A test vehicle fusion positioning method

Abstract

The present invention provides a fusion positioning method for a test vehicle, comprising: obtaining a UWB positioning parameter vector and an AI visual positioning parameter vector of the test vehicle to form a vector matrix; training a positioning deep learning model based on multiple sets of vector matrices; predicting the test vehicle's position coordinates based on the positioning deep learning model; calculating the difference between the model's predicted position coordinates and the test vehicle's RTK position coordinates under normal signal conditions; and adjusting the model parameters of the positioning deep learning model based on the difference to obtain an optimized positioning deep learning model. The present invention integrates multiple positioning methods, including UWB positioning, AI visual positioning, deep learning model positioning, and RTK positioning, to achieve UWB (ultra-wideband) positioning data as the underlying data. Using AI vision and deep learning model fusion technology, the method overcomes low positioning accuracy, generates standardized test positioning data, and improves test efficiency and pass rate.

Description

Technical Field

[0001] The present invention relates to the field of V2X testing, and more specifically, to a method for testing vehicle fusion positioning. Background Art

[0002] In today's era of booming intelligent transportation and autonomous driving, V2X, as a key supporting technology, plays a vital role in vehicle positioning and information exchange. To ensure the safety and reliability of V2X technology and autonomous driving systems, extensive testing is carried out in closed test fields.

[0003] Closed test fields can simulate a wide variety of real-world scenarios, creating a controlled testing environment for V2X equipment and vehicles, facilitating comprehensive performance evaluation and optimization. However, relevant standards, such as CSAE 246-2022, require high positioning accuracy of up to two centimeters for V2X test vehicles in closed test fields. However, in reality, positioning accuracy in existing closed test field V2X testing is often limited by numerous factors, failing to meet the test standard's requirements.

[0004] Satellite positioning signals, a crucial basis for vehicle positioning, are susceptible to interference from the surrounding environment in practical applications. For example, buildings within a closed test site can block satellite signals, significantly reducing the accuracy of satellite-based positioning. This not only severely impacts the accuracy of test results, but also fails to meet the validity and passability requirements of the standard.

[0005] In closed test areas with special environments, such as signal-blocking chambers and underpasses, even with the use of UWB positioning equipment, its positioning accuracy still failed to meet the stringent requirements of the test standards. This significantly compromised the reliability of the test data, making it unable to provide strong support for the optimization and improvement of V2X technology, and further hindering the development and application of intelligent transportation and autonomous driving technologies. Summary of the Invention

[0006] The present invention aims to solve the problem of insufficient vehicle positioning accuracy in existing closed test sites under weak satellite signal environments, and provides a test vehicle fusion positioning method, comprising:

[0007] Get the UWB positioning parameter vector and AI visual positioning parameter vector of the test vehicle. The UWB positioning parameter vector is expressed as ,in, To test the moment when the vehicle is powered on and starts positioning, is the time required for the jth processing stage of the i-th base station, and the AI ​​visual positioning parameter vector is expressed as , where t is the camera's translation vector, , The horizontal and vertical coordinates of the AI ​​visual positioning position coordinates of the vehicle to be tested respectively;

[0008] Combining the UWB positioning parameter vector and the AI ​​visual positioning parameter vector into a vector matrix, and obtaining multiple sets of vector matrices as multiple training samples;

[0009] Train the positioning deep learning model based on multiple training samples;

[0010] Predicting the position coordinates of the test vehicle based on the trained positioning deep learning model;

[0011] Calculating the difference between the predicted position coordinates of the test vehicle by the positioning deep learning model and the RTK position coordinates of the test vehicle under normal signal conditions;

[0012] Adjusting model parameters of the positioning deep learning model based on the gap to obtain an optimized positioning deep learning model;

[0013] The position coordinates of the vehicle to be tested are predicted based on the optimized positioning deep learning model.

[0014] The present invention provides a test vehicle fusion positioning method that integrates multiple positioning methods such as UWB positioning, AI vision positioning, deep learning model positioning, and RTK positioning, realizing UWB (ultra-wideband) positioning data as the underlying data, using AI vision and deep learning model fusion technology to compensate for the problem of low positioning accuracy, generate standard test positioning data, and improve the efficiency and pass rate of the test. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a test vehicle fusion positioning method provided by the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention can be arbitrarily combined with each other to form a feasible technical solution. This combination is not restricted by the sequence of steps and / or structural composition mode, but must be based on the ability of ordinary technicians in this field to implement it. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0017] While closed test sites currently provide a controlled testing environment for V2X equipment and vehicles, existing positioning technologies struggle to achieve the desired accuracy in complex environments, such as signal-blocking chambers and underpasses. Satellite navigation-based positioning methods often experience reduced accuracy due to the easy obstruction of satellite signals by surrounding buildings. While UWB positioning devices have been employed as a supplement, their effectiveness remains unsatisfactory, failing to meet the stringent two-centimeter positioning accuracy requirements of relevant standards.

[0018] To overcome these challenges, the present invention incorporates innovations and improvements in multiple areas. First, by integrating multiple positioning technologies, rather than relying solely on satellite navigation or UWB positioning, the system leverages the strengths of various technologies, including AI vision and deep learning, to achieve the fusion and complementarity of multi-source data. Furthermore, after entering a normal signal area, the system uses the normal RTK signal docking data to perform prediction corrections, further improving positioning accuracy.

[0019] See also Figure 1 The present invention provides a test vehicle fusion positioning method, the method comprising:

[0020] Step 1: Obtain the UWB positioning parameter vector and AI visual positioning parameter vector of the test vehicle. The UWB positioning parameter vector is expressed as ,in, To test the moment when the vehicle is powered on and starts positioning, is the time required for the jth processing stage of the i-th base station, and the AI ​​visual positioning parameter vector is expressed as , where t is the camera's translation vector, , The horizontal and vertical coordinates of the AI ​​visual positioning position coordinates of the vehicle to be tested.

[0021] It is understood that the embodiments of the present invention integrate UWB positioning methods and AI visual positioning methods, and then train a deep learning model. The UWB positioning parameter vector and AI visual positioning parameter vector of the test vehicle are obtained separately, and then the positioning deep learning model is trained based on the UWB positioning parameter vector and the AI ​​visual positioning parameter vector.

[0022] Among them, the principle of positioning the test vehicle based on the UWB positioning method is:

[0023] The position coordinates of the vehicle to be tested can be obtained using the trilateration method, which requires at least three base stations. Let i be the base station number, and the signal transmission time from the i-th base station to the test vehicle is:

[0024] ;

[0025] Where i represents the serial number of the base station, 、 、 and They represent the time required by the i-th base station in the four signal processing stages.

[0026] The distance from the i-th base station to the test vehicle is:

[0027]

[0028] Where c is the pulse velocity.

[0029] The specific positioning of the test vehicle using UWB positioning is as follows:

[0030] Based on the coordinates of the three base stations and the coordinates of the test vehicle, calculate the distance between each base station and the test vehicle:

[0031] ;

[0032] in, are the coordinates of the test vehicle to be tested, represents the coordinates of the i-th base station.

[0033] joint:

[0034] ;

[0035] Solve the coordinates of the test vehicle to be tested .

[0036] Among them, As the UWB positioning parameter vector.

[0037] The vehicle is positioned based on the AI ​​visual positioning method. The specific process is as follows:

[0038] The camera is captured at a frequency of 100 Hz. First, the camera is calibrated to obtain the camera parameters, including the rotation matrix R and translation vector t of the camera in the world coordinate system, focal length fx, fy, principal point coordinates cx, cy and distortion coefficients, which are obtained through the calibration tool:

[0039] .

[0040] Then use the Faster R-CNN algorithm to detect the vehicle and output the pixel coordinates (u, v) of the center or corner of the bounding box of the test vehicle. At time T, the pixel coordinates are converted to world coordinates through inverse perspective mapping (IPM), the image is corrected using the calibrated distortion coefficients, and the mapping matrix H from the image coordinate system to the ground coordinate system is obtained through calibration. Convert the pixel coordinates (u, v) to ground coordinates :

[0041] .

[0042] Will Added as a parameter to the AI ​​visual positioning parameter vector.

[0043] Step 2: Combining the UWB positioning parameter vector and the AI ​​visual positioning parameter vector into a vector matrix, and obtaining multiple groups of vector matrices as multiple training samples.

[0044] It can be understood that the UWB positioning parameter vector and the AI ​​visual positioning parameter vector are combined into a vector matrix, which is expressed as follows:

[0045] ;

[0046] Among them, the UWB parameter vector It needs to be normalized to be expressed as:

[0047] .

[0048] Each vector matrix is ​​taken as a training sample to obtain multiple training samples.

[0049] Step 3: Train the positioning deep learning model based on multiple training samples.

[0050] It is understandable that multiple training samples are obtained and the positioning deep learning model is trained based on the multiple training samples.

[0051] First, a positioning deep learning model is designed. In an embodiment of the present invention, the positioning deep learning model adopts Same convolution. The positioning deep learning model includes an input layer, a convolution layer, a pooling layer, a Flatten layer, a fully connected layer, and an output layer. The positioning deep learning model is trained based on multiple training samples, including:

[0052] Inputting a vector matrix into the input layer, and performing convolution on the vector matrix based on the convolution layer;

[0053] Performing maximum pooling processing on the output result of the convolutional layer based on the pooling layer;

[0054] The output matrix of the pooling layer is pulled into a flat one-dimensional vector based on the Flatten layer;

[0055] Outputting the one-dimensional vector output by the Flatten layer into a two-dimensional vector based on the fully connected layer;

[0056] The predicted position coordinates of the test vehicle are output through the output layer.

[0057] In the process of training the positioning deep learning model, the loss value of the positioning deep learning model is calculated based on the fusion objective function, and the fusion objective function is:

[0058] ;

[0059] Among them, Loss represents the loss value of the positioning deep learning model. represents the predicted position coordinates of the test vehicle at time k by the deep learning model, represents the UWB positioning coordinates of the test vehicle at time k, It represents the AI ​​visual positioning coordinates of the test vehicle at time k, and n represents the number of moments, which can also be understood as n training samples.

[0060] Adjust the positioning deep learning model based on the loss value, for example, adjust the network structure of the positioning deep learning model, etc., to minimize the loss as much as possible.

[0061] Step 4: Predict the position coordinates of the test vehicle based on the trained positioning deep learning model.

[0062] It is understandable that after step 3 trains the positioning deep learning model, the trained positioning deep learning model is used to test the test set, and the positioning deep learning model is optimized based on the test results of the test set.

[0063] Specifically, the UWB positioning parameter vector and the AI ​​visual positioning parameter vector of the test vehicle are obtained as test samples according to the preset sampling time interval T5; m test samples with time conversion gaps are obtained, where m is a positive integer and can be understood as test samples at m moments; the m test samples are predicted based on the trained positioning deep learning model to obtain the predicted position coordinates of the m test samples, which are expressed as:

[0064] .

[0065] Step 5: Calculate the difference between the predicted position coordinates of the test vehicle by the positioning deep learning model and the RTK position coordinates of the test vehicle under normal signals.

[0066] It is understandable that the RTK position coordinates of the test vehicle at the current moment are obtained after the test vehicle is in the normal signal area. , sample the RTK signal at the same preset sampling time interval T5, and test the real-time speed of the vehicle at the current moment and the angle between the test vehicle and the horizontal axis of the coordinate system , estimate the RTK position coordinates of the test vehicle at the previous moment ,in: ;

[0067] ;

[0068] in, is the RTK position coordinate of the test vehicle at the current time m, is the estimated RTK position coordinate of the test vehicle at the last moment m-1.

[0069] The RTK position coordinates of the test vehicle at m moments are estimated using the same estimation method.

[0070] Calculate the difference between the predicted position coordinates of the m test samples of the test vehicle and the estimated RTK position coordinates of the m test vehicles according to the positioning deep learning model. .gap The calculation formula is:

[0071] ;

[0072] in, Indicates the RTK position coordinates at time i, Represents the predicted position coordinates of the test vehicle at time g output by the positioning deep learning model.

[0073] Step 6: Adjust the model parameters of the positioning deep learning model based on the gap to obtain the optimized positioning deep learning model.

[0074] Understandably, based on the gap Adjust the model parameters of the positioning deep learning model to obtain an optimized positioning deep learning model. Specifically including:

[0075] Set the step size adjustment parameter ω of the positioning deep learning model. Based on the step size adjustment parameter ω, update the fusion objective function of the positioning deep learning model: ;

[0076] in, Represents the loss value of the positioning deep learning model, represents the predicted position coordinates of the test vehicle at time k by the deep learning model, represents the UWB positioning coordinates of the test vehicle at time k, represents the AI ​​visual positioning coordinates of the test vehicle at time k, and n represents the number of moments.

[0077] Continuously adjust the step size adjustment parameter ω of the positioning deep learning model to reduce the gap between the predicted position coordinates of the test vehicle by the positioning deep learning model and the RTK position coordinates under normal signals. As small as possible, the positioning deep learning model after preliminary training is optimized to obtain an optimized positioning deep learning model.

[0078] Step 7: Predict the position coordinates of the vehicle to be tested based on the optimized positioning deep learning model.

[0079] It can be understood that the above steps optimize the positioning deep learning model according to the RTK position coordinates of the test vehicle, and predict the position coordinates of the test vehicle based on the optimized positioning deep learning model.

[0080] An embodiment of the present invention provides a test vehicle fusion positioning method, which effectively integrates UWB positioning, AI vision positioning, deep learning models and RTK positioning, and uses the positioning information provided by AI vision to complement the UWB positioning data after optimization in the deep learning model, thereby improving positioning accuracy and reliability with the help of a fusion algorithm. When the satellite signal is blocked, AI vision can provide continuous positioning, overcoming the limitations of a single technology in complex environments. In addition, AI vision can accurately extract vehicle position data and establish a positioning model with centimeter-level accuracy, solving the problem of large positioning errors in closed test sites with satellite signal obstruction, thereby improving positioning accuracy and test efficiency and pass rate.

[0081] It should be noted that, in the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0082] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0084] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0086] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0087] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A test vehicle fusion positioning method, characterized in that: include: Get the UWB positioning parameter vector and AI visual positioning parameter vector of the test vehicle. The UWB positioning parameter vector is expressed as ,in, To test the moment when the vehicle is powered on and starts positioning, is the time required for the jth processing stage of the i-th base station, and the AI ​​visual positioning parameter vector is expressed as , where t is the camera's translation vector, , The horizontal and vertical coordinates of the AI ​​visual positioning position coordinates of the vehicle to be tested respectively; Combining the UWB positioning parameter vector and the AI ​​visual positioning parameter vector into a vector matrix, and obtaining multiple sets of vector matrices as multiple training samples; Train the positioning deep learning model based on multiple training samples; Predicting the position coordinates of the test vehicle based on the trained positioning deep learning model; Calculating the difference between the predicted position coordinates of the test vehicle by the positioning deep learning model and the RTK position coordinates of the test vehicle under normal signal conditions; Adjusting model parameters of the positioning deep learning model based on the gap to obtain an optimized positioning deep learning model; The position coordinates of the vehicle to be tested are predicted based on the optimized positioning deep learning model.

2. The test vehicle fusion positioning method according to claim 1, characterized in that: Obtaining the AI ​​visual positioning parameter vector includes: Calibrate the camera parameters, including the camera's rotation matrix R and translation vector t in the world coordinate system, the camera's focal lengths fx and fy, the principal point coordinates cx and cy, and the camera's distortion coefficients; Detect the test vehicle based on the Faster R-CNN algorithm and output the pixel position coordinates (u, v) of the test vehicle; Based on the camera parameters, the pixel position coordinates of the test vehicle are converted to ground coordinates: ； ； in, Altitude for testing vehicles; Will As the AI ​​visual positioning parameter vector.

3. The test vehicle fusion positioning method according to claim 1, characterized in that: The UWB positioning parameter vector and the AI ​​visual positioning parameter vector are combined into a vector matrix, including: The UWB positioning parameter vector Perform normalization: ； The normalized UWB positioning parameter vector and the AI ​​visual positioning parameter vector are combined into a vector matrix.

4. The test vehicle fusion positioning method according to claim 1, characterized in that: The positioning deep learning model includes an input layer, a convolutional layer, a pooling layer, a flatten layer, a fully connected layer, and an output layer. The positioning deep learning model is trained based on multiple training samples, including: Inputting a vector matrix into the input layer, and performing convolution on the vector matrix based on the convolution layer; Performing maximum pooling processing on the output result of the convolutional layer based on the pooling layer; The output matrix of the pooling layer is pulled into a flat one-dimensional vector based on the Flatten layer; Outputting the one-dimensional vector output by the Flatten layer into a two-dimensional vector based on the fully connected layer; The predicted position coordinates of the test vehicle are output through the output layer.

5. The test vehicle fusion positioning method according to claim 1 or 4, characterized in that: When training the positioning deep learning model, the loss value of the positioning deep learning model is calculated based on the fusion objective function, and the fusion objective function is: ； in, Represents the loss value of the positioning deep learning model, represents the predicted position coordinates of the test vehicle at time k by the deep learning model, represents the UWB positioning coordinates of the test vehicle at time k, It represents the AI ​​visual positioning coordinates of the test vehicle at time k, and n represents the number of moments, that is, the number of training samples.

6. The test vehicle fusion positioning method according to claim 5, characterized in that: Obtaining the UWB positioning coordinates of the test vehicle includes: At least three base stations are set up around the test vehicle. The signal transmission time of each base station is: ； Where i represents the serial number of the base station, 、 、 and They represent the time required by the i-th base station in the four signal processing stages; The distance from the i-th base station to the test vehicle is: ； Where c is the pulse velocity; Based on the coordinates of the three base stations and the coordinates of the test vehicle, calculate the distance between each base station and the test vehicle: ； in, are the coordinates of the test vehicle to be tested, represents the coordinates of the i-th base station; Solve the combined equations to find the coordinates of the test vehicle .

7. The test vehicle fusion positioning method according to claim 1, characterized in that: The obtaining of the UWB positioning parameter vector and the AI ​​visual positioning parameter vector of the test vehicle includes: Obtain the UWB positioning parameter vector and AI visual positioning parameter vector of the test vehicle as test samples according to the preset sampling time interval T5; Get m test samples with time conversion gaps, where m is a positive integer; Based on the trained positioning deep learning model, m test samples are predicted to obtain the predicted position coordinates of the m test samples, which are expressed as: 。 8. The test vehicle fusion positioning method according to claim 7, characterized in that: Calculating the difference between the predicted position coordinates of the test vehicle by the positioning deep learning model and the RTK position coordinates of the test vehicle under normal signals includes: Get the RTK position coordinates of the test vehicle at the current moment after the test vehicle is in the normal signal area , according to the same preset sampling time interval T5, the real-time speed of the test vehicle at the current moment and the angle between the test vehicle and the horizontal axis of the coordinate system , estimate the RTK position coordinates of the test vehicle at the previous moment ,in: ; ； in, is the RTK position coordinate of the test vehicle at the current time m, is the estimated RTK position coordinate of the test vehicle at the last moment m-1; According to the predicted position coordinates of the m test samples of the test vehicle by the positioning deep learning model and the estimated RTK position coordinates of the m test vehicles, the gap between the predicted position coordinates of the positioning deep learning model and the RTK position coordinates is calculated.

9. The test vehicle fusion positioning method according to claim 8, characterized in that: The calculation of the gap between the predicted position coordinates of the positioning deep learning model and the RTK position coordinates ,include: ； in, Indicates the RTK position coordinates at time i, Represents the predicted position coordinates of the test vehicle at time g output by the positioning deep learning model.

10. The test vehicle fusion positioning method according to claim 9, characterized in that: Adjusting model parameters of the positioning deep learning model based on the gap to obtain an optimized positioning deep learning model includes: The step size adjustment parameter ω of the positioning deep learning model is set, and the fusion objective function of the positioning deep learning model is updated based on the step size adjustment parameter ω: ; in, Represents the loss value of the positioning deep learning model, represents the predicted position coordinates of the test vehicle at time k by the deep learning model, represents the UWB positioning coordinates of the test vehicle at time k, represents the AI ​​visual positioning coordinates of the test vehicle at time k, and n represents the number of moments.

Images