Vehicle-road perception fusion radar self-calibration method and system for Internet of Vehicles

Abstract

The invention relates to a vehicle-road perception fusion radar self-calibration method and system oriented to the Internet of Vehicles, which are used for calibrating parameters of a roadside millimeter wave radar by using mass network-connected vehicle GPS positioning data in a road network to realize positioning calibration of the radar and are used for solving the problem of positioning deviation of the radar caused by change of an installation angle. The method comprises the following steps: (1) constructing a Bayesian network model by introducing a sensing mechanism of sensors on two sides of a vehicle and a road and a conversion mechanism of different coordinate systems, and determining a probability relation between variables; (2) probabilistic reasoning is carried out according to the Bayesian network graph; and (3) estimating key parameters by adopting an EM algorithm. According to the method, automatic calibration of the radar is realized, the road is not blocked in the calibration process, the traffic flow is not influenced, and the method has important significance in reducing the vehicle positioning deviation and improving the robustness of a road area sensing system.

Description

Technical Field

[0001] The present invention relates to the technical fields of Internet of Vehicles and radar calibration, and in particular to a vehicle-road perception fusion radar self-calibration method and system for Internet of Vehicles. Background Art

[0002] Roadside sensors are an important component of the smart road perception system. Sensors are affected by many factors, such as installation location, environmental interference, and their own performance, and there are certain deviations in vehicle positioning. Smart roads can solve the positioning deviation problem of relying on single-source sensors to a certain extent by integrating roadside and vehicle-mounted sensor data, thereby providing richer perception information. However, the vehicle-road perception fusion method is limited by the accuracy of sensor parameters, and most existing parameter calibration methods require closed roads, which still have limitations that affect daily traffic flow.

[0003] With the trend of Internet of Vehicles and vehicle-road information interaction, the method of calibrating the parameters of roadside sensors with the help of massive networked vehicle positioning data in the road network is one of the effective calibration methods. For on-board sensors, most vehicles can currently achieve their own positioning through the Global Navigation Satellite System (GNSS), which can provide a large amount of sample data for parameter calibration, and the error of vehicle GPS trajectory data is relatively stable. For roadside sensors, millimeter-wave radar has the advantages of strong penetration, low cost, no light influence, all-weather, all-day, and long-distance perception, and is a commonly used roadside sensing device for smart roads. However, due to the limitations of the radar's own performance such as resolution and generated point cloud density, as well as the influence of multipath effects, the current roadside radar still has large deviations in vehicle positioning and tracking, and it is difficult to meet the demand for high-precision positioning technology in the construction of smart road perception systems. For vehicle-road data interaction, the roadside radar and the on-board unit (OBU) transmit information to the roadside computing unit (RSU) through V2X communication technology. Due to the limited data storage and computing capabilities of the RSU, the requirements for computing resources for calibration at different data levels are also different. The current method of using high-density point cloud data to achieve high-precision radar calibration requires a large computational cost and is difficult to ensure real-time performance. In terms of calibration parameters, the installation position and angle of the radar will affect the coordinate conversion, which is the main source of radar positioning error and is also a key parameter for calibration.

[0004] To address the above problems, a radar calibration mechanism is established based on the inference of coordinate transformation mechanism with the help of a large amount of sample data from connected vehicles in the road network. This is conducive to improving the perception ability and stability of smart roads and avoiding the impact of the calibration process on traffic flow. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a vehicle-road perception fusion radar self-calibration method and system for the Internet of Vehicles, so as to solve or partially solve the problems of large positioning deviation and low calibration automation of roadside radar.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] One aspect of the present invention provides a vehicle-road perception fusion radar self-calibration method for Internet of Vehicles, comprising the following steps:

[0008] Obtaining the global coordinates of the roadside radar to be calibrated, the vehicle positioning data of the roadside radar before calibration, and the vehicle side positioning data;

[0009] Performing time alignment processing on the vehicle positioning data of the roadside radar before calibration and the vehicle side positioning data;

[0010] Taking the rotation angle parameter between the radar coordinate system and the global coordinate system as the estimation object, the probability relationship between the vehicle positioning coordinates of the roadside radar and the vehicle side positioning is established, and a Bayesian network model is constructed;

[0011] The Bayesian network model is iteratively solved using the EM algorithm to obtain converged rotation angle parameters and vehicle positioning data of the calibrated roadside radar, thereby realizing radar self-calibration.

[0012] As a preferred technical solution, the input of the Bayesian network model includes vehicle positioning data of the roadside radar, vehicle side positioning data, and the global coordinates of the roadside radar, and the output includes the rotation parameters of the calibrated roadside radar and the expected estimated value of the true value of the vehicle position.

[0013] As a preferred technical solution, the process of iteratively solving the Bayesian network model using the EM algorithm to obtain converged rotation angle parameters includes:

[0014] Initial rotation parameters;

[0015] Calculate the conditional expectation of the true value of the vehicle side positioning coordinates, bring the conditional expectation into the pre-constructed likelihood function, solve the rotation parameters with the goal of maximizing the likelihood function value, and repeat this step until the solved rotation parameters converge.

[0016] As a preferred technical solution, the conditions are expected to be:

[0017]

[0018] in, They represent the estimated x and y coordinates of the nth vehicle in the global coordinate system at the i-th moment, respectively. The x and y coordinates of the nth vehicle in the global coordinate system at the i-th moment are measured by the roadside radar. They represent the x and y coordinate values ​​of the nth vehicle in the global coordinate system at the i-th moment measured by the vehicle side, μ Rx , μ Ry Respectively represent the mean of the positioning error of the roadside radar in the x and y directions, σ Rx , σ Ry They represent the standard deviation of the positioning error of the roadside radar in the x and y directions, μ Gx , μ Gy Respectively represent the mean of the positioning error of the vehicle side in the x and y directions, σ Gx , σ Gy They represent the standard deviation of the positioning error of the vehicle side in the x and y directions respectively, and T represents the transposition.

[0019] As a preferred technical solution, the rotation parameters are solved by gradient descent or interior point method.

[0020] As a preferred technical solution, the condition for the convergence of the solved rotation parameters is that the number of iterations reaches a preset value, or the difference between the rotation parameters of two adjacent iterations is less than a preset value.

[0021] As a preferred technical solution, the likelihood function is:

[0022]

[0023] Among them, α R is the rotation parameter, p i,j represents the true value of the vehicle side positioning coordinate of the jth vehicle at the i-th time, t and n represent the total number of time and the total number of vehicles respectively, and Σ represents the The diagonal matrix composed of R Represents the position coordinate vector of the roadside radar in the global coordinate system, represents the position coordinate vector of the vehicle in the roadside radar coordinate system, Represents the vehicle position coordinate vector measured on the vehicle side, μ R =(μ Rx ,μ Ry ) T ,μ G =(μ Gx ,μ Gy ) T , μ Rx , μ Ry Respectively represent the mean of the positioning error of the roadside radar in the x and y directions, σ Rx , σ Ry They represent the standard deviation of the positioning error of the roadside radar in the x and y directions, μ Gx , μGy Respectively represent the mean of the positioning error of the vehicle side in the x and y directions, σ Gx , σ Gy They represent the standard deviation of the positioning error of the vehicle side in the x and y directions respectively, and T represents the transposition.

[0024] As a preferred technical solution, time alignment processing is achieved through linear interpolation.

[0025] Another aspect of the present invention provides a vehicle-road perception fusion radar self-calibration system for vehicle networking, characterized in that the system is used to implement the aforementioned vehicle-road perception fusion radar self-calibration method for vehicle networking, and the system includes:

[0026] Roadside radar to be calibrated;

[0027] Vehicle side positioning module;

[0028] A data acquisition module, used to obtain the global coordinates of the roadside radar to be calibrated, the vehicle positioning data of the roadside radar before calibration, and the vehicle side positioning data;

[0029] A data processing module, used for performing time alignment processing on the vehicle positioning data of the roadside radar before calibration and the vehicle side positioning data;

[0030] The rotation angle parameter solving module is used to estimate the rotation angle parameter between the radar coordinate system and the global coordinate system, establish the probability relationship between the vehicle positioning coordinates of the roadside radar and the vehicle side positioning, construct a Bayesian network model, and use the EM algorithm to iteratively solve the Bayesian network model to obtain the converged rotation angle parameter and the calibrated vehicle positioning data of the roadside radar to achieve radar self-calibration.

[0031] As a preferred technical solution, the roadside radar is a millimeter wave radar, and the data acquisition module includes a GPS positioning unit.

[0032] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0033] (1) High degree of automation: The present invention can realize the automation of calibration. The calibration process does not require the closure of roads and does not affect traffic flow. When the installation position and angle of the roadside radar change, resulting in the inability to accurately locate the vehicle, the radar can automatically calibrate the parameters and restore the positioning capability. In addition, the calibration process does not rely on road environment information and will not be affected by changes in weather conditions.

[0034] (2) High stability and reliability: The present invention combines the information of roadside and vehicle-mounted sensors, and uses the vehicle-mounted GPS as an additional information source to supplement the information loss of the roadside radar. Multi-source perception at both ends of the road and the vehicle can further enhance the stability and reliability of calibration. Compared with the single-source perception method, the present invention has stronger calibration capabilities. When the traffic volume on the road is large, more sample data can be provided for the calibration process, so the calibration effect is also better. Using the advantages of heterogeneous sensors to compensate for the calibration can enhance the perception ability of the overall perception system.

[0035] (3) Low consumption of computing resources: The computation process of the present invention is relatively simple, the calibration cost is low, and excessive computing resources are not required. The present invention does not calibrate from the original data layer, but implements parameter calibration from the decision level, that is, it does not consider the signal processing process inside the radar and GPS sensors, simplifies the calibration steps, can quickly implement the calibration method, and also reduces the transmission pressure of high-density point cloud data for vehicle-to-road communication. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flowchart of a vehicle-road perception fusion radar self-calibration method for Internet of Vehicles in an embodiment;

[0037] Figure 2 This is a diagram of a Bayesian network model in a bicycle scenario in an embodiment;

[0038] Figure 3 This is a diagram of the Bayesian network model in a multi-vehicle scenario;

[0039] Figure 4 It is a specific implementation scenario example diagram;

[0040] Figure 5 Schematic diagram of a vehicle-road perception fusion radar self-calibration system for Internet of Vehicles in an embodiment;

[0041] Figure 6 Schematic diagram of an electronic device in an embodiment. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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 creative work should fall within the scope of protection of the present invention.

[0043] Example 1

[0044] In response to the problems existing in the aforementioned prior art, the present embodiment provides a self-calibration method for vehicle-road perception fusion radar for the Internet of Vehicles, which realizes self-calibration of coordinate conversion parameters of roadside radars of types such as millimeter waves, can calibrate radar parameters more accurately, and the calibration process does not require road closure. By utilizing a large amount of networked vehicle positioning data on traffic roads, the problems of large positioning deviation of roadside radars and low calibration automation can be solved.

[0045] See also Figure 1 In this method, the sensors on the vehicle side and the road side first collect data, and then the conversion mechanism between the sensor coordinate systems determines the topological structure of the Bayesian network graph. The EM algorithm is used to estimate the parameters of the latent variables in the probability graph. When the parameter estimation is completed, the trajectory positioning calibration of the entire vehicle is also completed accordingly. Specifically, this method includes the following steps:

[0046] S1, data collection. Collect the global coordinates of the roadside millimeter wave radar, the vehicle positioning data of the radar before calibration, and the GPS data of the connected car (i.e., vehicle side positioning).

[0047] S2, Bayesian network model construction. The collected radar data and connected vehicle GPS data are time-aligned using the linear interpolation method.

[0048] like Figure 2 As shown in the figure, the shaded variables represent hidden variables, and the unshaded variables represent explicit variables, that is, except for α R , β R , γ R 、p i,n The key parameters of the radar considering the three-dimensional coordinates are α R , β R , γ R Three, for two-dimensional coordinates only α R parameter. It indicates the polar coordinates of the jth vehicle measured by the radar at the i-th time, including the distance and angle. R is the position coordinate of the radar in the global coordinate system. In the Bayesian network diagram constructed in this step, it is assumed that the positioning errors of GPS and radar obey the normal distribution, and the positioning results of the two are conditionally independent. The following is a further explanation of the probability diagram.

[0049] For the radar part on the left, the radar needs to perform two coordinate transformations after measurement: from polar coordinates to the radar local coordinate system, and then to the global coordinate system. The first coordinate transformation does not require parameters, and the second transformation is a rigid transformation between rectangular coordinate systems. The transformation process can be decomposed into two parts: rotation and translation. The second transformation requires the use of parameter α RFor the GPS part on the right, it is modeled as GPS error obeying normal distribution, and the radar parameters are calibrated based on GPS positioning.

[0050] For multi-vehicle scenarios, such as Figure 3 As shown in the figure, the constructed Bayesian network model is similar to the single-vehicle scenario. The relationship between the coordinate variables of each vehicle in the probabilistic graphical model is the same as that in the single-vehicle scenario, and the radar error and GPS error parameters of all vehicles are the same.

[0051] S3, parameter estimation. Based on the established Bayesian network model, the EM algorithm is used to iteratively estimate radar parameters.

[0052] This step builds a probability graph model and performs probability reasoning based on the analysis of the distribution characteristics of radar positioning deviation and positioning mechanism. The Bayesian network model includes the rotation angle variable between the radar coordinate system and the global coordinate system. This parameter is a hidden variable that needs to be estimated and cannot be directly observed. In this Bayesian network model, the mechanism of radar and coordinate system conversion is used to establish the probability relationship between radar positioning coordinates and GPS positioning coordinate variables.

[0053] Specifically, for the Bayesian network model with incomplete data, the EM algorithm is used to estimate the parameter α R The EM algorithm consists of two parts: the E step and the M step.

[0054] Step E: Randomly initialize parameter α R , and then calculate the conditional expectation of the vehicle position coordinates. According to the probability graph, the calculation method is as follows:

[0055]

[0056] The f function is the probability density function. Taking the calculation of the x-axis coordinate as an example (the same applies to the y-axis coordinate), it can be calculated through Bayesian theorem and probability derivation:

[0057]

[0058] Where f((x R i,n ,x G i,n )|x i,n ) is the probability density function of the normal distribution, f(x i,n ) is the probability density function of the vehicle position coordinate distribution, which is assumed to obey uniform distribution in the model, so f(x i,n) is a specific constant, which is the prior probability. Since the density function of the normal distribution is simple, it is easy to derive and solve the subsequent likelihood function. After multiplying the probability density functions of multiple normal distributions, they can be converted into a function form similar to the normal distribution density function through the matching method. The final calculation method is:

[0059]

[0060] Step M: Substitute the expected vehicle position condition calculated in the previous step into the likelihood function of the probability graph, maximize the likelihood function, and solve for the parameter α R Repeat the E and M steps and iterate until the parameter α R convergence.

[0061] For multi-vehicle scenarios, during the specific calibration process, the E step of the single-vehicle and multi-vehicle scenarios is the same, and the likelihood function of the M step is changed. The multi-vehicle scenario needs to calculate the probability product of each vehicle at each moment. The specific form is as follows. In the formula, the subscript i represents the time sequence number, and j represents the vehicle sequence number.

[0062]

[0063] For the optimization method of the likelihood function, gradient descent or interior point method can be used to solve it, ensuring that the optimal value of the likelihood function can be quickly found in one iteration of the EM algorithm, so as not to affect the overall calculation time too long. For the convergence judgment of the EM algorithm, the maximum number of iterations or parameter convergence criteria are set to determine whether the algorithm is terminated. When the number of iterations reaches 100, or the parameter α of two adjacent iterations is R When the difference is less than 0.01, the algorithm stops and the vehicle position condition expectation calculated in the last E step is used as the calibrated vehicle coordinates.

[0064] Preferably, step S3 includes steps S301-S305.

[0065] S301, randomly initialize the radar coordinate rotation parameter α in the range [0,2π] R .

[0066] S302, calculating the conditional expectation of the true value of the vehicle positioning coordinates, the specific calculation formula is as follows:

[0067]

[0068] Where: The estimated x and y coordinates of the nth vehicle in the global coordinate system at the i-th moment; The millimeter-wave radar measures the x and y coordinates of the nth vehicle in the global coordinate system at the i-th moment; GPS measures the x and y coordinates of the nth vehicle in the global coordinate system at the i-th moment; μ Rx , μ Ry : The mean value of the positioning error of the millimeter wave radar in the x and y directions; σ Rx , σ Ry : standard deviation of the positioning error of the millimeter wave radar in the x and y directions; μ Gx , μ Gy : the mean of GPS positioning errors in the x and y directions; σ Gx , σ Gy : standard deviation of GPS positioning error in x and y directions;

[0069] S303, substituting the calculated vehicle position condition expectation into the likelihood function, the specific likelihood function is:

[0070]

[0071] In the formula, p i,j : The conditional expectation vector of the true value of the jth vehicle coordinate at the i-th moment; Σ: The covariance matrix, which is composed of The diagonal matrix composed of i,j The calculation of is as follows:

[0072]

[0073] In the formula, p R : The position coordinate vector of the millimeter wave radar in the global coordinate system; The position coordinate vector of the vehicle in the millimeter wave radar coordinate system; Vehicle position coordinate vector measured by GPS; μ R =(μ Rx ,μ Ry ) T ,μ G =(μ Gx ,μ Gy ) T ; M: rotation matrix; and M are calculated as follows:

[0074]

[0075] In the formula, α R : The rotation angle parameter between the millimeter wave radar coordinate system and the global coordinate system. is the straight-line distance between the vehicle and the radar, is the azimuth of the vehicle in the radar polar coordinate system.

[0076] S304, maximize the likelihood function to obtain α R , the optimization method adopted is the interior point method.

[0077] S305, return to step S302, and iterate until the parameter α R Convergence. When converged, the radar parameters have been calibrated, and the vehicle position coordinates calculated in the last iteration are used as the calibration result to achieve the calibration of the vehicle's global positioning coordinates.

[0078] The following is a detailed description of the specific implementation scenarios. Figure 4 As shown in the figure, the roadside radar is installed on the road gantry or crossbar, and the radar detection range faces the direction of vehicle travel. The internal transmitter of the radar transmits electromagnetic waves, which will be reflected back after encountering the target object. The radar receiver receives the reflected signal and calculates the distance by measuring the frequency difference between the transmitted and received signals. The operating frequency allowed by the millimeter-wave radar for road traffic is generally 77GHz, and the maximum detection distance is generally 300m. For GPS, it uses the interaction between the satellite transmission signal and the vehicle receiver to achieve positioning. After the vehicle-mounted GPS receiver completes the processing of the received signal, it transmits the positioning data to the roadside calculation unit RSU through V2X communication technology. Similarly, the radar also transmits the positioning result to the RSU. Both the roadside radar and the RSU can be installed together on the road crossbar. Therefore, the calibration calculation process is finally completed by the RSU without closing the road, avoiding the defects of the time-consuming and labor-intensive traditional manual calibration method.

[0079] The following is a further explanation of the results evaluation of specific cases. Figure 4 As shown in the figure, the radar and RSU are installed on a unidirectional road crossbar, the lane width is 3.5m wide, and the maximum detection distance of the radar is 300m. The detection direction faces the direction of traffic. The red in the figure is the local coordinate system of the radar, and the blue is the global coordinate system. The example in the figure shows that the rotation angle between the two two-dimensional coordinate systems is 180°. Therefore, 180° is the parameter α R The true value of needs to be estimated. After adopting the calibration model proposed in this method, the model's calculation program needs to be deployed on the RSU without taking road closure measures. The final calibration result can be obtained by collecting real trajectory data through the connected vehicle, comparing it with the calibrated trajectory positioning result, and evaluating it by taking the maximum error, percentage error and root mean square error.

[0080] Example 2

[0081] Based on Example 1, this embodiment provides a vehicle-road perception fusion radar self-calibration system for the Internet of Vehicles, which is used to implement the aforementioned vehicle-road perception fusion radar self-calibration method for the Internet of Vehicles, see Figure 5 , the system includes:

[0082] (1) Roadside radar to be calibrated.

[0083] (2) Vehicle side positioning module.

[0084] (3) A data acquisition module, which is used to obtain the global coordinates of the roadside radar to be calibrated, the vehicle positioning data of the roadside radar before calibration, and the vehicle side positioning data.

[0085] (4) A data processing module, used to perform time alignment processing on the vehicle positioning data of the roadside radar before calibration and the vehicle side positioning data.

[0086] (5) A rotation angle parameter solving module is used to estimate the rotation angle parameter between the radar coordinate system and the global coordinate system, establish the probabilistic relationship between the vehicle positioning coordinates of the roadside radar and the vehicle side positioning, construct a Bayesian network model, and use the EM algorithm to iteratively solve the Bayesian network model to obtain the converged rotation angle parameter and the calibrated vehicle positioning data of the roadside radar, thereby realizing radar self-calibration.

[0087] In summary, the present invention has the following characteristics:

[0088] (1) Automatic calibration is achieved. The calibration process does not require road closures and does not affect traffic flow. When the installation position and angle of the roadside radar change, resulting in the inability to accurately locate the vehicle, the radar can automatically calibrate the parameters and restore the positioning capability. In addition, the calibration process does not rely on road environment information and will not be affected by changes in weather conditions.

[0089] (2) Combining the information from the roadside and vehicle-mounted sensors, using the vehicle-mounted GPS as an additional information source to supplement the information loss of the roadside radar, the multi-source perception at both ends of the road and the vehicle can further enhance the stability and reliability of the calibration. Compared with the single-source perception method, the present invention has a stronger calibration capability. When the traffic volume on the road is large, more sample data can be provided for the calibration process, so the calibration effect is also better. Using the advantages of heterogeneous sensors to compensate for the calibration can enhance the perception capability of the overall perception system.

[0090] (3) The calculation process is relatively simple, the calibration cost is low, and excessive computing resources are not required. The present invention does not calibrate from the original data layer, but implements parameter calibration from the decision level, that is, it does not consider the signal processing process inside the radar and GPS sensors, simplifies the calibration steps, can quickly implement the calibration method, and also reduces the transmission pressure of high-density point cloud data for vehicle-road communication.

[0091] Example 3

[0092] This embodiment provides an electronic device, including: one or more processors and a memory, wherein the memory stores one or more programs, and the one or more programs include instructions for executing the vehicle-road perception fusion radar self-calibration method for the Internet of Vehicles as described in Example 1.

[0093] like Figure 6 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 The vehicle-road perception fusion radar self-calibration method for the Internet of Vehicles. Of course, in addition to the software implementation, the present invention does not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0094] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0095] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0096] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A vehicle-road perception fusion radar self-calibration method for Internet of Vehicles, characterized in that: The steps include: Obtaining the global coordinates of the roadside radar to be calibrated, the vehicle positioning data of the roadside radar before calibration, and the vehicle side positioning data; Performing time alignment processing on the vehicle positioning data of the roadside radar before calibration and the vehicle side positioning data; Taking the rotation angle parameter between the radar coordinate system and the global coordinate system as the estimation object, the probability relationship between the vehicle positioning coordinates of the roadside radar and the vehicle side positioning is established, and a Bayesian network model is constructed; The Bayesian network model is iteratively solved using the EM algorithm to obtain converged rotation angle parameters and vehicle positioning data of the calibrated roadside radar, thereby realizing radar self-calibration.

2. The vehicle-road perception fusion radar self-calibration method for Internet of Vehicles according to claim 1 is characterized in that: The input of the Bayesian network model includes vehicle positioning data of the roadside radar, vehicle side positioning data, and global coordinates of the roadside radar, and the output includes the rotation parameters of the calibrated roadside radar and the expected estimated value of the true value of the vehicle position.

3. The vehicle-road perception fusion radar self-calibration method for Internet of Vehicles according to claim 1 is characterized in that: The process of iteratively solving the Bayesian network model using the EM algorithm to obtain converged rotation angle parameters includes: Initial rotation parameters; Calculate the conditional expectation of the true value of the vehicle side positioning coordinates, bring the conditional expectation into the pre-constructed likelihood function, solve the rotation parameters with the goal of maximizing the likelihood function value, and repeat this step until the solved rotation parameters converge.

4. The vehicle-road perception fusion radar self-calibration method for Internet of Vehicles according to claim 3 is characterized in that: The conditional expectation is: in, They represent the estimated x and y coordinates of the nth vehicle in the global coordinate system at the i-th moment, respectively. The x and y coordinates of the nth vehicle in the global coordinate system at the i-th moment are measured by the roadside radar. They represent the x and y coordinate values ​​of the nth vehicle in the global coordinate system at the i-th moment measured by the vehicle side, μ Rx , μ Ry Respectively represent the mean of the positioning error of the roadside radar in the x and y directions, σ Rx , σ Ry They represent the standard deviation of the positioning error of the roadside radar in the x and y directions, μ Gx , μ Gy Respectively represent the mean of the positioning error of the vehicle side in the x and y directions, σ Gx , σ Gy They represent the standard deviation of the positioning error of the vehicle side in the x and y directions respectively, and T represents the transposition.

5. The vehicle-road perception fusion radar self-calibration method for Internet of Vehicles according to claim 3 is characterized in that: The rotation parameters are solved by gradient descent or interior point method.

6. The vehicle-road perception fusion radar self-calibration method for Internet of Vehicles according to claim 3 is characterized in that: The condition for the rotation parameter to be solved to converge is that the number of iterations reaches a preset value, or the difference between the rotation parameters of two adjacent iterations is less than a preset value.

7. The vehicle-road perception fusion radar self-calibration method for Internet of Vehicles according to claim 3 is characterized in that: The likelihood function is: Among them, α R is the rotation parameter, p i,j represents the true value of the vehicle side positioning coordinate of the jth vehicle at the i-th time, t and n represent the total number of time and the total number of vehicles respectively, and Σ represents the The diagonal matrix composed of R Represents the position coordinate vector of the roadside radar in the global coordinate system, represents the position coordinate vector of the vehicle in the roadside radar coordinate system, Represents the vehicle position coordinate vector measured on the vehicle side, μ R =(μ Rx ,μ Ry ) T ,μ G =(μ Gx ,μ Gy ) T , μ Rx , μ Ry Respectively represent the mean of the positioning error of the roadside radar in the x and y directions, σ Rx , σ Ry They represent the standard deviation of the positioning error of the roadside radar in the x and y directions, μ Gx , μ Gy Respectively represent the mean of the positioning error of the vehicle side in the x and y directions, σ Gx , σ Gy They represent the standard deviation of the positioning error of the vehicle side in the x and y directions respectively, and T represents the transposition.

8. The vehicle-road perception fusion radar self-calibration method for Internet of Vehicles according to claim 1 is characterized in that: Time alignment is achieved through linear interpolation.

9. A vehicle-road perception fusion radar self-calibration system for Internet of Vehicles, characterized in that: The system is used to implement the vehicle-road perception fusion radar self-calibration method for the Internet of Vehicles as described in any one of claims 1 to 8, and the system includes: Roadside radar to be calibrated; Vehicle side positioning module; A data acquisition module, used to obtain the global coordinates of the roadside radar to be calibrated, the vehicle positioning data of the roadside radar before calibration, and the vehicle side positioning data; A data processing module, used for performing time alignment processing on the vehicle positioning data of the roadside radar before calibration and the vehicle side positioning data; The rotation angle parameter solving module is used to estimate the rotation angle parameter between the radar coordinate system and the global coordinate system, establish the probability relationship between the vehicle positioning coordinates of the roadside radar and the vehicle side positioning, construct a Bayesian network model, and use the EM algorithm to iteratively solve the Bayesian network model to obtain the converged rotation angle parameter and the calibrated vehicle positioning data of the roadside radar to achieve radar self-calibration.

10. The vehicle-road perception fusion radar self-calibration system for Internet of Vehicles according to claim 9, characterized in that: The roadside radar is a millimeter wave radar, and the data acquisition module includes a GPS positioning unit.

Images