Distributed two-stage roadside multi-millimeter-wave radar cooperative calibration method

Abstract

The invention belongs to the field of highway roadside sensor calibration, and provides a distributed two-stage roadside multi-millimeter-wave radar collaborative calibration method. Comprising the steps of construction of an engineering coordinate system, installation of radar equipment, construction of a radar calibration equation, fitting of a road estimation curve, optimization of rotation parameters, determination of matched trajectory data, fitting of an X-direction motion equation and a Y-direction motion equation, optimization of stretching parameters, updating of the radar calibration equation and periodic online collaborative calibration. According to the method, the calibration precision of the radar is improved by constructing the radar calibration equation and optimizing the rotation parameters and the stretching parameters, and the method has high adaptability and flexibility, can be integrated into an existing intelligent traffic system and is compatible with different radar models and configurations, and existing infrastructures do not need to be transformed on a large scale.

Description

Technical Field

[0001] The present invention relates to the field of highway roadside sensor calibration, and in particular to a distributed two-stage roadside multi-millimeter wave radar collaborative calibration method. Background Art

[0002] Installing continuous millimeter-wave radars on the side of highways to extract vehicle motion trajectory data is a key technology for providing a digital twin digital base in smart highways. The calibration of millimeter-wave radars refers to converting the collected raw data into the required coordinate system, which is a necessary prerequisite for the use of radar data. Large-scale radar groups involve the calibration of multiple millimeter-wave radars. The calibration results must not only make the calibration results of each radar reasonable, but also improve the coordination between the calibration results of different radars as much as possible.

[0003] There are two main solutions for the calibration of millimeter-wave radars on highways. One is based on static target calibration, which requires the setting of fixed reflectors for calibration. It is dangerous for highway environments and has low efficiency. The other is based on the comparison of other data, such as aligning high-precision trajectory data from other data sources with trajectory data collected synchronously by millimeter-wave radars to calibrate radars. Although this type of dynamic calibration method greatly improves safety compared to traditional static calibration methods and does not rely on closed roads, it is currently mainly used for the calibration of single radars. Using it in multiple radars may also require a lot of organizational costs. When multiple radars work together, a slight rotation of the radar may cause large fluctuations in trajectory data. In highway environments, sensors will inevitably fluctuate in position due to vibration, wind, or aging of facilities, and repeated calibration work is cumbersome. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a distributed two-stage roadside multi-millimeter-wave radar collaborative calibration method to solve the problems of insufficient utilization of highway alignment and vehicle motion consistency information between radars and low degree of automation in the prior art.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A distributed two-stage roadside multi-millimeter wave radar collaborative calibration method, comprising:

[0007] Construct the engineering coordinate system and road line shape discretization representation based on the pre-collected horizontal curve data of the target road section, and use the RTK measurement equipment to align the radar equipment to the target installation position;

[0008] Construct radar calibration equations;

[0009] Using the radar device to collect first trajectory data of the vehicle on the target road section, and using RANSAC robust regression to fit the first trajectory data to obtain a road estimation curve;

[0010] The road estimation curve is converted into an engineering estimation curve by using the rotation parameters of the radar calibration equation, and the rotation parameters of the radar calibration equation are determined by using the Nelder-Mead optimization algorithm when the variance of the engineering estimation curve and the road linear discretization representation is minimized, so as to obtain the optimized rotation parameters;

[0011] In the overlapping area of ​​the adjacent radar devices, two sets of second trajectory data of the same vehicle are identified and matched, and the two sets of second trajectory data are matched by using a distance sorting method to obtain matching trajectory data;

[0012] Extracting motion features of the second trajectory data in the X and Y directions, and fitting the motion features with a time-based cubic polynomial function using a least squares method to obtain an X-direction motion equation and a Y-direction motion equation;

[0013] Calculate the difference between the observed value of the X-direction motion equation and the Y-direction motion equation and the fitting value of the matching trajectory data at the same time to obtain the X-direction residual and the Y-direction residual, input the X-direction residual and the Y-direction residual into the optimization target, and use the Nelder-Mead optimization algorithm to determine the stretching parameter of the radar calibration equation when the output result of the optimization target is minimized, and obtain the optimized stretching parameter;

[0014] Updating the optimized rotation parameter and the optimized stretching parameter to the radar calibration equation, and calibrating the vehicle on the target road section using the updated radar calibration equation;

[0015] The step of "collecting the first trajectory data of the vehicle on the target road section by using the radar device, and fitting the first trajectory data by using RANSAC robust regression to obtain a road estimation curve" is returned to periodically at a fixed time to perform periodic online collaborative calibration on the radar calibration equation.

[0016] Preferably, the flat curve data includes: straight line parameters, curve parameters and corner parameters.

[0017] Preferably, the radar calibration equation is: B=AH;

[0018] in, B is the coordinate point after conversion; A is the original coordinate point collected by the radar device; H is the conversion matrix; θ is the rotation parameter; α is the stretching parameter; c 1 、c 2are respectively the coordinates of the radar device in the X and Y directions in the engineering coordinate system.

[0019] Preferably, the expression of the optimization objective is:

[0020]

[0021] Among them, γ and β are the error weights in the X and Y directions respectively; e x 、e y are the X-direction residual and the Y-direction residual respectively; n is the number of data points; i is e x and e y The corresponding point number.

[0022] Preferably, the distance between adjacent pile numbers in the engineering coordinate system is less than or equal to 2m.

[0023] Preferably, the first trajectory data includes: a time field, a horizontal coordinate field, a vertical coordinate field and a speed field.

[0024] Preferably, the road estimation curve, the X-direction motion equation and the Y-direction motion equation are all three-variable cubic equations.

[0025] Preferably, the fixed time is 30 minutes.

[0026] Preferably, the installation distance between adjacent radar devices is 350 m; and the length of the overlapping area is 100 m.

[0027] Preferably, the error between the actual installation position of the radar device and the target installation position is less than 1 cm.

[0028] The present invention discloses the following technical effects:

[0029] The present invention provides a distributed two-stage roadside multi-millimeter-wave radar collaborative calibration method. By constructing radar calibration equations and optimizing rotation parameters and stretching parameters, the present invention solves the problems of insufficient utilization of highway alignment and vehicle motion consistency information between radars and low degree of automation in the prior art, and realizes adaptive calibration that quickly determines monitored road conditions and trajectory data. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0031] Figure 1A schematic diagram of a distributed two-stage roadside multi-millimeter-wave radar collaborative calibration process provided by an embodiment of the present invention;

[0032] Figure 2 Illustration of the distributed deployment scenario and overlapping area of ​​millimeter wave mines provided in an embodiment of the present invention;

[0033] Figure 3 A technical diagram provided for an embodiment of the present invention;

[0034] Figure 4 A schematic diagram of a calibration process provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] 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 only 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 are within the scope of protection of the present invention.

[0036] The purpose of the present invention is to provide a distributed two-stage roadside multi-millimeter-wave radar collaborative calibration method to solve the problems of insufficient utilization of highway alignment and vehicle motion consistency information between radars and low degree of automation in the prior art.

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] Figure 1 A schematic diagram of a distributed two-stage roadside multi-millimeter wave radar collaborative calibration process provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the present invention provides a distributed two-stage roadside multi-millimeter wave radar collaborative calibration method, comprising:

[0039] Step 100: construct an engineering coordinate system and a road line discretization representation based on the pre-collected horizontal curve data of the target road section, and align the radar device to the target installation position using an RTK measurement device;

[0040] Step 200: construct radar calibration equation;

[0041] Step 300: using a radar device to collect first trajectory data of a vehicle on a target road section, and using RANSAC robust regression to fit the first trajectory data to obtain a road estimation curve;

[0042] Step 400: using the rotation parameters of the radar calibration equation to convert the road estimation curve into the engineering estimation curve, and using the Nelder-Mead optimization algorithm to determine the rotation parameters of the radar calibration equation when the variance of the engineering estimation curve and the road linear discretization representation is minimized, and obtain the optimized rotation parameters;

[0043] Step 500: in the overlapping area of ​​adjacent radar devices, two sets of second trajectory data of the same vehicle are identified and matched, and the two sets of second trajectory data are matched using a distance sorting method to obtain matching trajectory data;

[0044] Step 600: extracting motion features of the second trajectory data in the X and Y directions, and fitting the motion features with a time-based cubic polynomial function using the least squares method to obtain an X-direction motion equation and a Y-direction motion equation;

[0045] Step 700: Calculate the difference between the observed value of the X-direction motion equation and the Y-direction motion equation and the fitted value of the matching trajectory data at the same time to obtain the X-direction residual and the Y-direction residual, input the X-direction residual and the Y-direction residual into the optimization target, and use the Nelder-Mead optimization algorithm to determine the stretching parameter of the radar calibration equation when the output result of the optimization target is minimized, and obtain the optimized stretching parameter;

[0046] Step 800: updating the optimized rotation parameters and the optimized stretching parameters to the radar calibration equation, and calibrating the vehicles on the target road section using the updated radar calibration equation;

[0047] Step 900: Return to step “collecting first trajectory data of vehicles on the target road section using radar equipment, and fitting the first trajectory data using RANSAC robust regression to obtain a road estimation curve” with a fixed time as a cycle to perform periodic online collaborative calibration on the radar calibration equation.

[0048] Specifically, the flat curve data includes: straight line parameters, curve parameters and corner parameters.

[0049] Furthermore, the radar calibration equation is: B = AH;

[0050] in, B is the coordinate point after conversion; A is the original coordinate point collected by the radar device; H is the conversion matrix; θ is the rotation parameter; α is the stretching parameter; c 1 、c 2 They are respectively the X and Y coordinates of the radar device in the engineering coordinate system.

[0051] Specifically, the expression of the optimization objective is:

[0052]

[0053] Among them, γ and β are the error weights in the X and Y directions respectively; e x 、e y are the X-direction residual and the Y-direction residual respectively; n is the number of data points; i is e x and e y The corresponding point number.

[0054] Preferably, the distance between adjacent pile numbers in the engineering coordinate system is less than or equal to 2m.

[0055] Specifically, the first trajectory data includes: a time field, a horizontal coordinate field, a vertical coordinate field and a speed field.

[0056] Preferably, the road estimation curve, the X-direction motion equation and the Y-direction motion equation are all three-variable cubic equations.

[0057] Optionally, the fixed time is 30 minutes.

[0058] Specifically, the installation distance between adjacent radar devices is 350m; the length of the overlapping area is 100m.

[0059] Preferably, the error between the actual installation position of the radar device and the target installation position is less than 1 cm.

[0060] Furthermore, a relatively simple distance sorting is used for matching the second trajectory data. When the distance between the two vehicles is detected to be less than 5m for five consecutive timestamps, it is determined that the match is successful.

[0061] Specifically, using the horizontal curve data of the highway, the straight line, curve, corner and other parameters, the pile coordinates of the road centerline can be solved to obtain the discrete representation of the road. The corresponding pile coordinate system point is {(k 1 ,b 1 ),(k 2 ,b 2 ),…,(k m ,b m )}, where the spacing between each pile number is generally no more than 2m, and the corresponding engineering coordinate G road :{(x g1 ,y g1 ),(x g2 ,y g2 )…(x gm ,y gm )}, which is used to represent the road linear reference line in the detection area where the radar is located. Where m is the number of discrete points on the road; k m 、b m are the pile number position and the lateral position of the road section of the corresponding point respectively; gm ,y gm are the horizontal and vertical positions of the corresponding points in the engineering coordinate system.

[0062] Furthermore, the calibration equation is initialized by placing the RTK device under the installed radar device to collect the longitude and latitude of the corresponding radar device. At this time, the longitude and latitude coordinate system positions of a series of radar installations can be recorded as {(lat r1 ,long r1 ),(lat r2 ,long r2 )…(lat rd ,long rd )}, and convert the coordinates into the corresponding radar position in the engineering coordinate system {(x r1 ,y r1 ),(x r2 ,y r2 )…(x rd ,y rd )}. The corresponding pile number coordinate system can be read from the drawing or on-site as {(k r1 ,b r1 ),(k r1 ,b r1 )…(k rd ,b rd )}. Among them, lat rd 、long rd They represent the longitude and latitude positions of the radar in the latitude and longitude coordinate system respectively; rd ,y rd is the horizontal and vertical position of the corresponding point in the engineering coordinate system; k rd , b rd is the pile number position and the lateral position of the road section of the corresponding point in the pile number coordinate system, r represents the radar, and d represents the radar number.

[0063] Specifically, the calibration of the radar involves the coordinate system transformation. The original coordinate point recorded in the radar is A, which is [X A ,Y A ,1], where X A , Y A They represent all the initial positions of the radar's first trajectory data in the engineering coordinate system, 1 is the offset, and to facilitate the calculation of the displacement in the calibration relationship, the required coordinate point after conversion is B and is recorded as [X B ,Y B ], X B , Y BThey represent the calibrated position of the radar's first trajectory data in the engineering coordinate system. The calibration process of the millimeter-wave radar refers to solving a transformation matrix H so that B = AH. In order to express the calibration equation, the radar calibration process is defined as the change in two dimensions of rotation and stretching. The core parameters are converted into two estimated parameters θ and α. The transformation matrix can be expressed as follows:

[0064]

[0065] Furthermore, the radar collects and accumulates raw trajectory data for a period of time, which contains the trajectory data of the vehicle in the radar coordinate system, denoted as Tra rd , which contains {T rd ,X rd ,Y rd ,V rd …}, representing the time T of the detected vehicle rd , horizontal and vertical position and speed data X rd ,Y rd , V rd , where r indicates radar data and d represents the radar number. The data collected by the radar are all in the radar's own coordinate system, and the radar's own coordinate system is converted to the corresponding position in the engineering coordinate system.

[0066] Preferably, the trajectory data is fitted with a corresponding cubic curve y=ax using RANSAC 3 +bx 2 +cx+d, there are four parameters, a, b, c, d; RANSAC can effectively reduce the impact of outliers in the preliminary data on the results. This fitting process can be expressed as in represents the above fitting process, i represents the trajectory data set of the i-th radar, and θ is the parameter to be optimized. The fitted curve is represented by the estimated road geometry curve feature point G est :{(x e1 ,y e1 ),(x e2 ,y e2 )…(x em ,y em )}where x em ,y em The corresponding road feature points after fitting correspond to the road feature points discretized in the previous step.

[0067] Furthermore, the fitting curve G est It can represent the movement trend of the vehicle along the road line, which should be consistent with the line of the highway, that is, the curve and the pile-by-pile coordinates G obtained by S1 road It should be consistent.

[0068] Furthermore, the optimization equation of the first stage is expressed as minVar(G est -G road ), optimize and solve to get θ', Var represents the variance of the two.

[0069] Specifically, in the overlapping area of ​​two radars, the trajectory data of the same vehicle detected by different radars are identified and matched, and the matched trajectory data is Tra match , including the vehicle position X match , Y match and speed V match , time T match Extract the motion characteristics of the vehicle in the X and Y directions, and use the cubic polynomial function of time t to fit the motion trajectory of the vehicle in the X and Y directions: P y =a 3 t 3 +a 2 t 2 +a 1 t+a 0 and P x =b 3 t 3 +b 2 t 2 +b 1 t+b 0 , where a i and b i (i=0, 1, 2, 3) are coefficients of the polynomial, and the coefficients are determined using the least squares method.

[0070] Furthermore, the residual error in the fitting process is calculated, that is, the difference between the observed value and the fitted value: x =P x -x match , e y =P y -y match , where x match ,y match are the vehicle positions in the X and Y directions in the matched trajectory data, respectively. An optimization goal of minimizing the sum of squares of the residuals is established to ensure that the motion state of the same vehicle detected by the two radars is consistent in the overlapping area:

[0071]

[0072] By optimizing the above objective function, the longitudinal scaling parameter α of the radar is determined to further compensate for the errors caused by manufacturing tolerance or installation error, so that the motion patterns of adjacent radar vehicles are consistent and the radars have stronger collaborative working capabilities.

[0073] Preferably, for all radars, two radars facing each other are taken as research objects, and optimization parameters are solved every 30 minutes. The above steps are repeated to complete the calibration of all radars, and the calibration of all radars is completed regularly.

[0074] Further, the following is described with reference to the specific drawings:

[0075] refer to Figure 2 In this embodiment, multiple distributed radars are arranged and the overlap of the sensing range is shown in the figure. A pair of radars are selected to be installed about 350m apart from each other, and the length of the overlapping area is about 100m. The implementation scenario is a general section of a highway.

[0076] S1: In this embodiment, the precise installation position of the radar must be determined first, so that the data detected by the radar can be accurately matched with the actual road geometry in the subsequent calibration process. To this end, this embodiment uses the flat curve data of the highway, which includes parameters such as straight lines, curves, and corners, which are key information for describing the coordinates of each pile of the road centerline. The specific data generally comes from the design drawings. The result of the road line discretization solved at this time, G road :{(x g1 ,y g1 ),(x g2 ,y g2 )…(x gm ,y gm )}, in this embodiment, every 2m interval is used as a discrete point.

[0077] Specifically, after the radar is installed, the RTK measurement equipment is used to align the radar installation position, collect its longitude and latitude coordinates, ensure that the deviation between the radar installation position and the design position is at the centimeter level, and convert the longitude and latitude coordinates into engineering coordinates. In the embodiment, the results of the stake number coordinate system and the engineering coordinate system of the radar installation point refer to Table 1. Among them, through this initialization method, the relative position relationship between the radar and the road is determined to ensure the benchmark of the subsequent calibration process of the radar.

[0078] Table 1

[0079]

[0080]

[0081] S2: In this embodiment, the original coordinate point recorded in the radar is A. In the radar original coordinate system, the radar is generally used as the origin. The coordinate point after calibration conversion is B, and all radar coordinate systems are unified into the engineering coordinate system. Each radar needs to calibrate the following conversion matrix, which contains the unknowns α, θ, c 1 ,c2 .

[0082]

[0083] Specifically, taking radar points 1 and 2 as examples, c 1 ,c 2 Take 148765.3, 228130; 149220.3, 229811.3 respectively; the physical meaning is to move the radar position from the original position to the corresponding c 1 ,c 2 At this time, the unknowns only include α and θ, which are the solution targets for the next optimization.

[0084] S3: The first stage optimization minimizes the difference between the vehicle trajectory detected by the radar and the actual road geometry by optimizing the radar's rotation parameters. This process is as follows Figure 3 As shown. The pre-collected data of the trajectory is filtered to remove the influence of outliers, including filtering to remove points that obviously deviate from the normal trajectory, to improve the fitting accuracy. RANSAC robust regression is used on the trajectory data to obtain the estimated road curve G est , which is used to represent the curve line shape of the road where the radar is located. At this time, the curve equation fitted by radar points 1 and 2 is similar to y A = This type of form, where x A ,y A All are represented in the original coordinate system of the radar.

[0085] Furthermore, the fitting process can be expressed as in represents the above fitting process, i represents the trajectory data set of the i-th radar, and θ is the parameter to be optimized. The fitted curve is represented by the estimated road geometry curve feature point G est :{(x e1 ,y e1 ),(x e2 ,y e2 )…(x em ,y em )}where x em ,y em The corresponding road feature points after fitting correspond to the road feature points obtained in the previous step. In order to make the estimated results and the actual results as similar as possible, solve minVar(G est -G road ), specifically aligning x and using the corresponding y for subtraction.

[0086] Furthermore, in this embodiment, the Nelder-Mead optimization algorithm is used to solve the problem. The optimization parameters θ of radar points 1 and 2 are obtained by solving the problem.1 ,θ 2 They are 0.4958 and 3.5468 respectively. The optimization results are as follows Figure 4 As shown in the figure, as a result of the first step of optimization, the radar and road coordinate systems are initially aligned. However, due to the influence of the radar's pitch angle and other factors, the radar is still not aligned in the longitudinal direction of the road, so further calibration is required.

[0087] S4: The second stage optimization, optimizes the longitudinal scaling parameter α of the radar to ensure that the motion state of the same vehicle detected by the two radars is consistent in the overlapping area. The purpose of step S4 is to identify and match the trajectory data of the same vehicle detected by different radars in the overlapping area of ​​the two radars. This process is as follows Figure 3 As shown. The overlapping area between radars is the key prerequisite for kinematic consistency optimization. Since the millimeter-wave radar cannot perceive the geometric information of the vehicle, especially at the far end of the radar's perception range, the vehicle reflection centroid cannot reflect the vehicle's geometric centroid to a certain extent. Therefore, this embodiment pays special attention to the matching of the longitudinal distance in this step, while taking into account the lateral distance of vehicles in the same lane, only the trajectory data that can be detected by both radars and successfully matched is retained, and incomplete trajectory pairs and those with excessive errors are eliminated. In the embodiment, 100 pairs of successfully paired trajectory pairs are retained.

[0088] The matched trajectory data is divided into horizontal and vertical directions, and the position and time are fitted respectively. Optimize the following objective function. In order to make the deviation of the horizontal fitting of the road smaller, in the embodiment, β=2,γ=1, and the corresponding optimization equation is

[0089] Furthermore, the scaling parameters α of radar points 1 and 2 can be obtained by using the nelder-mead method. 1 ,α 2 They are 1.02 and 1.03 respectively. The optimization results are shown in Figure 4 As shown, the vehicles now maintain better kinematic consistency within the same overlapping area.

[0090] S5: In this embodiment, the periodic online calibration of all radars ensures the long-term stability of the entire radar network and the accuracy of the data. Before calibration, first ensure the normal operation of all radar hardware and the accuracy of the initial calibration parameters. The calibration cycle between each pair of radars is set to 30 minutes. This periodic calibration ensures that the system can adapt to changes in road conditions, such as fluctuations in traffic flow, weather changes, etc., so as to keep the calibration parameters updated and accurate in real time.

[0091] The beneficial effects of the present invention are as follows:

[0092] (1) The present invention innovatively combines the two stages of road geometry similarity optimization and motion consistency optimization to improve the calibration accuracy of radar pairs. In the first stage, the rotation parameters of the radar are optimized by utilizing the road geometry features and the vehicle trajectory data detected by the radar, and the first stage makes the radar roughly aligned. In the second stage, by analyzing the motion consistency between different radars, the longitudinal scaling parameters of the radar are further optimized to ensure that the vehicle motion data detected in the radar overlap area has a high degree of consistency. This two-stage method makes full use of different prior knowledge and combines it with the transformation matrix to improve the efficiency of calibration.

[0093] (2) The calibration method of the present invention has strong adaptability and flexibility. It can be easily integrated into the existing intelligent transportation system, only considering the trajectory data and existing road facilities conditions, and can be compatible with different radar models and configurations without the need for large-scale transformation of existing infrastructure. This flexibility allows the method of the present invention to be quickly deployed in different roads and traffic environments, and has broad application prospects. In summary, the distributed two-stage calibration method of the present invention shows advantages in calibration efficiency, adaptability to different traffic environments, cost-effectiveness, and system stability, and provides a new adaptive calibration solution for roadside millimeter-wave radars in intelligent transportation systems.

[0094] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0095] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A distributed two-stage roadside multi-millimeter wave radar collaborative calibration method, characterized in that: include: Construct the engineering coordinate system and road line shape discretization representation based on the pre-collected horizontal curve data of the target road section, and use the RTK measurement equipment to align the radar equipment to the target installation position; Construct radar calibration equations; Using the radar device to collect first trajectory data of the vehicle on the target road section, and using RANSAC robust regression to fit the first trajectory data to obtain a road estimation curve; The road estimation curve is converted into an engineering estimation curve by using the rotation parameters of the radar calibration equation, and the rotation parameters of the radar calibration equation are determined by using the Nelder-Mead optimization algorithm when the variance of the engineering estimation curve and the road linear discretization representation is minimized, so as to obtain the optimized rotation parameters; In the overlapping area of ​​the adjacent radar devices, two sets of second trajectory data of the same vehicle are identified and matched, and the two sets of second trajectory data are matched by using a distance sorting method to obtain matching trajectory data; Extracting motion features of the second trajectory data in the X and Y directions, and fitting the motion features with a time-based cubic polynomial function using a least squares method to obtain an X-direction motion equation and a Y-direction motion equation; Calculate the difference between the observed value of the X-direction motion equation and the Y-direction motion equation and the fitting value of the matching trajectory data at the same time to obtain the X-direction residual and the Y-direction residual, input the X-direction residual and the Y-direction residual into the optimization target, and use the Nelder-Mead optimization algorithm to determine the stretching parameter of the radar calibration equation when the output result of the optimization target is minimized, and obtain the optimized stretching parameter; Updating the optimized rotation parameter and the optimized stretching parameter to the radar calibration equation, and calibrating the vehicle on the target road section using the updated radar calibration equation; Return to step "using the radar device to collect the first trajectory data of the vehicle on the target road section, and using RANSAC robust regression to fit the first trajectory data to obtain a road estimation curve" with a fixed time as a cycle to perform periodic online collaborative calibration on the radar calibration equation.

2. A distributed two-stage roadside multi-millimeter wave radar collaborative calibration method according to claim 1, characterized in that: The flat curve data includes: straight line parameters, curve parameters and corner parameters.

3. The distributed two-stage roadside multi-millimeter wave radar collaborative calibration method according to claim 1 is characterized in that: The radar calibration equation is: B = AH; in, B is the coordinate point after conversion; A is the original coordinate point collected by the radar device; H is the conversion matrix; θ is the rotation parameter; α is the stretching parameter; c1 and c2 are the coordinates of the radar device in the X and Y directions in the engineering coordinate system, respectively.

4. A distributed two-stage roadside multi-millimeter wave radar collaborative calibration method according to claim 1, characterized in that: The expression of the optimization objective is: Among them, γ and β are the error weights in the X and Y directions respectively; e x 、e y are the X-direction residual and the Y-direction residual respectively; n is the number of data points; i is e x and e y The corresponding point number.

5. The distributed two-stage roadside multi-millimeter wave radar collaborative calibration method according to claim 1 is characterized in that: The distance between adjacent pile numbers in the engineering coordinate system is less than or equal to 2m.

6. A distributed two-stage roadside multi-millimeter wave radar collaborative calibration method according to claim 1, characterized in that: The first trajectory data includes: a time field, a horizontal coordinate field, a vertical coordinate field and a speed field.

7. A distributed two-stage roadside multi-millimeter wave radar collaborative calibration method according to claim 1, characterized in that: The road estimation curve, the X-direction motion equation, and the Y-direction motion equation are all three-variable cubic equations.

8. The distributed two-stage roadside multi-millimeter wave radar collaborative calibration method according to claim 1 is characterized in that: The fixing time is 30 minutes.

9. A distributed two-stage roadside multi-millimeter wave radar collaborative calibration method according to claim 1, characterized in that: The installation distance between adjacent radar devices is 350m; the length of the overlapping area is 100m.

10. The distributed two-stage roadside multi-millimeter wave radar collaborative calibration method according to claim 1, characterized in that: The error between the actual installation position of the radar device and the target installation position is less than 1 cm.

Images