging the trajectory point calculated position data sequence of the vehicle obtained by the first trajectory submodule 4131 to the position data sequence of the positioning module input by the first input module 401.

In a specific embodiment, and so on, the first parameter setting submodule 4431, the first trajectory submodule 4131, the first error calculation submodule 4231, and the first parameter obtaining submodule 4331 are executed in a loop, and the parameters of each layer of neurons in the n layers of neurons in the full connection layer of the acceleration data model are obtained layer by layer in sequence, so that the training of the acceleration data model is completed.

The second trajectory submodule 4132 is configured to enable the angular velocity data model to sequentially output an angular velocity estimation data sequence according to the angular velocity sequence of the measurement data sequence input by the first input module 401 by using the i-th layer neuron of the full connection layer, and calculate and obtain a trajectory point calculated position of the vehicle at each time of the same time period according to the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, thereby obtaining a trajectory point calculated position data sequence of the vehicle.

In a specific embodiment, taking the first layer of neurons and the second layer of neurons of the angular velocity data model as an example, the first layer of neurons of the angular velocity data model is trained to obtain the parameters of the first layer of neurons. The second trajectory submodule 4132 may cause the angular velocity data model to predict an output angular velocity estimation data sequence using a first layer neuron of the full connection layer from the angular velocity sequence of the measurement data sequence of the inertial measurement unit for one time period input by the first input module 401; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module of the same time period input by the first input module 401; and the angular velocity data model can calculate the track point calculated position of the vehicle at each moment in a time period by utilizing the first layer neuron of the full connection layer according to the initial pose of the vehicle, the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

The second error calculation submodule 4232 is configured to calculate an i-th group of errors between the track point estimated position data sequence of the vehicle obtained by the second track submodule 4132 and the position data sequence of the positioning module input by the first input module 401.

In a specific embodiment, the second error calculation submodule 4232 calculates a difference between the trajectory point estimated position data of the vehicle obtained by the second trajectory submodule 4132 and the position data of the positioning module at each same time in the same time period and the position data of the positioning module input by the first input module 401, so as to calculate a first set of errors between the trajectory point estimated position data of the vehicle and the position data of the positioning module in the same time period.

The second parameter obtaining submodule 4332 is configured to obtain parameters of i-th layer neurons of the full-link layer of the angular velocity data model by making the i-th group of error convergence calculated and obtained by the second error calculating submodule 4232 smaller than a preset threshold.

In a specific embodiment, the second parameter obtaining submodule 4332 may continuously optimize the parameters of the first layer of neurons in the full-link layer of the angular velocity data model by adjusting the parameters of the first layer of neurons in the full-link layer of the angular velocity data model, so that the first group of errors calculated by the second error calculating submodule 4232 is the smallest as possible, and the convergence is smaller than the preset threshold, that is, the position data sequence of the positioning module input to the first input module 401 is converged by the trajectory point estimated position data sequence of the vehicle obtained by the second trajectory submodule 4132, thereby obtaining the parameters of the first layer of neurons in the angular velocity data model.

The second parameter setting submodule 4432 is configured to set the parameter of the i-th layer neuron of the full-link layer of the angular velocity data model obtained by the second parameter obtaining submodule 4332 as the initial parameter of the i + 1-th layer neuron.

In one embodiment, the second parameter setting sub-module 4432 may set the parameters of the first layer neurons of the full connectivity layer of the angular velocity data model as the initial parameters of the second layer neurons.

The second trajectory sub-module 4132 is further configured to enable the angular velocity data model to output the angular velocity estimation data sequence according to the angular velocity sequence of the measurement data sequence input by the first input module 401 by using the i +1 th layer neuron of the full connection layer, and calculate and obtain the trajectory point estimated position of the vehicle at each time of the same time period according to the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, thereby obtaining the trajectory point estimated position data sequence of the vehicle.

In one embodiment, the second trajectory submodule 4132 may enable the angular velocity data model to predict the output angular velocity estimation data sequence using the second layer neurons of the fully-connected layer from the angular velocity sequence of the measurement data sequence of the inertial measurement unit for a time period input by the first input module 401; acquiring an initial pose of the vehicle according to the position data sequence of the RTK positioning module of the same time period input by the first input module 401; and the angular velocity data model can calculate the track point calculated position of the vehicle at each moment in a time period by using a second layer neuron of the full connection layer according to the initial pose of the vehicle, the angular velocity estimation data sequence and the acceleration sequence of the measurement data sequence, so that the track point calculated position data sequence of the vehicle is obtained.

The second error calculation submodule 4232 is further configured to calculate an i +1 th group of errors between the track point estimated position data sequence of the vehicle obtained by the second track submodule 4132 and the position data sequence of the positioning module input by the first input module 401.

In a specific embodiment, the second error calculation submodule 4232 calculates a difference between the trajectory point estimated position data of the vehicle obtained by the second trajectory submodule 4132 and the position data of the positioning module at each same time in the same time period and the position data of the positioning module input by the first input module 401, so as to calculate a second set of errors between the trajectory point estimated position data of the vehicle and the position data of the positioning module in the same time period.

The second parameter obtaining submodule 4332 is further configured to obtain parameters of i +1 th layer neurons of the full-link layer of the angular velocity data model by making the i +1 th group of error convergence calculated and obtained by the second error calculating submodule 4232 smaller than a preset threshold.

In a specific embodiment, the second parameter obtaining submodule 4332 may continuously optimize the parameters of the second layer neurons of the full-link layer of the angular velocity data model by adjusting the parameters of the second layer neurons of the full-link layer of the angular velocity data model, so that the second group of errors calculated by the second error calculating submodule 4232 is the smallest as possible, and the convergence is smaller than the preset threshold, that is, the second parameter obtaining submodule 4332 converges the position data sequence of the positioning module input to the first input module 401 by the trajectory point estimated position data sequence of the vehicle obtained by the second trajectory submodule 4132, thereby obtaining the parameters of the second layer neurons of the angular velocity data model.

In a specific embodiment, and so on, the second parameter setting submodule 4432, the second trajectory submodule 4132, the second error calculation submodule 4232, and the second parameter obtaining submodule 4332 are executed in a loop, and the parameters of each layer of neurons in the n layers of neurons in the full connection layer of the angular velocity data model are obtained layer by layer in sequence, so as to complete the training of the angular velocity data model.

And a trajectory calculation module 405, configured to enable the trained neural network model to output an IMU estimation data sequence according to the measurement data sequence, and calculate a trajectory point calculation position sequence of the vehicle according to the IMU estimation data sequence, so as to obtain a motion trajectory of the vehicle.

In one specific embodiment, the trajectory estimation module 405 enables the trained acceleration data model to output acceleration estimation data at each time according to the acceleration of the vehicle accelerometer at each time input by the second input module 404, so as to obtain an acceleration estimation data sequence; enabling the trained angular velocity data model to output angular velocity estimation data at each moment according to the angular velocity of the vehicle gyroscope at each moment input by the second input module 404, and obtaining an angular velocity estimation data sequence; according to positioning information before the positioning module signal is unavailable, obtaining the initial pose of the vehicle, and the position information, the course angle information and the speed information of the vehicle at the initial moment of the initial pose of the vehicle; and calculating the track point calculated position of the vehicle and the track point calculated position sequence of the vehicle according to the initial pose, the acceleration estimated data sequence and the angular velocity estimated data sequence of the vehicle by using the trained neural network model, thereby obtaining the motion track of the vehicle.

According to the technical scheme provided by the embodiment of the application, neurons of a full connection layer of a neural network model to be trained are divided into multiple layers of neurons according to a set rule, a measurement data sequence of an inertial measurement unit and a position data sequence of a positioning module are taken as samples, parameters of the multiple layers of neurons of the full connection layer of the neural network model to be trained are sequentially obtained layer by layer, training of the neurons of the neural network model to be trained is completed sequentially layer by layer, the training difficulty of the neural network model is reduced, the parameters of the neurons of the full connection layer of the neural network model to be trained can be rapidly obtained, and the training efficiency of the neural network model is improved; meanwhile, the trained neural network model is used for outputting an IMU estimation data sequence according to the measurement data sequence of the inertia measurement unit, the motion track of the vehicle is obtained according to the IMU estimation data sequence, the accumulated error caused by the fact that the vehicle track is calculated by the measurement data sequence of the inertia measurement unit can be reduced by the neural network model, and the accuracy of calculating the vehicle track by the measurement data of the inertia measurement unit is improved.

With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.

Referring to fig. 6, the electronic device 60 includes a memory 601 and a processor 602.

The Processor 602 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic device, discrete hardware component, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory 601 may include various types of storage units, such as system memory, Read Only Memory (ROM), and permanent storage. Wherein the ROM may store static data or instructions for the processor 602 or other modules of the computer. The persistent storage device may be a read-write storage device. The persistent storage may be a non-volatile storage device that does not lose stored instructions and data even after the computer is powered off. In some embodiments, the persistent storage device employs a mass storage device (e.g., magnetic or optical disk, flash memory) as the persistent storage device. In other embodiments, the permanent storage may be a removable storage device (e.g., floppy disk, optical drive). The system memory may be a read-write memory device or a volatile read-write memory device, such as a dynamic random access memory. The system memory may store instructions and data that some or all of the processors require at runtime. In addition, the memory 601 may include any combination of computer-readable storage media, including various types of semiconductor memory chips (DRAM, SRAM, SDRAM, flash memory, programmable read-only memory), magnetic and/or optical disks, may also be employed. In some embodiments, memory 601 may include a removable storage device that is readable and/or writable, such as a Compact Disc (CD), a digital versatile disc read only (e.g., DVD-ROM, dual layer DVD-ROM), a Blu-ray disc read only, an ultra-dense disc, a flash memory card (e.g., SD card, min SD card, Micro-SD card, etc.), a magnetic floppy disk, or the like. Computer-readable storage media do not contain carrier waves or transitory electronic signals transmitted by wireless or wired means.

The memory 601 has stored thereon executable code that, when processed by the processor 602, may cause the processor 602 to perform some or all of the methods described above.

Furthermore, the method according to the present application may also be implemented as a computer program or computer program product comprising computer program code instructions for performing some or all of the steps of the above-described method of the present application.

Alternatively, the present application may also be embodied as a non-transitory machine-readable storage medium (or computer-readable storage medium, or machine-readable storage medium) having stored thereon executable code (or a computer program, or computer instruction code) which, when executed by a processor of an electronic device (or electronic device, server, etc.), causes the processor to perform some or all of the various steps of the above-described methods in accordance with the present application.

Having described embodiments of the present application, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A vehicle track calculation method based on a neural network model is characterized by comprising the following steps:

2. The method of claim 1, wherein the dividing neurons of the fully-connected layer of the neural network model to be trained into a plurality of layers of neurons according to a set rule comprises:

3. The method of claim 2, wherein the dividing the neurons of the fully-connected layer of the neural network model to be trained into n layers of neurons sequentially, the number of the i layer of neurons being 2 to the power i, the i +1 layer of neurons including the i layer of neurons, wherein n is a positive integer, i =1,2,3, …, n, comprises:

4. The method according to claim 3, wherein the enabling the neural network model to be trained to sequentially and layer-by-layer utilize the multiple layers of neurons to output an IMU estimation data sequence according to the measurement data sequence, obtain a trajectory point estimation position data sequence of a vehicle according to the IMU estimation data sequence, and sequentially and layer-by-layer obtain parameters of the multiple layers of neurons according to the trajectory point estimation position data sequence of the vehicle and the position data sequence of the positioning module, thereby completing the training of the neural network model to be trained, comprises:

5. The method of claim 4, further comprising:

7. The apparatus of claim 6, further comprising:

8. The apparatus of claim 7, wherein the training module comprises:

9. The apparatus of claim 8, wherein the training module further comprises:

10. A non-transitory machine-readable storage medium having executable code stored thereon, which when executed by a processor of an electronic device, causes the processor to perform the method of any one of claims 1-5.